naturalistic approach in research

• Bipolar Disorder
• Therapy Center
• When To See a Therapist
• Types of Therapy
• Best Online Therapy
• Best Couples Therapy
• Best Family Therapy
• Managing Stress
• Sleep and Dreaming
• Understanding Emotions
• Self-Improvement
• Healthy Relationships
• Student Resources
• Personality Types
• Guided Meditations
• Verywell Mind Insights
• 2023 Verywell Mind 25
• Mental Health in the Classroom
• Editorial Process
• Meet Our Review Board
• Crisis Support

What Is Naturalistic Observation?

Kendra Cherry, MS, is a psychosocial rehabilitation specialist, psychology educator, and author of the "Everything Psychology Book."

Amy Morin, LCSW, is a psychotherapist and international bestselling author. Her books, including "13 Things Mentally Strong People Don't Do," have been translated into more than 40 languages. Her TEDx talk,  "The Secret of Becoming Mentally Strong," is one of the most viewed talks of all time.

Illustration by Brianna Gilmartin, Verywell

• How Naturalistic Observation Works
• Pros and Cons
• Data Collection Methods

How Often Is Data Collected?

Naturalistic observation is a research method that involves observing subjects in their natural environment. This approach is often used by psychologists and other social scientists. It is a form of qualitative research , which focuses on collecting, evaluating, and describing non-numerical data.

It can be useful if conducting lab research would be unrealistic, cost-prohibitive, or would unduly affect the subject's behavior. The goal of naturalistic observation is to observe behavior as it occurs in a natural setting without interference or attempts to manipulate variables.

This article discusses how naturalistic observation works and the pros and cons of doing this type of research. It also covers how data is collected and examples of when this method might be used in psychology research.

How Does Naturalistic Observation Work?

People do not necessarily behave in a lab setting the way they would in a natural environment. Researchers sometimes want to observe their subject's behavior as it happens ("in the wild," so to speak). Psychologists can get a better idea of how and why people react the way that they do by watching how they respond to situations and stimuli in real life.

Naturalistic observation is different than structured observation because it involves looking at a subject's behavior as it occurs in a natural setting, with no attempts at intervention on the part of the researcher.

For example, a researcher interested in aspects of classroom behavior (such as the interactions between students or teacher-student dynamics) might use naturalistic observation as part of their research.

Performing these observations in a lab would be difficult because it would involve recreating a classroom environment. This would likely influence the behavior of the participants, making it difficult to generalize the observations made.

By observing the subjects in their natural setting (the classroom where they work and learn), the researchers can more fully observe the behavior they are interested in as it occurs in the real world.

Naturalistic Observation Pros and Cons 

Like other research methods, naturalistic observation has advantages and disadvantages.

More realistic

More affordable

Can detect patterns

Inability to manipulate or control variables

Cannot explain why behaviors happen

Risk of observer bias

An advantage of naturalistic observation is that it allows the investigators to directly observe the subject in a natural setting. The method gives scientists a first-hand look at social behavior and can help them notice things that they might never have encountered in a lab setting.

The observations can also serve as inspiration for further investigations. The information gleaned from naturalistic observation can lead to insights that can be used to help people overcome problems and lead to healthier, happier lives.

Other advantages of naturalistic observation include:

• Allows researchers to study behaviors or situations that cannot be manipulated in a lab due to ethical concerns . For example, it would be unethical to study the effects of imprisonment by actually confining subjects. But researchers can gather information by using naturalistic observation in actual prison settings.
• Can support the external validity of research . Researchers might believe that the findings of a lab study can be generalized to a larger population, but that does not mean they would actually observe those findings in a natural setting. They may conduct naturalistic observation to make that confirmation.

Naturalistic observation can be useful in many cases, but the method also has some downsides. Some of these include:

• Inability to draw cause-and-effect conclusions : The biggest disadvantage of naturalistic observation is that determining the exact cause of a subject's behavior can be difficult.
• Lack of control : Another downside is that the experimenter cannot control for outside variables .
• Lack of validity : While the goal of naturalistic observation is to get a better idea of how it occurs in the real world, experimental effects can still influence how people respond. The Hawthorne effect and other demand characteristics can play a role in people altering their behavior simply because they know they are being observed.
• Observer bias : The biases of the people observing the natural behaviors can influence the interpretations that experimenters make.

It is also important to note that naturalistic observation is a type of correlational research (others include surveys and archival research). A correlational study is a non-experimental approach that seeks to find statistical relationships between variables. Naturalistic observation is one method that can be used to collect data for correlational studies.

While such methods can look at the direction or strength of a relationship between two variables, they cannot determine if one causes the other. As the saying goes, correlation does not imply causation.

Data Collection Methods 

Researchers use different techniques to collect and record data from naturalistic observation. For example, they might write down how many times a certain behavior occurred in a specific period of time or take a video recording of subjects.

• Audio or video recordings : Depending on the type of behavior being observed, the researchers might also decide to make audio or videotaped recordings of each observation session. They can then later review the recordings.
• Observer narrative : The observer might take notes during the session that they can refer back to. They can collect data and discern behavior patterns from these notes.
• Tally counts : The observer writes down when and how many times certain behaviors occurred.

It is rarely practical—or even possible—to observe  every  moment of a subject's life. Therefore, researchers often use sampling to gather information through naturalistic observation.

The goal is to make sure that the sample of data is representative of the subject's overall behavior. A representative sample is a selection that accurately depicts the characteristics that are present in the total subject of interest. A  representative sample  can be obtained through:

• Time sampling : This involves taking samples at different intervals of time (random or systematic). For example, a researcher might observe a person in the workplace to notice how frequently they engage in certain behaviors and to determine if there are patterns or trends.
• Situation sampling : This type of sampling involves observing behavior in different situations and settings. An example of this would be observing a child in a classroom, home, and community setting to determine if certain behaviors only occur in certain settings.
• Event sampling : This approach involves observing and recording each time an event happens. This allows the researchers to better identify patterns that might be present. For example, a researcher might note every time a subject becomes agitated. By noting the event and what was occurring around the time of each event, researchers can draw inferences about what might be triggering those behaviors.

Examples of Naturalistic Observation

Imagine that you want to study risk-taking behavior in teenagers. You might choose to observe behavior in different settings, such as a sledding hill, a rock-climbing wall, an ice-skating rink, and a bumper car ride. After you operationally define "risk-taking behavior," you would observe your teen subjects in these settings and record every incidence of what you have defined as risky behavior.

Famous examples of naturalistic observations include Charles Darwin's journey aboard the  HMS Beagle , which served as the basis for his theory of natural selection, and Jane Goodall's work studying the behavior of chimpanzees in their natural habitat.

Naturalistic observation can play an important role in the research process. It offers a number of advantages, including often being more affordable and less intrusive than other types of research.

In some cases, researchers may utilize naturalistic observation as a way to learn more about something that is happening in a certain population. Using this information, they can then formulate a hypothesis that can be tested further.

Mehl MR, Robbins ML, Deters FG. Naturalistic observation of health-relevant social processes: the electronically activated recorder methodology in psychosomatics . Psychosom Med. 2012;74(4):410-7. doi:10.1097/PSY.0b013e3182545470

U.S. National Library of Medicine. Rewriting the book of nature - Darwin and the Beagle voyage .

Angrosino MV. Naturalistic Observation . Left Coast Press.

DiMercurio A, Connell JP, Clark M, Corbetta D. A naturalistic observation of spontaneous touches to the body and environment in the first 2 months of life . Front Psychol . 2018;9:2613. doi:10.3389/fpsyg.2018.02613

Pierce K, Pepler D. A peek behind the fence: observational methods 25 years later . In: Smith PK, Norman JO, eds. The Wiley Blackwell Handbook of Bullying. 1st ed . Wiley; 2021:215-232. doi:10.1002/9781118482650.ch12

By Kendra Cherry, MSEd Kendra Cherry, MS, is a psychosocial rehabilitation specialist, psychology educator, and author of the "Everything Psychology Book."

Observation Method in Psychology: Naturalistic, Participant and Controlled

Saul Mcleod, PhD

Editor-in-Chief for Simply Psychology

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul Mcleod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

Learn about our Editorial Process

Olivia Guy-Evans, MSc

Associate Editor for Simply Psychology

BSc (Hons) Psychology, MSc Psychology of Education

Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.

On This Page:

The observation method in psychology involves directly and systematically witnessing and recording measurable behaviors, actions, and responses in natural or contrived settings without attempting to intervene or manipulate what is being observed.

Used to describe phenomena, generate hypotheses, or validate self-reports, psychological observation can be either controlled or naturalistic with varying degrees of structure imposed by the researcher.

There are different types of observational methods, and distinctions need to be made between:

1. Controlled Observations 2. Naturalistic Observations 3. Participant Observations

In addition to the above categories, observations can also be either overt/disclosed (the participants know they are being studied) or covert/undisclosed (the researcher keeps their real identity a secret from the research subjects, acting as a genuine member of the group).

In general, conducting observational research is relatively inexpensive, but it remains highly time-consuming and resource-intensive in data processing and analysis.

The considerable investments needed in terms of coder time commitments for training, maintaining reliability, preventing drift, and coding complex dynamic interactions place practical barriers on observers with limited resources.

Controlled Observation

Controlled observation is a research method for studying behavior in a carefully controlled and structured environment.

The researcher sets specific conditions, variables, and procedures to systematically observe and measure behavior, allowing for greater control and comparison of different conditions or groups.

The researcher decides where the observation will occur, at what time, with which participants, and in what circumstances, and uses a standardized procedure. Participants are randomly allocated to each independent variable group.

Rather than writing a detailed description of all behavior observed, it is often easier to code behavior according to a previously agreed scale using a behavior schedule (i.e., conducting a structured observation).

The researcher systematically classifies the behavior they observe into distinct categories. Coding might involve numbers or letters to describe a characteristic or the use of a scale to measure behavior intensity.

The categories on the schedule are coded so that the data collected can be easily counted and turned into statistics.

For example, Mary Ainsworth used a behavior schedule to study how infants responded to brief periods of separation from their mothers. During the Strange Situation procedure, the infant’s interaction behaviors directed toward the mother were measured, e.g.,

• Proximity and contact-seeking
• Contact maintaining
• Avoidance of proximity and contact
• Resistance to contact and comforting

The observer noted down the behavior displayed during 15-second intervals and scored the behavior for intensity on a scale of 1 to 7.

Sometimes participants’ behavior is observed through a two-way mirror, or they are secretly filmed. Albert Bandura used this method to study aggression in children (the Bobo doll studies ).

A lot of research has been carried out in sleep laboratories as well. Here, electrodes are attached to the scalp of participants. What is observed are the changes in electrical activity in the brain during sleep ( the machine is called an EEG ).

Controlled observations are usually overt as the researcher explains the research aim to the group so the participants know they are being observed.

Controlled observations are also usually non-participant as the researcher avoids direct contact with the group and keeps a distance (e.g., observing behind a two-way mirror).

• Controlled observations can be easily replicated by other researchers by using the same observation schedule. This means it is easy to test for reliability .
• The data obtained from structured observations is easier and quicker to analyze as it is quantitative (i.e., numerical) – making this a less time-consuming method compared to naturalistic observations.
• Controlled observations are fairly quick to conduct which means that many observations can take place within a short amount of time. This means a large sample can be obtained, resulting in the findings being representative and having the ability to be generalized to a large population.

Limitations

• Controlled observations can lack validity due to the Hawthorne effect /demand characteristics. When participants know they are being watched, they may act differently.

Naturalistic Observation

Naturalistic observation is a research method in which the researcher studies behavior in its natural setting without intervention or manipulation.

It involves observing and recording behavior as it naturally occurs, providing insights into real-life behaviors and interactions in their natural context.

Naturalistic observation is a research method commonly used by psychologists and other social scientists.

This technique involves observing and studying the spontaneous behavior of participants in natural surroundings. The researcher simply records what they see in whatever way they can.

In unstructured observations, the researcher records all relevant behavior with a coding system. There may be too much to record, and the behaviors recorded may not necessarily be the most important, so the approach is usually used as a pilot study to see what type of behaviors would be recorded.

Compared with controlled observations, it is like the difference between studying wild animals in a zoo and studying them in their natural habitat.

With regard to human subjects, Margaret Mead used this method to research the way of life of different tribes living on islands in the South Pacific. Kathy Sylva used it to study children at play by observing their behavior in a playgroup in Oxfordshire.

Collecting Naturalistic Behavioral Data

Technological advances are enabling new, unobtrusive ways of collecting naturalistic behavioral data.

The Electronically Activated Recorder (EAR) is a digital recording device participants can wear to periodically sample ambient sounds, allowing representative sampling of daily experiences (Mehl et al., 2012).

Studies program EARs to record 30-50 second sound snippets multiple times per hour. Although coding the recordings requires extensive resources, EARs can capture spontaneous behaviors like arguments or laughter.

EARs minimize participant reactivity since sampling occurs outside of awareness. This reduces the Hawthorne effect, where people change behavior when observed.

The SenseCam is another wearable device that passively captures images documenting daily activities. Though primarily used in memory research currently (Smith et al., 2014), systematic sampling of environments and behaviors via the SenseCam could enable innovative psychological studies in the future.

• By being able to observe the flow of behavior in its own setting, studies have greater ecological validity.
• Like case studies , naturalistic observation is often used to generate new ideas. Because it gives the researcher the opportunity to study the total situation, it often suggests avenues of inquiry not thought of before.
• The ability to capture actual behaviors as they unfold in real-time, analyze sequential patterns of interactions, measure base rates of behaviors, and examine socially undesirable or complex behaviors that people may not self-report accurately.
• These observations are often conducted on a micro (small) scale and may lack a representative sample (biased in relation to age, gender, social class, or ethnicity). This may result in the findings lacking the ability to generalize to wider society.
• Natural observations are less reliable as other variables cannot be controlled. This makes it difficult for another researcher to repeat the study in exactly the same way.
• Highly time-consuming and resource-intensive during the data coding phase (e.g., training coders, maintaining inter-rater reliability, preventing judgment drift).
• With observations, we do not have manipulations of variables (or control over extraneous variables), meaning cause-and-effect relationships cannot be established.

Participant Observation

Participant observation is a variant of the above (natural observations) but here, the researcher joins in and becomes part of the group they are studying to get a deeper insight into their lives.

If it were research on animals , we would now not only be studying them in their natural habitat but be living alongside them as well!

Leon Festinger used this approach in a famous study into a religious cult that believed that the end of the world was about to occur. He joined the cult and studied how they reacted when the prophecy did not come true.

Participant observations can be either covert or overt. Covert is where the study is carried out “undercover.” The researcher’s real identity and purpose are kept concealed from the group being studied.

The researcher takes a false identity and role, usually posing as a genuine member of the group.

On the other hand, overt is where the researcher reveals his or her true identity and purpose to the group and asks permission to observe.

• It can be difficult to get time/privacy for recording. For example, researchers can’t take notes openly with covert observations as this would blow their cover. This means they must wait until they are alone and rely on their memory. This is a problem as they may forget details and are unlikely to remember direct quotations.
• If the researcher becomes too involved, they may lose objectivity and become biased. There is always the danger that we will “see” what we expect (or want) to see. This problem is because they could selectively report information instead of noting everything they observe. Thus reducing the validity of their data.

Recording of Data

With controlled/structured observation studies, an important decision the researcher has to make is how to classify and record the data. Usually, this will involve a method of sampling.

In most coding systems, codes or ratings are made either per behavioral event or per specified time interval (Bakeman & Quera, 2011).

The three main sampling methods are:

Event-based coding involves identifying and segmenting interactions into meaningful events rather than timed units.

For example, parent-child interactions may be segmented into control or teaching events to code. Interval recording involves dividing interactions into fixed time intervals (e.g., 6-15 seconds) and coding behaviors within each interval (Bakeman & Quera, 2011).

Event recording allows counting event frequency and sequencing while also potentially capturing event duration through timed-event recording. This provides information on time spent on behaviors.

Coding Systems

The coding system should focus on behaviors, patterns, individual characteristics, or relationship qualities that are relevant to the theory guiding the study (Wampler & Harper, 2014).

Codes vary in how much inference is required, from concrete observable behaviors like frequency of eye contact to more abstract concepts like degree of rapport between a therapist and client (Hill & Lambert, 2004). More inference may reduce reliability.

Macroanalytic coding systems

Macroanalytic coding systems involve rating or summarizing behaviors using larger coding units and broader categories that reflect patterns across longer periods of interaction rather than coding small or discrete behavioral acts. 

For example, a macroanalytic coding system may rate the overall degree of therapist warmth or level of client engagement globally for an entire therapy session, requiring the coders to summarize and infer these constructs across the interaction rather than coding smaller behavioral units.

These systems require observers to make more inferences (more time-consuming) but can better capture contextual factors, stability over time, and the interdependent nature of behaviors (Carlson & Grotevant, 1987).

Microanalytic coding systems

Microanalytic coding systems involve rating behaviors using smaller, more discrete coding units and categories.

For example, a microanalytic system may code each instance of eye contact or head nodding during a therapy session. These systems code specific, molecular behaviors as they occur moment-to-moment rather than summarizing actions over longer periods.

Microanalytic systems require less inference from coders and allow for analysis of behavioral contingencies and sequential interactions between therapist and client. However, they are more time-consuming and expensive to implement than macroanalytic approaches.

Mesoanalytic coding systems

Mesoanalytic coding systems attempt to balance macro- and micro-analytic approaches.

In contrast to macroanalytic systems that summarize behaviors in larger chunks, mesoanalytic systems use medium-sized coding units that target more specific behaviors or interaction sequences (Bakeman & Quera, 2017).

For example, a mesoanalytic system may code each instance of a particular type of therapist statement or client emotional expression. However, mesoanalytic systems still use larger units than microanalytic approaches coding every speech onset/offset.

The goal of balancing specificity and feasibility makes mesoanalytic systems well-suited for many research questions (Morris et al., 2014). Mesoanalytic codes can preserve some sequential information while remaining efficient enough for studies with adequate but limited resources.

For instance, a mesoanalytic couple interaction coding system could target key behavior patterns like validation sequences without coding turn-by-turn speech.

In this way, mesoanalytic coding allows reasonable reliability and specificity without requiring extensive training or observation. The mid-level focus offers a pragmatic compromise between depth and breadth in analyzing interactions.

Preventing Coder Drift

Coder drift results in a measurement error caused by gradual shifts in how observations get rated according to operational definitions, especially when behavioral codes are not clearly specified.

This type of error creeps in when coders fail to regularly review what precise observations constitute or do not constitute the behaviors being measured.

Preventing drift refers to taking active steps to maintain consistency and minimize changes or deviations in how coders rate or evaluate behaviors over time. Specifically, some key ways to prevent coder drift include:

• Operationalize codes : It is essential that code definitions unambiguously distinguish what interactions represent instances of each coded behavior. 
• Ongoing training : Returning to those operational definitions through ongoing training serves to recalibrate coder interpretations and reinforce accurate recognition. Having regular “check-in” sessions where coders practice coding the same interactions allows monitoring that they continue applying codes reliably without gradual shifts in interpretation.
• Using reference videos : Coders periodically coding the same “gold standard” reference videos anchors their judgments and calibrate against original training. Without periodic anchoring to original specifications, coder decisions tend to drift from initial measurement reliability.
• Assessing inter-rater reliability : Statistical tracking that coders maintain high levels of agreement over the course of a study, not just at the start, flags any declines indicating drift. Sustaining inter-rater agreement requires mitigating this common tendency for observer judgment change during intensive, long-term coding tasks.
• Recalibrating through discussion : Having meetings for coders to discuss disagreements openly explores reasons judgment shifts may be occurring over time. Consensus on the application of codes is restored.
• Adjusting unclear codes : If reliability issues persist, revisiting and refining ambiguous code definitions or anchors can eliminate inconsistencies arising from coder confusion.

Essentially, the goal of preventing coder drift is maintaining standardization and minimizing unintentional biases that may slowly alter how observational data gets rated over periods of extensive coding.

Through the upkeep of skills, continuing calibration to benchmarks, and monitoring consistency, researchers can notice and correct for any creeping changes in coder decision-making over time.

Reducing Observer Bias

Observational research is prone to observer biases resulting from coders’ subjective perspectives shaping the interpretation of complex interactions (Burghardt et al., 2012). When coding, personal expectations may unconsciously influence judgments. However, rigorous methods exist to reduce such bias.

Coding Manual

A detailed coding manual minimizes subjectivity by clearly defining what behaviors and interaction dynamics observers should code (Bakeman & Quera, 2011).

High-quality manuals have strong theoretical and empirical grounding, laying out explicit coding procedures and providing rich behavioral examples to anchor code definitions (Lindahl, 2001).

Clear delineation of the frequency, intensity, duration, and type of behaviors constituting each code facilitates reliable judgments and reduces ambiguity for coders. Application risks inconsistency across raters without clarity on how codes translate to observable interaction.

Coder Training

Competent coders require both interpersonal perceptiveness and scientific rigor (Wampler & Harper, 2014). Training thoroughly reviews the theoretical basis for coded constructs and teaches the coding system itself.

Multiple “gold standard” criterion videos demonstrate code ranges that trainees independently apply. Coders then meet weekly to establish reliability of 80% or higher agreement both among themselves and with master criterion coding (Hill & Lambert, 2004).

Ongoing training manages coder drift over time. Revisions to unclear codes may also improve reliability. Both careful selection and investment in rigorous training increase quality control.

Blind Methods

To prevent bias, coders should remain unaware of specific study predictions or participant details (Burghardt et al., 2012). Separate data gathering versus coding teams helps maintain blinding.

Coders should be unaware of study details or participant identities that could bias coding (Burghardt et al., 2012).

Separate teams collecting data versus coding data can reduce bias.

In addition, scheduling procedures can prevent coders from rating data collected directly from participants with whom they have had personal contact. Maintaining coder independence and blinding enhances objectivity.

Bakeman, R., & Quera, V. (2017). Sequential analysis and observational methods for the behavioral sciences. Cambridge University Press.

Burghardt, G. M., Bartmess-LeVasseur, J. N., Browning, S. A., Morrison, K. E., Stec, C. L., Zachau, C. E., & Freeberg, T. M. (2012). Minimizing observer bias in behavioral studies: A review and recommendations. Ethology, 118 (6), 511-517.

Hill, C. E., & Lambert, M. J. (2004). Methodological issues in studying psychotherapy processes and outcomes. In M. J. Lambert (Ed.), Bergin and Garfield’s handbook of psychotherapy and behavior change (5th ed., pp. 84–135). Wiley.

Lindahl, K. M. (2001). Methodological issues in family observational research. In P. K. Kerig & K. M. Lindahl (Eds.), Family observational coding systems: Resources for systemic research (pp. 23–32). Lawrence Erlbaum Associates.

Mehl, M. R., Robbins, M. L., & Deters, F. G. (2012). Naturalistic observation of health-relevant social processes: The electronically activated recorder methodology in psychosomatics. Psychosomatic Medicine, 74 (4), 410–417.

Morris, A. S., Robinson, L. R., & Eisenberg, N. (2014). Applying a multimethod perspective to the study of developmental psychology. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (2nd ed., pp. 103–123). Cambridge University Press.

Smith, J. A., Maxwell, S. D., & Johnson, G. (2014). The microstructure of everyday life: Analyzing the complex choreography of daily routines through the automatic capture and processing of wearable sensor data. In B. K. Wiederhold & G. Riva (Eds.), Annual Review of Cybertherapy and Telemedicine 2014: Positive Change with Technology (Vol. 199, pp. 62-64). IOS Press.

Traniello, J. F., & Bakker, T. C. (2015). The integrative study of behavioral interactions across the sciences. In T. K. Shackelford & R. D. Hansen (Eds.), The evolution of sexuality (pp. 119-147). Springer.

Wampler, K. S., & Harper, A. (2014). Observational methods in couple and family assessment. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social and personality psychology (2nd ed., pp. 490–502). Cambridge University Press.

What Is Naturalistic Observation? Definition and Examples

• Archaeology

• Ph.D., Psychology, Fielding Graduate University
• M.A., Psychology, Fielding Graduate University
• B.A., Film Studies, Cornell University

Naturalistic observation is a research method used in psychology and other social sciences in which research participants are observed in their natural environments. Unlike lab experiments that involve testing hypotheses and controlling variables, naturalistic observation simply requires recording what is observed in a specific setting.

Kay Takeaways: Naturalistic Observation

• Naturalistic observation is a research method in which people or other subjects are observed in their natural setting.
• Psychologists and other social scientists use naturalistic observation to study specific social or cultural settings that couldn’t be investigated in other ways, such as prisons, bars, and hospitals.
• Naturalistic observation has some drawbacks, including the inability to control for variables and a lack of replicability.

Naturalistic Observation Applications

Naturalistic observation involves observing subjects of interest in their normal, everyday setting . It is sometimes referred to as field work because it requires researchers to go out into the field (the natural setting) to collect data on their participants. Naturalistic observation traces its roots back to anthropology and animal behavior research. For example, cultural anthropologist Margaret Mead used naturalistic observation to study the daily lives of different groups in the South Pacific.

The approach doesn't always require researchers to observe people in such exotic environments, however. It can be conducted in any kind of social or organizational setting , including offices, schools, bars, prisons, dorm rooms, online message boards, or just about any other place where people can be observed. For example, psychologist Sylvia Scribner used naturalistic observation to investigate how people make decisions in various professions. To do so, she accompanied people—from milk men, to cashiers, to machine operators—as they went about their regular work routines.

Naturalistic observation is valuable when a researcher wants to learn more about people in a specific social or cultural setting but can’t gather the information any other way. Sometimes studying people in a lab can impact their behavior, be cost prohibitive, or both. For example, if a researcher wishes to study the behavior of shoppers in the weeks leading up to the Christmas holiday, it would be impractical to construct a store in the lab. Plus, even if the researcher did so, it would be unlikely to elicit the same response from participants as shopping at a store in the real world. Naturalistic observation offers the opportunity to observe shoppers’ behavior, and based on researchers' observations of the situation, has the potential to generate new ideas for specific hypotheses or avenues of research.

The method requires researchers to immerse themselves in the setting being studied. This typically involves taking copious field notes. Researchers may also interview specific people involved in the situation, collect documents from the setting, and make audio or video recordings. In her research on decision-making in different occupations, for instance, Scribner not only took detailed notes, she also gathered every scrap of written material her participants read and produced, and photographed the equipment they used.

Scope of the Observation

Before going into the field, researchers conducting naturalistic observation must define the scope of their research. While the researcher may want to study everything about the people in the chosen setting, this may not be realistic given the complexities of human behavior. As a result, the researcher must focus observations on the specific behaviors and responses they are most interested in studying.

For example, the researcher might choose to collect quantitative data by counting the number of times a specific behavior occurs. So, if the researcher is interested in dog owners' interactions with their dogs, they might tally the number of times the owner talks to their dog during a walk. On the other hand, much of the data collected during naturalistic observation, including notes, audio and video recordings and interviews, are qualitative data that require the researcher to describe, analyze, and interpret what was observed.

Sampling Methods

Another way researchers can limit the scope of a study is by using a specific sampling method. This will enable them to gather a representative sample of data on the subjects’ behavior without having to observe everything the subject does at all times. Sampling methods include:

• Time sampling, which means the researcher will observe subjects at different intervals of time. These intervals could be random or specific. For example, the researcher could decide to only observe subjects every morning for an hour.
• Situation sampling, which means the researcher will observe the same subjects in various situations. For instance, if a researcher wants to observe the behavior of Star Wars fans' responses to the release of the most recent movie in the franchise, the researcher might observe fans’ behavior at the red carpet of the movie's premiere, during screenings, and on online Star Wars message boards.
• Event sampling , which means the researcher will only record specific behaviors and ignore all others. For example, when observing interactions between children on a playground, the researcher might decide they’re only interested in observing how children decide to take turns on the slide while ignoring behavior on the other playground equipment.

Pros and Cons of Naturalistic Observation

There are a number of advantages to naturalistic observation. These include:

• Studies have greater external validity because the researcher’s data comes directly from observing subjects in their natural environment.
• Observing people in the field can lead to glimpses of behavior that could never occur in a lab, possibly leading to unique insights.
• The researcher can study things that would be impossible or unethical to reproduce in a lab. For example, while it would be unethical to study the way people cope with the aftermath of violence by manipulating exposure in a lab, researchers can gather data on this subject by observing participants in a support group.

Despite its value in certain situations, naturalistic observation can have a number of drawbacks, including:

• Naturalistic observation studies typically involve observing a limited number of settings . As a result, the subjects being studied are limited to certain ages, genders, ethnicities, or other characteristics, which means a study’s findings cannot be generalized to the population as a whole.
• Researchers can’t control for different variables like they can in a lab, which makes naturalistic observation studies less reliable and more difficult to replicate.
• Lack of control over external variables also makes it impossible to determine the cause of the behaviors the researcher observes.
• If subjects know they’re being observed, it has the potential to change their behavior.
• Cherry, Kendra. Naturalistic Observation in Psychology.” V erywellMind , 1 October, 2019. https://www.verywellmind.com/what-is-naturalistic-observation-2795391
• Cozby, Paul C. Methods in Behavioral Research . 10th ed., McGraw-Hill. 2009.
• McLeod, Saul A. “Observation Methods.” Simply Psychology , 6 June 2015. https://www.simplypsychology.org/observation.html
• An Overview of Qualitative Research Methods
• What Are Natural Experiments and How Do Economists Use Them?
• What Is Participant Observation Research?
• What Is a Cohort Effect? Definition and Examples
• What Is Direct Observation?
• Conducting Case Study Research in Sociology
• What Is Ethnography?
• How to Understand Interpretive Sociology
• The Different Types of Sampling Designs in Sociology
• How to Conduct a Sociology Research Interview
• The Study of Cultural Artifacts via Content Analysis
• Definition and Overview of Grounded Theory
• Pilot Study in Research
• Understanding Purposive Sampling
• What are Controlled Experiments?
• What Is Panel Data?

Naturalistic Observation: Definition, Examples, and Advantages

Categories Research Methods

Sharing is caring!

Naturalistic observation is a psychological research method that involves observing and recording behavior in the natural environment. Unlike experiments, researchers do not manipulate variables. This research method is frequently used in psychology to help researchers investigate human behavior.

This article explores how naturalistic observation is used in psychology. It offers examples and the potential advantages and disadvantages of this type of research. 

Table of Contents

What Is Naturalistic Observation?

In naturalistic observation, the researcher observes the participants’ behavior in their natural setting, taking notes on their behavior and interactions. The researcher may use various tools, such as video or audio recordings, to help capture the behavior accurately. The researcher may also use coding systems or other quantitative measures to systematically record observed behavior.

Naturalistic observation can be used to investigate a wide range of psychological phenomena, such as social interaction patterns, parental behavior, or animal behavior. 

Types of Naturalistic Observation

Naturalistic observation can be:

Unstructured or Structured

The observer can either watch and record everything that happens, or they can have a checklist or form to guide their observations.

Participant or Non-Participant

The observer can be an active participant, or they can remain separate from the subject and view from the sidelines.

Overt or Covert

The observer can either openly watch and record the subjects’ behaviors, or they can keep their presence hidden from the individual or group.

The specific type of naturalistic observation that researchers use depends on the situation, what they are researching, and the resources available. No matter the type, the observation must occur in a natural setting rather than in an experimental lab.

How to Collect Data in Naturalistic Observation

There are a number of methods that researchers might utilize to record data about the behaviors and events they observe. Some of these include:

• Note-taking : Research may opt to take notes about what they witness. This approach tends to be unstructured, allowing the observers to determine what they think is relevant and to include insights that may be helpful.
• Tally counts : In other cases, research may take a more structured approach where they count the frequency of a behavior.
• Audiovisual recordings : In other cases, research may want recordings of participant behavior. This not only allows researchers to refer to the recordings later, it can also be useful for sharing with others.

How Data Is Sampled in Naturalistic Observation

While naturalistic observation is not an experimental design, researchers still want to ensure that the data they collect represents what is happening in the group. To do this, researchers must collect a representative sample. When a sample is representative, it means that it accurately reflects what is happening in a given population.

To do this, researchers may utilize three primary sampling approaches:

Event Sampling

Event sampling involves the researcher creating a set of predefined categories and behaviors they will observe. This method is useful when the researcher wants to collect data on specific behaviors or events, allowing for more precise data collection.

Using this approach, the research would note every occurrence of a specific behavior.

Situation Sampling

Situation sampling involves observing participants in more than one situation. This approach can give researchers more insight and allow them to determine if certain behaviors only occur in specific contexts or settings. 

Time Sampling

Time sampling is a type of systematic observation that involves the researcher observing and recording the subjects’ behavior at predetermined intervals. This method is useful when the researcher wants to collect data on the frequency and duration of specific behaviors.

Each method of data collection has its strengths and weaknesses, and the choice of method depends on the research question and the nature of the subjects being observed.

Examples of Naturalistic Observation

It can be helpful to look at a few different examples to learn more about how naturalistic observation can be used:

• Researchers might observe children in a classroom to learn more about their social interaction patterns. 
• Naturalistic observation can also be used to study animal behavior in their natural habitat, such as observing chimpanzees in the wild to understand their social behavior.

Researchers use this research method in various fields, including animal researchers and anthropologists. 

The work of zoologist Konrad Lorenz, for example, relied on the use of naturalistic observation. Lorenz observed the behavior of ducklings after they hatched and noted that they became attached to the first possible parent figure they saw, a phenomenon known as imprinting. Once imprinted on a parent figure, the duckling would follow and learn from their parent.

From his naturalistic observations, Lorenz hypothesized that there was a critical period immediately after hatching where ducklings needed to imprint on a parent. Based on his observations, Lorenz conducted further experiments that confirmed his hypothesis.

More Examples of Naturalistic Observation

Naturalistic observation is a research method commonly used in various areas of psychology. 

Social Psychology

Naturalistic observation can provide valuable insights into people’s behavior in different social situations. By observing people’s behavior in a crowded public place like a shopping mall or train station, researchers can better understand how social norms are established and maintained and how people interact in various social groups.

Consumer Research

Consumer research is another area where naturalistic observation can be used effectively. By observing shoppers in a grocery store or shopping mall, researchers can study how people make purchasing decisions in real-life situations.

Researchers can gain valuable insights into consumer behavior by analyzing what catches their attention, how they interact with different products, and how they decide what to buy.

Developmental Psychology

Observing children playing in a playground or a classroom can help researchers understand how children develop and learn new skills in natural settings.

Researchers can gain insights into the developmental process by observing children as they interact with each other and learn social skills or as they learn new concepts and skills in a classroom.

Cognitive Psychology

Naturalistic observation can be used to study how people think and process information in real-life situations. For example, observing people using a computer program can help researchers understand how people navigate through it and solve problems.

Similarly, observing people in a conversation can provide insights into how they process and respond to information in real time.

Advantages of Naturalistic Observation

Naturalistic observation offers a number of benefits that can make it a good choice for research. 

Ecological Validity

One of the strengths of naturalistic observation is its ability to capture behavior in a natural setting, providing a more accurate and comprehensive picture of how people or animals behave in their everyday environment.

It is often more realistic than lab research, so it can give insight into how people behave authentically in everyday settings and situations.

Inspiration for Additional Research

Naturalistic observation can also generate new hypotheses and insights that may not be captured in other research methods. 

Research That Can’t Be Done in a Lab

Naturalistic observation allows the study of behaviors that cannot be replicated in a lab. Naturalistic observation is sometimes the only approach for studying behaviors that cannot be reproduced in a lab due to ethical reasons.

For example, researchers might use this approach to research prison behavior or the social impact of domestic violence on emotional health. Those are not situations they can manipulate in a lab, but they can observe the impact on people who have had those experiences.

Disadvantages of Naturalistic Observation

While naturalistic can be a valuable tool, it is not appropriate for every situation. Some potential downsides include: 

Bias and Lack of Control

Naturalistic observation is limited by its lack of environmental control and the potential for observer bias. Researchers must be careful to minimize the influence of their presence on the behavior being observed and to use systematic and objective methods for recording and analyzing the data. 

Inability to Infer Cause and Effect

Naturalistic observation is also limited by its inability to establish causality between variables.

Naturalistic Observation vs. Case Study

Naturalistic observation and case studies are both research methods used in psychology but differ in their approach and purpose. Naturalistic observation involves observing and recording the behavior of individuals or groups in their natural environment without any intervention or manipulation by the researcher.

On the other hand, a case study is an in-depth analysis of a single individual or a small group of individuals, often conducted through interviews, surveys, and other forms of data collection.

The key difference between naturalistic observation and a case study is that the former focuses more on observing and recording behaviors and interactions as they occur naturally, while the latter focuses on gathering detailed information about a specific individual or group.

Naturalistic observation is often used to study social interactions, group dynamics, and other natural behaviors in real-world settings. In contrast, case studies often explore complex psychological phenomena such as mental illness, personality disorders, or unusual behaviors.

Both naturalistic observation and case studies have their strengths and limitations. The choice of method depends on the research question, the level of detail needed, and the feasibility of conducting the study in a particular setting.

Naturalistic Observation Ideas

There are many potential ideas for studies that involve naturalistic observation. A few ideas include:

• Observe the behavior of animals in their natural habitats, studying their patterns of movement, foraging, and communication
• Observe human behavior in public spaces, such as parks or coffee shops, documenting patterns of social interaction and communication
• Focus on the behavior of individuals within specific social groups or communities, studying their interactions and relationships over time
• Watch the behavior of children in a classroom setting could provide insights into their learning and socialization processes

Frequently Asked Questions

Why do we use naturalistic observation.

Naturalistic observation is important because it allows researchers to better understand how individuals behave in their everyday lives. By observing behavior in a natural setting, researchers can obtain a more accurate representation of how people act and interact with each other in their normal environment. 

This method is particularly useful when studying social behavior, as it allows researchers to capture the complexity and nuances of social interactions that might not be apparent in a laboratory setting.

Naturalistic observation can also offer valuable insights into the development of certain behaviors, such as those related to child development or the formation of social groups.

What is the most famous example of naturalistic observation?

The most famous example of naturalistic observation is probably Jane Goodall’s study of chimpanzees in the wild. Goodall spent years observing the behavior of chimpanzees in Tanzania, documenting their social interactions, tool use, and other aspects of their lives. Her work helped to revolutionize our understanding of these animals and their place in the natural world.

In conclusion, naturalistic observation is a powerful research method that can be used effectively in various areas within psychology. Researchers can gain valuable insights into human behavior and cognition by observing people’s behavior in natural settings.

Bornstein MH, Cheah CSL. Audiovisual records, encoding of . In: Encyclopedia of Social Measurement . Elsevier; 2005:103-110. doi:10.1016/B0-12-369398-5/00400-X

Erdley CA, Jankowski MS. Assessing youth . In: Social Skills Across the Life Span . Elsevier; 2020:69-90. doi:10.1016/B978-0-12-817752-5.00004-4

Helmchen H. Ethical issues in naturalistic versus controlled trials . Dialogues Clin Neurosci . 2011;13(2):173-182. doi:10.31887/DCNS.2011.13.2/hhelmchen

Mehl MR, Robbins ML, Deters FG. Naturalistic observation of health-relevant social processes: the electronically activated recorder methodology in psychosomatics . Psychosom Med . 2012;74(4):410-417. doi:10.1097/PSY.0b013e3182545470

Morrison C, Lee JP, Gruenewald PJ, Mair C. The reliability of naturalistic observations of social, physical and economic environments of bars . Addict Res Theory . 2016;24(4):330-340. doi:10.3109/16066359.2016.1145674

2.2 Approaches to Research

Learning objectives.

By the end of this section, you will be able to:

• Describe the different research methods used by psychologists
• Discuss the strengths and weaknesses of case studies, naturalistic observation, surveys, and archival research
• Compare longitudinal and cross-sectional approaches to research
• Compare and contrast correlation and causation

There are many research methods available to psychologists in their efforts to understand, describe, and explain behavior and the cognitive and biological processes that underlie it. Some methods rely on observational techniques. Other approaches involve interactions between the researcher and the individuals who are being studied—ranging from a series of simple questions to extensive, in-depth interviews—to well-controlled experiments.

Each of these research methods has unique strengths and weaknesses, and each method may only be appropriate for certain types of research questions. For example, studies that rely primarily on observation produce incredible amounts of information, but the ability to apply this information to the larger population is somewhat limited because of small sample sizes. Survey research, on the other hand, allows researchers to easily collect data from relatively large samples. While this allows for results to be generalized to the larger population more easily, the information that can be collected on any given survey is somewhat limited and subject to problems associated with any type of self-reported data. Some researchers conduct archival research by using existing records. While this can be a fairly inexpensive way to collect data that can provide insight into a number of research questions, researchers using this approach have no control on how or what kind of data was collected. All of the methods described thus far are correlational in nature. This means that researchers can speak to important relationships that might exist between two or more variables of interest. However, correlational data cannot be used to make claims about cause-and-effect relationships.

Correlational research can find a relationship between two variables, but the only way a researcher can claim that the relationship between the variables is cause and effect is to perform an experiment. In experimental research, which will be discussed later in this chapter, there is a tremendous amount of control over variables of interest. While this is a powerful approach, experiments are often conducted in artificial settings. This calls into question the validity of experimental findings with regard to how they would apply in real-world settings. In addition, many of the questions that psychologists would like to answer cannot be pursued through experimental research because of ethical concerns.

Clinical or Case Studies

In 2011, the New York Times published a feature story on Krista and Tatiana Hogan, Canadian twin girls. These particular twins are unique because Krista and Tatiana are conjoined twins, connected at the head. There is evidence that the two girls are connected in a part of the brain called the thalamus, which is a major sensory relay center. Most incoming sensory information is sent through the thalamus before reaching higher regions of the cerebral cortex for processing.

Link to Learning

Watch this CBC video about Krista's and Tatiana's lives to learn more.

The implications of this potential connection mean that it might be possible for one twin to experience the sensations of the other twin. For instance, if Krista is watching a particularly funny television program, Tatiana might smile or laugh even if she is not watching the program. This particular possibility has piqued the interest of many neuroscientists who seek to understand how the brain uses sensory information.

These twins represent an enormous resource in the study of the brain, and since their condition is very rare, it is likely that as long as their family agrees, scientists will follow these girls very closely throughout their lives to gain as much information as possible (Dominus, 2011).

Over time, it has become clear that while Krista and Tatiana share some sensory experiences and motor control, they remain two distinct individuals, which provides invaluable insight for researchers interested in the mind and the brain (Egnor, 2017).

In observational research, scientists are conducting a clinical or case study when they focus on one person or just a few individuals. Indeed, some scientists spend their entire careers studying just 10–20 individuals. Why would they do this? Obviously, when they focus their attention on a very small number of people, they can gain a precious amount of insight into those cases. The richness of information that is collected in clinical or case studies is unmatched by any other single research method. This allows the researcher to have a very deep understanding of the individuals and the particular phenomenon being studied.

If clinical or case studies provide so much information, why are they not more frequent among researchers? As it turns out, the major benefit of this particular approach is also a weakness. As mentioned earlier, this approach is often used when studying individuals who are interesting to researchers because they have a rare characteristic. Therefore, the individuals who serve as the focus of case studies are not like most other people. If scientists ultimately want to explain all behavior, focusing attention on such a special group of people can make it difficult to generalize any observations to the larger population as a whole. Generalizing refers to the ability to apply the findings of a particular research project to larger segments of society. Again, case studies provide enormous amounts of information, but since the cases are so specific, the potential to apply what’s learned to the average person may be very limited.

Naturalistic Observation

If you want to understand how behavior occurs, one of the best ways to gain information is to simply observe the behavior in its natural context. However, people might change their behavior in unexpected ways if they know they are being observed. How do researchers obtain accurate information when people tend to hide their natural behavior? As an example, imagine that your professor asks everyone in your class to raise their hand if they always wash their hands after using the restroom. Chances are that almost everyone in the classroom will raise their hand, but do you think hand washing after every trip to the restroom is really that universal?

This is very similar to the phenomenon mentioned earlier in this chapter: many individuals do not feel comfortable answering a question honestly. But if we are committed to finding out the facts about hand washing, we have other options available to us.

Suppose we send a classmate into the restroom to actually watch whether everyone washes their hands after using the restroom. Will our observer blend into the restroom environment by wearing a white lab coat, sitting with a clipboard, and staring at the sinks? We want our researcher to be inconspicuous—perhaps standing at one of the sinks pretending to put in contact lenses while secretly recording the relevant information. This type of observational study is called naturalistic observation : observing behavior in its natural setting. To better understand peer exclusion, Suzanne Fanger collaborated with colleagues at the University of Texas to observe the behavior of preschool children on a playground. How did the observers remain inconspicuous over the duration of the study? They equipped a few of the children with wireless microphones (which the children quickly forgot about) and observed while taking notes from a distance. Also, the children in that particular preschool (a “laboratory preschool”) were accustomed to having observers on the playground (Fanger, Frankel, & Hazen, 2012).

It is critical that the observer be as unobtrusive and as inconspicuous as possible: when people know they are being watched, they are less likely to behave naturally. If you have any doubt about this, ask yourself how your driving behavior might differ in two situations: In the first situation, you are driving down a deserted highway during the middle of the day; in the second situation, you are being followed by a police car down the same deserted highway ( Figure 2.7 ).

It should be pointed out that naturalistic observation is not limited to research involving humans. Indeed, some of the best-known examples of naturalistic observation involve researchers going into the field to observe various kinds of animals in their own environments. As with human studies, the researchers maintain their distance and avoid interfering with the animal subjects so as not to influence their natural behaviors. Scientists have used this technique to study social hierarchies and interactions among animals ranging from ground squirrels to gorillas. The information provided by these studies is invaluable in understanding how those animals organize socially and communicate with one another. The anthropologist Jane Goodall , for example, spent nearly five decades observing the behavior of chimpanzees in Africa ( Figure 2.8 ). As an illustration of the types of concerns that a researcher might encounter in naturalistic observation, some scientists criticized Goodall for giving the chimps names instead of referring to them by numbers—using names was thought to undermine the emotional detachment required for the objectivity of the study (McKie, 2010).

The greatest benefit of naturalistic observation is the validity , or accuracy, of information collected unobtrusively in a natural setting. Having individuals behave as they normally would in a given situation means that we have a higher degree of ecological validity, or realism, than we might achieve with other research approaches. Therefore, our ability to generalize the findings of the research to real-world situations is enhanced. If done correctly, we need not worry about people or animals modifying their behavior simply because they are being observed. Sometimes, people may assume that reality programs give us a glimpse into authentic human behavior. However, the principle of inconspicuous observation is violated as reality stars are followed by camera crews and are interviewed on camera for personal confessionals. Given that environment, we must doubt how natural and realistic their behaviors are.

The major downside of naturalistic observation is that they are often difficult to set up and control. In our restroom study, what if you stood in the restroom all day prepared to record people’s hand washing behavior and no one came in? Or, what if you have been closely observing a troop of gorillas for weeks only to find that they migrated to a new place while you were sleeping in your tent? The benefit of realistic data comes at a cost. As a researcher you have no control of when (or if) you have behavior to observe. In addition, this type of observational research often requires significant investments of time, money, and a good dose of luck.

Sometimes studies involve structured observation. In these cases, people are observed while engaging in set, specific tasks. An excellent example of structured observation comes from Strange Situation by Mary Ainsworth (you will read more about this in the chapter on lifespan development). The Strange Situation is a procedure used to evaluate attachment styles that exist between an infant and caregiver. In this scenario, caregivers bring their infants into a room filled with toys. The Strange Situation involves a number of phases, including a stranger coming into the room, the caregiver leaving the room, and the caregiver’s return to the room. The infant’s behavior is closely monitored at each phase, but it is the behavior of the infant upon being reunited with the caregiver that is most telling in terms of characterizing the infant’s attachment style with the caregiver.

Another potential problem in observational research is observer bias . Generally, people who act as observers are closely involved in the research project and may unconsciously skew their observations to fit their research goals or expectations. To protect against this type of bias, researchers should have clear criteria established for the types of behaviors recorded and how those behaviors should be classified. In addition, researchers often compare observations of the same event by multiple observers, in order to test inter-rater reliability : a measure of reliability that assesses the consistency of observations by different observers.

Often, psychologists develop surveys as a means of gathering data. Surveys are lists of questions to be answered by research participants, and can be delivered as paper-and-pencil questionnaires, administered electronically, or conducted verbally ( Figure 2.9 ). Generally, the survey itself can be completed in a short time, and the ease of administering a survey makes it easy to collect data from a large number of people.

Surveys allow researchers to gather data from larger samples than may be afforded by other research methods . A sample is a subset of individuals selected from a population , which is the overall group of individuals that the researchers are interested in. Researchers study the sample and seek to generalize their findings to the population. Generally, researchers will begin this process by calculating various measures of central tendency from the data they have collected. These measures provide an overall summary of what a typical response looks like. There are three measures of central tendency: mode, median, and mean. The mode is the most frequently occurring response, the median lies at the middle of a given data set, and the mean is the arithmetic average of all data points. Means tend to be most useful in conducting additional analyses like those described below; however, means are very sensitive to the effects of outliers, and so one must be aware of those effects when making assessments of what measures of central tendency tell us about a data set in question.

There is both strength and weakness of the survey in comparison to case studies. By using surveys, we can collect information from a larger sample of people. A larger sample is better able to reflect the actual diversity of the population, thus allowing better generalizability. Therefore, if our sample is sufficiently large and diverse, we can assume that the data we collect from the survey can be generalized to the larger population with more certainty than the information collected through a case study. However, given the greater number of people involved, we are not able to collect the same depth of information on each person that would be collected in a case study.

Another potential weakness of surveys is something we touched on earlier in this chapter: People don't always give accurate responses. They may lie, misremember, or answer questions in a way that they think makes them look good. For example, people may report drinking less alcohol than is actually the case.

Any number of research questions can be answered through the use of surveys. One real-world example is the research conducted by Jenkins, Ruppel, Kizer, Yehl, and Griffin (2012) about the backlash against the US Arab-American community following the terrorist attacks of September 11, 2001. Jenkins and colleagues wanted to determine to what extent these negative attitudes toward Arab-Americans still existed nearly a decade after the attacks occurred. In one study, 140 research participants filled out a survey with 10 questions, including questions asking directly about the participant’s overt prejudicial attitudes toward people of various ethnicities. The survey also asked indirect questions about how likely the participant would be to interact with a person of a given ethnicity in a variety of settings (such as, “How likely do you think it is that you would introduce yourself to a person of Arab-American descent?”). The results of the research suggested that participants were unwilling to report prejudicial attitudes toward any ethnic group. However, there were significant differences between their pattern of responses to questions about social interaction with Arab-Americans compared to other ethnic groups: they indicated less willingness for social interaction with Arab-Americans compared to the other ethnic groups. This suggested that the participants harbored subtle forms of prejudice against Arab-Americans, despite their assertions that this was not the case (Jenkins et al., 2012).

Archival Research

Some researchers gain access to large amounts of data without interacting with a single research participant. Instead, they use existing records to answer various research questions. This type of research approach is known as archival research . Archival research relies on looking at past records or data sets to look for interesting patterns or relationships.

For example, a researcher might access the academic records of all individuals who enrolled in college within the past ten years and calculate how long it took them to complete their degrees, as well as course loads, grades, and extracurricular involvement. Archival research could provide important information about who is most likely to complete their education, and it could help identify important risk factors for struggling students ( Figure 2.10 ).

In comparing archival research to other research methods, there are several important distinctions. For one, the researcher employing archival research never directly interacts with research participants. Therefore, the investment of time and money to collect data is considerably less with archival research. Additionally, researchers have no control over what information was originally collected. Therefore, research questions have to be tailored so they can be answered within the structure of the existing data sets. There is also no guarantee of consistency between the records from one source to another, which might make comparing and contrasting different data sets problematic.

Longitudinal and Cross-Sectional Research

Sometimes we want to see how people change over time, as in studies of human development and lifespan. When we test the same group of individuals repeatedly over an extended period of time, we are conducting longitudinal research. Longitudinal research is a research design in which data-gathering is administered repeatedly over an extended period of time. For example, we may survey a group of individuals about their dietary habits at age 20, retest them a decade later at age 30, and then again at age 40.

Another approach is cross-sectional research. In cross-sectional research , a researcher compares multiple segments of the population at the same time. Using the dietary habits example above, the researcher might directly compare different groups of people by age. Instead of studying a group of people for 20 years to see how their dietary habits changed from decade to decade, the researcher would study a group of 20-year-old individuals and compare them to a group of 30-year-old individuals and a group of 40-year-old individuals. While cross-sectional research requires a shorter-term investment, it is also limited by differences that exist between the different generations (or cohorts) that have nothing to do with age per se, but rather reflect the social and cultural experiences of different generations of individuals that make them different from one another.

To illustrate this concept, consider the following survey findings. In recent years there has been significant growth in the popular support of same-sex marriage. Many studies on this topic break down survey participants into different age groups. In general, younger people are more supportive of same-sex marriage than are those who are older (Jones, 2013). Does this mean that as we age we become less open to the idea of same-sex marriage, or does this mean that older individuals have different perspectives because of the social climates in which they grew up? Longitudinal research is a powerful approach because the same individuals are involved in the research project over time, which means that the researchers need to be less concerned with differences among cohorts affecting the results of their study.

Often longitudinal studies are employed when researching various diseases in an effort to understand particular risk factors. Such studies often involve tens of thousands of individuals who are followed for several decades. Given the enormous number of people involved in these studies, researchers can feel confident that their findings can be generalized to the larger population. The Cancer Prevention Study-3 (CPS-3) is one of a series of longitudinal studies sponsored by the American Cancer Society aimed at determining predictive risk factors associated with cancer. When participants enter the study, they complete a survey about their lives and family histories, providing information on factors that might cause or prevent the development of cancer. Then every few years the participants receive additional surveys to complete. In the end, hundreds of thousands of participants will be tracked over 20 years to determine which of them develop cancer and which do not.

Clearly, this type of research is important and potentially very informative. For instance, earlier longitudinal studies sponsored by the American Cancer Society provided some of the first scientific demonstrations of the now well-established links between increased rates of cancer and smoking (American Cancer Society, n.d.) ( Figure 2.11 ).

As with any research strategy, longitudinal research is not without limitations. For one, these studies require an incredible time investment by the researcher and research participants. Given that some longitudinal studies take years, if not decades, to complete, the results will not be known for a considerable period of time. In addition to the time demands, these studies also require a substantial financial investment. Many researchers are unable to commit the resources necessary to see a longitudinal project through to the end.

Research participants must also be willing to continue their participation for an extended period of time, and this can be problematic. People move, get married and take new names, get ill, and eventually die. Even without significant life changes, some people may simply choose to discontinue their participation in the project. As a result, the attrition rates, or reduction in the number of research participants due to dropouts, in longitudinal studies are quite high and increase over the course of a project. For this reason, researchers using this approach typically recruit many participants fully expecting that a substantial number will drop out before the end. As the study progresses, they continually check whether the sample still represents the larger population, and make adjustments as necessary.

As an Amazon Associate we earn from qualifying purchases.

This book may not be used in the training of large language models or otherwise be ingested into large language models or generative AI offerings without OpenStax's permission.

Want to cite, share, or modify this book? This book uses the Creative Commons Attribution License and you must attribute OpenStax.

Access for free at https://openstax.org/books/psychology-2e/pages/1-introduction

• Authors: Rose M. Spielman, William J. Jenkins, Marilyn D. Lovett
• Publisher/website: OpenStax
• Book title: Psychology 2e
• Publication date: Apr 22, 2020
• Location: Houston, Texas
• Book URL: https://openstax.org/books/psychology-2e/pages/1-introduction
• Section URL: https://openstax.org/books/psychology-2e/pages/2-2-approaches-to-research

© Jan 6, 2024 OpenStax. Textbook content produced by OpenStax is licensed under a Creative Commons Attribution License . The OpenStax name, OpenStax logo, OpenStax book covers, OpenStax CNX name, and OpenStax CNX logo are not subject to the Creative Commons license and may not be reproduced without the prior and express written consent of Rice University.

Using naturalistic inquiry to inform qualitative description

Affiliations.

• 1 University of Wollongong, Wollongong, New South Wales, Australia.
• 2 School of Nursing, University of Wollongong, Wollongong, New South Wales, Australia.
• PMID: 33855820
• DOI: 10.7748/nr.2021.e1788

Background: Choosing how to answer a research question requires an understanding of philosophical and theoretical assumptions and how these inform a study's methodology and methods. This can be a challenge for all researchers, but for novice researchers, such as doctoral candidates, this can feel like an overwhelming task. Ensuring there is clear alignment between philosophy, theory, methodology and methods is an essential part of the research process, that enables research to be undertaken with clarity and integrity. This alignment must be a good fit for the research aim, and to ensure the researcher's intrinsic values and beliefs do not affect the analysis undertaken.

Aim: To describe the alignment between qualitative description and naturalistic inquiry and how it was applied to a doctoral candidate's exploration of the meaning of safety for people with experience of admission to an acute mental health unit.

Discussion: Understanding the alignment between qualitative descriptive methodology and naturalistic inquiry provided a clear pathway for the doctoral candidate.

Conclusion: The assumptions that underpin a methodological approach need to be unpacked to understand how to answer a research question effectively.

Implications for practice: Qualitative description, informed by naturalistic inquiry, offers a practical way to explore and answer research questions.

Keywords: qualitative research; research; research methods; study design.

©2021 RCN Publishing Company Ltd. All rights reserved. Not to be copied, transmitted or recorded in any way, in whole or part, without prior permission of the publishers.

• Qualitative Research
• Research Design*
• Research Personnel*

Have a language expert improve your writing

Run a free plagiarism check in 10 minutes, automatically generate references for free.

• Knowledge Base
• Methodology
• Naturalistic Observation | Definition, Guide & Examples

Naturalistic Observation | Definition, Guide & Examples

Published on 6 May 2022 by Pritha Bhandari . Revised on 13 March 2023.

Naturalistic observation is a qualitative research method where you record the behaviours of your research subjects in real-world settings. You avoid interfering with or influencing any variables in a naturalistic observation.

You can think of naturalistic observation as ‘people watching’ with a purpose.

Table of contents

What is naturalistic observation, types of naturalistic observation methods, how to collect data, data sampling, advantages of naturalistic observation, disadvantages of naturalistic observation, frequently asked questions about naturalistic observation.

In naturalistic observations, you study your research subjects in their own environments to explore their behaviours without any outside influence or control. It’s a research method used in field studies.

Traditionally, naturalistic observation studies have been used by animal researchers, psychologists, ethnographers , and anthropologists. Naturalistic observations are helpful as a hypothesis -generating approach, because you gather rich information that can inspire further research.

Based on his naturalistic observations, he believed that these birds imprinted on the first potential parent in their surroundings, and they quickly learned to follow them and their actions.

Naturalistic observation is especially valuable for studying behaviours and actions that may not be replicable in controlled lab settings.

Prevent plagiarism, run a free check.

Naturalistic observations can be:

• Covert or overt: You either hide or reveal your identity as an observer to the participants you observe.
• Participant or non-participant: You participate in the activity or behaviour yourself, or you observe from the sidelines.

There are four main ways of using naturalistic observations.

Importantly, all of these take place in naturalistic settings rather than experimental laboratory settings. While you may actively participate in some types of observations, you refrain from influencing others or interfering with the activities you are observing too much.

You can use a variety of data collection methods for naturalistic observations.

Audiovisual recordings

Nowadays, it’s common to collect observations through audio and video recordings so you can revisit them at a later stage or share them with other trained observers. It’s best to place these recording devices discreetly so your participants aren’t distracted by them.

However, make sure you receive informed consent in a written format from each participant prior to recording them.

Note-taking

You can take notes while conducting naturalistic observations. Note down anything that seems relevant or important to you based on your research topic and interests in an unstructured way.

Tally counts

If you’re studying specific behaviours or events, it’s often helpful to make frequency counts of the number of times these occur during a certain time period. You can use a tally count to easily note down each instance that you observe in the moment.

There’s a lot of information you can collect when you conduct research in natural, uncontrolled environments. To simplify your data collection, you’ll often use data sampling.

Data sampling allows you to narrow down the focus of your data recording to specific times or events.

Time sampling

You record observations only at specific times. These time intervals can be randomly selected (e.g., at 8:03, 10:34, 12:51) or systematic (e.g., every 2 hours). You record whether your behaviours of interest occur during these time periods.

Event sampling

You record observations only when specific events occur. You may use a tally count to note the frequency of the event or take notes each time you see the event occurring.

Naturalistic observation is a valuable tool because of its flexibility, external validity, and suitability for research topics that can’t be studied in a lab.

Flexibility

Because naturalistic observation is a non-experimental method, you’re not bound to strict procedures. You can avoid using rigid protocols and also change your methods midway if you need to.

Ecological validity

Naturalistic observations are particularly high in ecological validity, because you use real life environments instead of lab settings. People don’t always act in the same ways in and outside the lab. Your participants behave in more authentic ways when they are unaware they’re being observed.

Naturalistic observations help you study topics that you can’t in the lab for ethical reasons.  You can also use technology to record conversations, behaviours, or other noise, provided you have consent or it’s otherwise ethically permissible .

The downsides of naturalistic observation include its lack of scientific control, ethical considerations, and potential for research bias from observers and subjects.

Lack of control

Since you perform research in natural environments, you can’t control the setting or any variables . Without this control, you won’t be able to draw conclusions about causal relationships . You also may not be able to replicate your findings in other contexts, with other people, or at other times.

Ethical considerations

Most people don’t want to be observed as they’re going about their day without their explicit consent or awareness. It’s important to always respect privacy and try to be unobtrusive. It’s also best to use naturalistic observations only in public situations where people expect they won’t be alone.

Observer bias

Because you indirectly collect data , there’s always a risk of observer bias in naturalistic observations. Your perceptions and interpretations of behaviour may be influenced by your own experiences and inaccurately represent the truth. This type of bias is particularly likely to occur in participant observation methods.

Subject bias

When you observe subjects in their natural environment, they may sometimes be aware they’re being observed. As a result, they may change their behaviours to act in more socially desirable ways to confirm your expectations.

Naturalistic observation is a qualitative research method where you record the behaviours of your research subjects in real-world settings. You avoid interfering or influencing anything in a naturalistic observation.

Naturalistic observation is a valuable tool because of its flexibility, external validity , and suitability for topics that can’t be studied in a lab setting.

The downsides of naturalistic observation include its lack of scientific control , ethical considerations , and potential for bias from observers and subjects.

Social desirability bias is the tendency for interview participants to give responses that will be viewed favourably by the interviewer or other participants. It occurs in all types of interviews and surveys , but is most common in semi-structured interviews , unstructured interviews , and focus groups .

Social desirability bias can be mitigated by ensuring participants feel at ease and comfortable sharing their views. Make sure to pay attention to your own body language and any physical or verbal cues, such as nodding or widening your eyes.

This type of bias in research can also occur in observations if the participants know they’re being observed. They might alter their behaviour accordingly.

You can use several tactics to minimise observer bias .

• Use masking (blinding) to hide the purpose of your study from all observers.
• Triangulate your data with different data collection methods or sources.
• Use multiple observers and ensure inter-rater reliability.
• Train your observers to make sure data is consistently recorded between them.
• Standardise your observation procedures to make sure they are structured and clear.

Cite this Scribbr article

If you want to cite this source, you can copy and paste the citation or click the ‘Cite this Scribbr article’ button to automatically add the citation to our free Reference Generator.

Bhandari, P. (2023, March 13). Naturalistic Observation | Definition, Guide & Examples. Scribbr. Retrieved 15 April 2024, from https://www.scribbr.co.uk/research-methods/naturalistic-observations/

Is this article helpful?

Pritha Bhandari

Other students also liked, qualitative vs quantitative research | examples & methods, descriptive research design | definition, methods & examples, what is an observational study | guide & examples.

• Table of Contents
• Random Entry
• Chronological
• Editorial Information
• About the SEP
• Editorial Board
• How to Cite the SEP
• Special Characters
• Advanced Tools
• Support the SEP
• PDFs for SEP Friends
• Make a Donation
• SEPIA for Libraries
• Entry Contents

Bibliography

Academic tools.

• Friends PDF Preview
• Author and Citation Info
• Back to Top

Naturalistic Approaches to Social Construction

Social “construction,” “constructionism” and “constructivism” are terms in wide use in the humanities and social sciences, and are applied to a diverse range of objects including the emotions, gender, race, sex, homo- and hetero-sexuality, mental illness, technology, quarks, facts, reality, and truth. This sort of terminology plays a number of different roles in different discourses, only some of which are philosophically interesting, and fewer of which admit of a “naturalistic” approach—an approach that treats science as a central and successful (if sometimes fallible) source of knowledge about the world. If there is any core idea of social constructionism, it is that some object or objects are caused or controlled by social or cultural factors rather than natural factors, and if there is any core motivation of such research, it is the aim of showing that such objects are or were under our control: they could be, or might have been, otherwise.

Determination of our representations of the world (including our ideas, concepts, beliefs, and theories of the world) by factors other than the way the world is may undermine our faith that any independent phenomena are represented or tracked, undermining the idea that there is a fact of the matter about which way of representing is correct. And determination of the non-representational facts of the world by our theories seems to reverse the “direction of fit” between representation and reality presupposed by our idea of successful epistemic activity. For both of these reasons, proponents and opponents of constructionist thought have held it to embody a challenge to the naturalism endemic in contemporary philosophy. But social constructionist themes can be and have been picked up by naturalists who hope to accommodate the interesting and important cultural phenomena documented by constructionist authors while denying more radical anti-scientific and anti-realist theses widely associated with social constructionism.

I begin by discussing social constructionism, and I then discuss some threads of contemporary naturalism. I go on to consider two different sorts of objects of social construction—representations and human traits—and discuss naturalistic, constructionist approaches to them.

1.1 What Constructs?

1.2 what is constructed, 1.3 what is it to construct, 2. naturalism and social construction, 3.1 the social construction of representations, 3.2 construction, human kinds and human traits, 4. conclusion, other internet resources, related entries, 1. what is social construction.

While constructionist claims often take the passive form of a declaration that “ Y is socially constructed,” it is more useful to think of social constructionist claims as having the form of a two-part relation:

X socially constructs Y .

We can then think of different accounts of social construction as differing in their accounts either of the relation itself, or of one or both relata.

While philosophers have carefully engaged various constructionist claims over the last several decades, much of the attention has been paid to various objects of construction (e.g., ideas? knowledge? facts? human nature?). In contrast, comparatively little attention has been paid to distinguishing different sorts of agents of construction. Many of the agents in social constructionist claims can be neatly divided into two groups: those that view the agents as primarily impersonal agents, and those that view the agents as personal agents (i.e., persons or groups).

Work in the first group emphasizes a causal role for impersonal causes like cultures, conventions, or institutions in producing some phenomenon. For example, the claim that what we perceive is determined by our background theories emphasizes an impersonal causal agent—culture—in determining some phenomena. Perhaps the most influential version of this claim came in Thomas Kuhn’s suggestion that, “what a man sees depends both upon what he looks at and also upon what his previous visual-conceptual experience has taught him to see” (1962/1970, 113), a suggestion with some foundation in “New Look ” psychology (e.g. Briner, Postman, and Rodrigues 1951). This view was subsequently taken up by a range of other authors across disciplines. For example, the historian Thomas Laqueur writes that, “powerful prior notions of difference or sameness determine what one sees and reports about the body” (1990, 21). [ 1 ] Provocative claims like Kuhn’s and Laqueur’s suggest that perception is so dependent upon the background theories that the observational data becomes compromised as an independent constraint on empirical inquiry. Impersonal cultural accounts of construction are also found in explanations of nonrepresentational phenomena, for example, of sex-differentiated behavior. Here a core claim might admit that there is sex difference, but claim that the cause of difference is rooted in different conceptions of sex (and the practices caused by those conceptions) rather than biological facts (see Feminist Perspectives on Sex and Gender ).

A second group of constructionist claims emphasizes personal social agents that construct through their choices. For example, Andrew Pickering’s (1984) influential work Constructing Quarks emphasizes the role of scientists’ judgments in a variety of roles in scientific process including, e.g., theory selection, experiment evaluation, assessments of research fecundity, and so forth, and such an emphasis on apparently highly contingent choices by researchers and scientific institutions is a mainstay of the social studies of knowledge literature. In emphasizing personal choices, some constructionist work (including some of Pickering’s) seems primarily aimed at emphasizing the contingency of the scientific theory that we come to accept (cf. Hacking 1999). [ 2 ] Other constructionists—those we might call critical constructionists—emphasize personal choices not just to establish the contingency of the acceptance of some representation as to emphasize the role of an agent’s interests or power relations in determining the content of an accepted representation. For example, Charles Mills suggests that the borders of American racial categories were determined in such a way as to “establish and maintain the privileges of different groups. So, for example, the motivation for using the one-drop rule to determine black racial membership is to maintain the subordination of the products of ‘miscegenation’” (1998, 48). And a range of constructionist research, especially research on human classifications like “race” and “gender,” documents shifts in human classification in response to shifts of interests or power.

Social constructionist claims are made about so many different objects that it is perhaps not surprising to find that such claims have different implications depending upon the different objects at which they are directed. Most uses of “construction”-talk (and related talk to the effect that that objects are, surprisingly, “invented” or “made up”) are directed at three very different sorts of entities: representations (e.g. ideas, theories, concepts, accounts, taxonomies, and so forth), (non-representational) facts quite generally, and a special sort of non-representational fact: facts about human traits.

Most philosophical discussion of social constructionism has been concerned with the so-called “science wars” which means that they have been concerned with evaluating the inference from the numerous and complex social influences operating in the production of scientific theories to the social construction of the facts those theories purport to represent, or to the failure of accounts of scientific rationality, or scientific realism, or scientific process (e.g. Laudan 1981, Nelson 1994, Fine 1996, Kukla 2000).

But “construction” talk has a more or less independent, but equally contentious life in the “human nature wars” where it labels the position that human traits (for example the emotions) or human kinds (which we can think of categories whose members share traits or clusters of traits, including, especially, dispositions to think and behave) are produced by culture rather than by biology or nature.

This kind constructionist view contrasts with the view that human kinds or traits are to be explained in terms of non-cultural mechanisms – especially internal, biological or natural states of the organism. The most pronounced disputes are prima facie concerned with whether the clustering of traits in, for example, sex difference, emotional behavior, or mental illness, are caused by a cultural practice of differentiating persons or are instead caused by natural processes operating in relative independence from culture.

But this kind constructionist view has also (especially in the philosophy of race) come to contrast with the skeptical view that a kind does not exist. In the context of race, constructionism amounts to the positive assertion that race is real even though it is not constituted by, or grounded in, biological facts such as genetic difference. (See, e.g., Haslanger 2012, Taylor 2013, Sundstrom 2002, Outlaw 1995, and the section “Race: Do Races Exist? Contemporary Philosophical Debates” in the entry on race .)

We consider naturalistic approaches to the construction of representations and human traits in more detail below, but it is useful to first distinguish global constructionist claims that hold that every fact is a social construction, from local constructionist claims that hold that only particular facts are. [ 3 ] Because of their provocative nature, many philosophers associate the term “social construction” with a global thesis, and a standard argument against global constructionism concerns whether such a program is sustainable in the face of the regress such a global thesis engenders regarding the thesis of constructionism itself (e.g. Boghossian 2006, Kukla 2000). Philosophers may have focused on these more radical claims in part because of the recognition that, relying on something like the general idea of construction sketched above, claims that are relatively global in scope are quite provocative and surprising while claims that would count as locally socially constructionist are quite familiar in many areas of philosophy, perhaps most importantly in meta-ethics, aesthetics, and social ontology. The domain of social ontology is especially interesting because here many facts are widely recognized as social constructions: for example, facts about being a U.S. Senator or a licensed dog are social constructions. [ 4 ] Call such constructions overt constructions. [ 5 ]

But even local constructionist claims can be interesting to the extent that they try to show some object may be produced by unacknowledged social practices—when they are covert constructions. This is the role that they play in the philosophy of psychiatry (Hacking 1995a, Scheff 1984, Showalter 1996, cf. Murphy 2006), the philosophy of the emotions (Averill 1980a, 1980b, Armon-Jones 1986, Harré 1986, cf. Griffiths 1997), the philosophy of race (e.g. Outlaw 1990, 1995; Mills 1998; Taylor 2013), and the philosophy of gender (see Feminist Theories of Sex and Gender: Gender as Socially Constructed ). Here the local claim that some kind (for example mental illness , emotion , race , or gender ) is explained by received culture or practice retains its interest because it offers a metaphysical alternative to other explanations (biological, religious, etc.) of the differential features of the kind members as well as an alternative to skepticism about the reality of the kind. [ 6 ]

We have already suggested that the core idea of constructionism is that some social agent produces or controls some object. Of course, “construction” talk is meant to evoke a variety of connotations that attend more paradigmatic construction: intentional activity, engaged in step-by step fashion, producing a designed, artifactual product. While different objects lead constructionist talk to be interpreted in different ways, we can distinguish two different sorts of relationship: causal or constitutive . [ 7 ] On the first, X constructs Y if Y is caused to come to exist, to continue to exist, or to have the properties that it does by X . On the second, Y is constructed if it is constituted by X ’s conceptual or social activity (perhaps even independently of X ’s causal influence on Y ).

The first, and more straightforward idea is causal construction :

X causally constructs Y if and only if X causes Y to exist or to persist or X controls the kind-typical properties of Y . [ 8 ]

There is no special problem posed by the claim that human social and linguistic activities cause certain things to exist or persist, or cause certain facts to be. More obscure is the idea that X ’s construction of Y is some sort of constitutive relationship. Many constructionist claims seem to involve the idea that the world is itself “made up” by social and cultural activities in ways that suggest our socio-linguistic behaviors are at least necessary to the object in question. This suggests a relationship such as:

X constitutively constructs Y if and only if X ’s conceptual or social activity regarding an individual y is metaphysically necessary for y to be a Y .

Consider the ways in which causal and constitutive claims might pull apart in a case of a socially produced artifact. Representations expressing the concept watch are normally causally necessary for some materials to come to have the intrinsic features of a watch, but they are not metaphysically necessary. It is metaphysically possible, however unlikely, that we could walk across a heath and find (something with the intrinsic features of) a watch that had “always been there.”

In contrast, the best candidates for constitutive construction are social facts:

For social facts, the attitude that we take toward the phenomenon is partly constitutive of the phenomenon … Part of being a cocktail party is being thought to be a cocktail party; part of being a war is being thought to be a war. This is a remarkable feature of social facts; it has no analogue among physical facts. (Searle 1995, 33–34)

On Searle’s view, a particular gathering of persons can be a cocktail party only with the conceptual and social recognition of those gathered. A similar idea has been influential in constructionist discussions. For example, the provocative claims that there were no homosexuals before the concept homosexual came to be expressed in Western culture in the nineteenth century (e.g. Foucault 1978, Halperin 1990) or that race is a modern invention (e.g Taylor 2004) seem to make sense if we see sexual or racial kinds as in part constituted by our concepts of them.

But Searle is right that there is something remarkable here, at least in the case of social facts: somehow our conceptual scheme or practice are necessary to make it true that some event instantiates cocktail party or war . What is wanted is, at a minimum, a model of this production—a model of exactly how the conceptual practice constitutes the fact. Perhaps the most obvious model to explain such constitutive claims is to hold that the relevant necessity is analytic , it holds in virtue of the meaning of the relevant term or concept. It is a fact about the meaning of “cocktail party” and perhaps “homosexual” and “race”) that it does not apply to a thing unless it is recognized to do so.

Whether any such meaning claims can be accommodated has been a contentious question since Quine (1953), but it is a question we can put aside for now (see The Analytic/Synthetic Distinction ). Instead, we should ask whether such model of constitutivity as analyticity is plausible for objects of social construction.

On the one hand, if Searle’s general account of social facts is correct, there may be many terms that operate like “cocktail party” in that the participants produce them only when they share certain intentional states about what they are doing. On the other hand, this does not seem plausible for the objects of many social constructionist claims. Remember, it is a mainstay of constructionist research to claim that social influence is exercised in surprising and provocative ways, especially on objects that we take to be produced naturally. But just this feature suggests that it cannot be part of our ordinary concepts of covertly constructed kinds that instances require our social-conceptual imprimatur to be members of these kinds (Machery 2014, Mallon 2017). This point is highlighted in a more general way by Paul Boghossian’s query:

isn’t it part of the very concept of an electron, or of a mountain, that these things were not constructed by us? Take electrons, for example. Is it not part of the very purpose of having such a concept that it is to designate things that are independent of us? (2006, 39)

If this is right, constructionists who view construction as a constitutive relation need another account of the necessity of our conceptual practice: it is implausible and inconsistent to claim that the necessity arises out of concept or word meanings in cases of covert construction.

There is a different model of necessity for the constructionist, however, which is to hold that the necessity in question is revealed a posteriori by our investigations of the phenomenon in question. Saul Kripke (1980), Hilary Putnam (1975) and others defended a causal theory of reference on which some terms (notably natural kind terms) referred to some sort of stuff or essence underlying the central uses of the term (see Reference: Causal Theories ). Crucially, however, because the reference relation is external, competent users of a term can be radically mistaken about what the term refers to and still successfully refer. In the case of water, for example, Putnam suggests that “water” picks out the sort of stuff that bears the appropriate causal-historical relation to paradigmatic instances in our own causal history (viz. H 2 O), and this was true even when we did not know what sort of stuff that was (i.e. before we knew the chemical structure). Kripke, Putnam, and others emphasized that claims such as “water=H 2 O” express necessary though a posteriori truths.

While the causal theory of reference (and its correct interpretation) remains controversial, in many quarters of philosophy it has become accepted wisdom. It is thus an option for interpreters of social constructionism to claim that certain terms—for example, “race”—actually refer to a kind that is produced by our socio-linguistic behavior, even if that fact is revealed only a posteriori . [ 9 ] Such a constitutive constructionist could grant, then, that it is part of our ordinary conception of the concept (e.g. of race) that – like electron – it refers to an independent, natural fact about the world, but such a constructionist would insist that further exploration of the world reveals that conventional features of our practice produce the object of our study. As with the case of “water” before modern chemistry, the conception widely associated with “race” (viz. that it is a biological kind) is wrong, but the term successfully refers all the same. Ideally, for such an approach to work, the constitutive constructionist would like an independent characterization of the sorts of social objects that investigation reveals to be identical with the kinds in question (e.g. Ásta 2016; Bach 2012; Mallon 2003, 2016), but they also need to fend off critics of applying the causal theory of reference in the context of reference to socially produced objects (e.g. Thomasson 2003) as well as more general critiques of employing theories of reference as premises in arguments with philosophically significant conclusions (Mallon et al. 2009, Mallon 2007b). Still, if it can be made to work, this strategy would make sense of constitutive constructionist claims while respecting Boghossian’s idea (one that is also central to constructionism) that these kinds are ordinarily believed to be natural and independent of us. For this reason, this strategy has been suggested in the case of race, gender, and other human kinds (Haslanger 2003, 2005; Mallon 2003, 2016), and more generally for scientific facts (Boyd 1992).

Of course, there may well be other models of necessity available. For example, it is sometimes suggested that a neo-Kantian interpretation of social constructionism is possible, an interpretation on which our socio-linguistic activities could provide a transcendental basis for any knowledge of the world. Such an interpretation might allow certain apparently radical constitutive claims, but the challenge would remain to reconcile the view with a naturalistic conception of ourselves, something such a proposal may fail to do (e.g. Boyd 1992, Rosen 1994).

Any discussion of naturalistic approaches to social construction is complicated by the fact that “naturalism” itself has no very widespread and uniform understanding (see Naturalism ). Still, the prospect seems provocative, in part, because social construction has come to be associated with a critical anti-realist attitude towards science.

Above, we identified naturalism with a certain attitude towards science, and for present purposes, we develop this idea by identifying three naturalistic attitudes toward science that have been picked up by naturalists addressing social constructionist themes.

• Accommodating Science: Most contemporary naturalists take science to be an enormously successful enterprise, and so other knowledge claims must either cohere with the findings of our best science or explain those findings away.
• Empiricism: Knowledge comes from careful study of the world, not a priori theorizing.
• Causal Modeling: The world is a set of entities related by natural laws. In attempting to understand it, we produce causal models that idealize these relationships to varying degrees.
• Supervenience: There are more and less fundamental entities, and the less fundamental depend on the more fundamental. Naturalists understand (at least) these fundamental entities to be natural (as opposed to supernatural). Naturalists typically hold these fundamental entities to be physical entities.
• Reduction: The regularities in which less fundamental entities participate are explained by natural laws governing the more fundamental entities upon which they supervene.
• Nonanomalism: Human beings and their products (e.g. culture or society) are natural things within the world that science explains. They are not metaphysically anomalous.
• Methodological Naturalism: In studying human nature, human culture, and social life, the methods of the natural sciences are to be employed.

These features characterize substantial threads of contemporary naturalist thought—threads that arise repeatedly in discussions of constructionism. Still, it is worth noting that something may be naturalistic in one sense but not another, and that the various threads we have characterized may sometimes be at odds. For example, rational choice explanations in economics might count as naturalist in that they attempt to reduce complex macro-level phenomena to simple, micro-level phenomena at the level of individuals (exhibiting some variety of metaphysical fundamentalism), and in the sense that they employ idealized causal modeling to do so (as in 1c). But they seem nonnaturalist insofar as they offer a highly idealized account of human behavior, one that seems frequently contradicted by the psychological facts about human reasoning (see, e.g., Nisbett and Ross 1980, Tversky and Kahneman, 1974) (against, perhaps, 1a and b, and 3).

We now review various naturalistic approaches to social construction, considering different sorts of entities in turn.

3. Naturalizing Social Construction

As we noted above, the production of facts by social agents poses no special problem for the naturalist where that production is understood causally, though naturalists of many stripes may want to produce causal models to show how the macro-level social phenomena of interest to many social theorists and social scientists are causally realized given what we know about, e.g. human nature or the causal structure of the universe. In contrast, constitutive claims of construction seem difficult to make sense of (except on an account of construction on which social activity involving a representation comes to produce and causally sustain an object that is referred to by that representation).

In recognition of this state of affairs, many naturalist approaches to constructed phenomena have involved attempts to causally model matters of interest to constructionists in ways that engage more or less completely with existing scientific knowledge. By way of illustrating such naturalistic approaches, I’ll discuss the social construction of representations and of human nature in more detail.

In talking about the construction of representations, we address the range of mental states, group beliefs, scientific theories, and other representations that express concepts or propositions. Such representations are, among other things, the vehicles of our thought as well as the means by which we store, organize, and further our knowledge of the world, and we do this in virtue of their role as bearers of meaning. A number of commentators have noted that many provocative constructionist claims are, in the first instance, claims that some sort of representation is constructed (e.g. Andreasen 1998, Hacking 1999, Haslanger 2012, Mallon 2004). Specifically, these are claims that social causes produce or control the selection of some representations with some meanings rather than others: for example, when Pickering (1984) writes of the construction of quarks or Laqueur (1990) suggests that sex is “made up,” they seem to be most directly addressing the process by which the theories of the quark or theories of sex are produced, viz. they are showing how a theory with one meaning was selected or endorsed rather than another theory or no theory at all. Where we limit the objects of constructionist claims to representations (such as theories), the claims cease to be particularly metaphysically provocative though detailed constructionist accounts of how certain representations came to be selected may still teach us much about science (e.g. Latour and Woolgar 1979l Collins and Pinch 2012).

In light of this, philosophers may be wont to diagnose some constructionist talk as a careless (or even an intentionally provocative) error of talking about the object of construction using a representation when one should be mentioning it (thereby expressing a view about the referent of the representation rather than the representation itself). When Claudius Ptolemy offered a geo-centric theory of the universe in the second century CE, he thereby contributed to the social construction of something: namely, a geocentric theory of the universe . We can talk about how and when that theory arose, and how it changed over time, but in doing so we are simply talking about a representation (or perhaps a lineage of related representations). It would be a mistake simply to slip from those claims into saying that in constructing this theory he thereby constructed a geocentric universe . Hence, charity in interpretation alone may suggest attributing only the weaker claim to a constructionist author. [ 10 ]

Still some constructionists endorse a stronger claim as well—that in constructing the theories, the facts described by those theories are thereby made to be. But if we leave at least the global versions of these additional claims aside as impossible to reconcile with naturalism, the distinctive feature of social constructionist explanations of representations is that they explain how we came to have those representations not by reference to the facts in the world they represent (as in realism), nor by reference to associations among our sensations (as in some forms of empiricism), nor by reference to innate knowledge or concepts (as in rationalism), nor by reference to the conditions of our thought or experience (as in transcendental arguments) but rather by reference to social and cultural background facts.

Naturalist work on constructionist approaches to representations can be grouped according to the debate the naturalist is addressing. Naturalists addressing the challenge posed by social construction to the authority of science have attempted to respond to this challenge in a variety of ways that pit various versions of realism and empiricism against constructionism (e.g. Boyd 1992; see Social Dimensions of Scientific Knowledge ). Because naturalists are typically committed to science as a central, if fallible, avenue of knowledge about the world (i.e. some variety of epistemic fundamentalism), naturalists will want to explain how this can be if, as social constructionists about scientific representations note, empirical observation is theory-laden and scientific theories are themselves subject to massive social influences.

For example, Jerry Fodor’s account of the modularity of perception (e.g. 1983, 1984, 1988) is, in part, a response to the implication that perception is so theory-laden that it lacks the independence required to constrain belief (see above for this implication in such diverse thinkers as Kuhn 1962/1970 and Laqueur 1990). Fodor suggests that sensory perception is modular by which he means (in part) “mandatory” and “informationally encapsulated” in its operations—i.e., it operates independently of our will and of our background theories and expectations. Fodor illustrates this effect by pointing to cases of optical illusions like the Muller-Lyer illusion (Fodor 1984). Here, two parallel line segments continue to appear to be different lengths even when one knows them be the same length, suggesting the independence of the process that produces sensory phenomena from one’s background theoretical beliefs. And while some philosophers (e.g. Churchland 1988, cf. Fodor 1988) have resisted this conclusion, some social scientists of knowledge have attempted to restate a constructionist view in ways that allow that Fodor may be correct. Barry Barnes, David Bloor and John Henry, for example, shift from emphasis on the determination of perceptual experience by culture to an emphasis on the underdetermination of belief by perceptual experience (a view which leaves room for cultural determination of belief) (1996, Ch. 1). More generally, epistemologists and philosophers of science have taken up the project of accommodating social influence in the production of knowledge, and this project is well underway in contemporary social epistemology and philosophy of science (e.g. Boyd 1992; Kitcher 1993, 2001). These issues are taken up elsewhere ( Social Epistemology ) so we address them no further here. Instead, I focus on attempts by naturalists to accommodate the cultural and personal processes at the heart of constructionist phenomena in naturalistic terms.

In contrast to naturalistic responses to the threat of scientific anti-realism, naturalistic responses to constructionist claims about representations (including beliefs) understood as human traits have been far more sympathetic to constructionist approaches. Indeed, an emphasis on the cultural and social causes of belief is quite amenable to range of naturalists, and naturalistic approaches to these causes are well represented in constructionist precursors, including such luminaries as Karl Marx, Friedrich Nietzsche (see the section on the critique of the descriptive component of MPS in Nietzsche’s Moral and Political Philosophy ), and Karl Mannheim (1936). In contemporary naturalistic philosophy of science and psychology, the naturalistic explanation of culturally produced cognition is picked up by at least three distinct strands of work taking up constructionist themes of culture. The first is centered on the idea that culture can be understood by analogy with population genetics, and that cultural items might be understood to be more or less successful based upon their success in spreading in a population. Various versions of this sentiment find expression in such diverse thinkers as Robert Boyd and Peter Richerson (1985, 2005a, 2005b), D.T. Campbell (1960), Luca Cavalli-Sforza and Marcus Feldman (1981), David Hull (1988), Jesse Prinz (2007, Ch. 6), Daniel Sperber (1996), and one version of it has a substantial popular following (Richard Dawkins’s (1976) widely read discussion of “memes”). While only some of these thinkers link the project to the understanding of constructionist research themes, the project in every case is to formally model cultural processes, understanding these complex processes as depending on simpler ones (See also Cultural Evolution .)

The second, overlapping strand of naturalistic inquiry also views culture as a system of representations upon which selection acts, but attempts to integrate this idea with the idea, common in evolutionary cognitive psychology, that the mind is comprised of a great many domain-specific mental mechanisms, and uses these as the selective mechanisms that act as a primary mechanism of selection (so called “massive modularity”; see Evolutionary Psychology: Massive Modularity ; cf. Carruthers 2006), and it is most firmly represented among cognitive anthropologists and psychologists like Scott Atran (1998), Pascal Boyer (1994, 2001), Laurence Hirschfeld (1996), and Daniel Sperber (1996). Such an approach represents naturalism in most (or perhaps all) of the above senses, and it is finding its way into the work of naturalist philosophers of science and psychology (Machery and Faucher 2005, Mallon 2013, Nichols 2002, Prinz 2007, Sripada 2006, Sterelny 2003).

A third, philosophically underdeveloped strand naturalizes crucial elements of critical constructionist approaches by suggesting the influence of sometimes implicit evaluations on judgments and theoretical activities. For example, a growing body of empirical evidence on so-called “motivated cognition” (cf. Kunda 1999) suggests mechanisms for (and some empirical validation of) the critical social constructionist tradition of explaining the content of accepted theories in part by appeal to the interests of the theorists.

Any sort of human trait could be an object of social construction, but many of the most interesting and contested cases are ones in which clusters of traits—traits that comprise human kinds—are purported to co-occur and to correlate with mental states, including dispositions to think and behave in particular ways. [ 11 ]

Because discussion of kinds of persons with dispositions to think and behave quickly gives rise to other questions about freedom of the will and social regulation, debates over constructionism about kinds are central to social and political debates regarding human categorization, including debates over sex and gender, race, emotions, hetero- and homo-sexuality, mental illness, and disability. Since the constructionist strategy explains a trait by appeal to highly contingent factors (including culture), partisans of these debates often come inquire whether a trait or cluster of traits is culturally specific, or can be found across cultures.

3.2.1 The Conceptual Project

These issues can quickly come to generate more heat than light, and so one role that philosophers in general, and naturalists in particular, have played is to carefully analyze constructionist positions and their alternatives. For example, in reflecting on debates over cultural specificity or universality, a number of commentators have noted that constructionist claims of cultural specificity often hinge not on genuine empirical disagreement about what is or is not found through history and across cultures, but also on a strategy of individuating the phenomena in question in ways that do or do not involve contextual features that vary across cultures (Mallon and Stich 2000; Boghossian 2006, 28; Pinker 2003, 38).

Philosophers have also distinguished claims of social construction from the possibility of cultural control (Mallon 2007a, Stein 1999), disentangled claims of social construction from claims of voluntariness and nonessentialism (Stein 1999), set out alternate forms of constructionism or anti-constructionism (Griffiths 1997, Mallon 2007c, Andreasen 1998), disentangled questions regarding the neural basis of a human kind from the innate/constructed dichotomy (Murphy 2006, Ch. 7) and so forth.

This conceptual project is a philosophical project par excellence , and it has contributed a great deal to clarifying just what conceptual and empirical issues are at stake in constructionist work.

3.2.2 Explaining the Development and Distribution of Human Traits

Naturalist interpretations of constructionism have also taken up the distinct, open-ended, empirical project of defending substantive claims regarding the development and distribution of human traits via the suggestions that human socio-linguistic behaviors shape human traits (including behavior) via different avenues, both developmental and situational.

One “social role” family of theories emphasizes the way that our socio-linguistic practices produce social roles that structure and shape human life and behavior. Perhaps the most influential philosophical project in this area has been Ian Hacking’s work on “making up people” (1986, 1992, 1995a, 1995b, 1998). In a series of papers and books, Hacking argues that the creation and promulgation of bureaucratic, technical, and medical classifications like “child abuse,” “multiple personality disorder,” and “fugue” create “new ways to be a person” (1995b, p. 239). The idea is that the conception of a certain kind of person shapes both a widespread social response (e.g. one that exculpates and perhaps encourages kind-typical behaviors), while at the same time, the conception shapes individual “performances” of the behavior in question (by suggesting highly specific avenues of behavior). On Hacking’s model, one he calls “the looping effect of human kinds,” the conception of the behavior may be part of an epistemic project of understanding a human kind that in turn gives rise to the clusters of traits that the theory represents (thereby providing epistemic support for the conception). [ 12 ] Much of Hacking’s own recent work has been aimed at providing detailed historical and cultural evidence that suggests that looping effects really are a feature of (at least modern) human social life, e.g. for the American epidemic of multiple personality disorder that started in the 1980s (Hacking 1995) or the European epidemic of fugue in the late nineteenth century (Hacking 1998). Hacking makes further claims about the “looping effect,” for example, that looping effects mark “a cardinal difference between the traditional natural and social sciences” because “the targets of the natural sciences are stationary” while “the targets of the social sciences are on the move” (1999, 108) ),claims that themselves have spurred lively discussions over the nature of looping effects (e.g. Cooper 2004, Laimann forthcoming) and of their mechanisms in human groups (e.g. Mallon 2016, Kuorikoski and Pöyhönen 2012).

Others have drawn on Hacking’s account to offer similar accounts of constructed kinds of person, including K. Anthony Appiah (1996) on racial identities, and Paul Griffiths (1997) on performed emotional syndromes. Together with Hacking’s work, these accounts provide partial, causal interpretation of even quite radical claims about kinds of person. For example, Judith Butler has provocatively claimed that the sex-differentiated behavior is a performance, writing, “That the gendered body is performative suggests that it has no ontological status apart from the various acts which constitute its reality. … In other words, acts and gestures, articulated and enacted desires create the illusion of an interior and organizing gender core…” (1990, 136). Following on the work of Hacking, Appiah, Griffiths, and others, we can naturalistically (re)interpret Butler’s claim as one that explains gender differences in actions, gestures, desires, and so on by reference to the social role that a person occupies. Such a causal model of the way in which social roles might shape behavior is at least arguably naturalistic in all of the above senses.

This “social role project” amounts to only one way of developing constructionist ideas in the service of explaining the development of human kinds, traits, or behaviors. For example, constructionist ideas find diverse manifestations in the theory of emotions (e.g. Armon-Jones 1986, Barrett 2017, Harré 1986, cf. Griffiths 1997 and Prinz 2004 for discussion). Because social constructionism offers a general set of explanatory approaches, constructionist approaches can be expected to reemerge in a variety of ways in the attempt to explain a wide range of human phenomena.

3.2.3 Formal Approaches to the Social Construction of Kinds

Still a different way of developing naturalistic constructionist accounts of kinds involves using various formal methods to model such kinds. Among recent work in social ontology, Francesco Guala has distinguished “rules-based” approaches to social institutions from “equilibrium-based” approaches (2016, xxv). The former attempts to understand social structure as emerging from the collective adoption of rules, while the latter sees it as emerging along with various solutions to coordination and cooperation problems. As an example of the former, Searle (1995) influentially argues that we can understand social institutions as brought into being by collective endorsement of rules of the form:

X counts as Y in C .

Here, “ X ” is a specification of the individual or type to which the status “ Y ” applies. And “ C ” specifies the context in which this imposition occurs. For instance, it might specify that tokens of a certain type produced by the U.S. mint count as money in the United States. Such statuses obtain in virtue of collective acceptance of one or more status functions. (See the entry on social ontology .)

In contrast, the latter family of approaches attempts to understand social structure by using the tools of economic and evolutionary game theory to understand culture (e.g. Bicchieri 2006, 2016; Guala 2016; O’Connor 2017). Here, norms, behaviors, and social regularities are seen as produced and stabilized by the preferences of individual actors making decisions in a social context of other actors. For example, Richard McElreath, Robert Boyd, and Peter Richerson (2003) have argued that ethnic-group based “markers” (e.g. things like styles of dress or other indicators of membership in an ethnic group) culturally evolved because they allowed actors to differentially interact with those who shared common norms, thus reaping the benefits of coordination and cooperation with greater efficiency.

While rules-based approaches have been much discussed across a range of philosophical fields (including metaphysics, social philosophy, empirically-informed philosophy of mind), equilibrium-based approaches have so far received comparatively little philosophical attention.

3.2.4 Human Kinds and Normativity

Many constructionist projects concerning human kinds are, or are pursued as part of, normative projects. Thinkers interested in gender, race, mental illness and disability, are often motivated not only by concern with the metaphysics of these categories, but with questions of social morality and justice that connect with them. For instance, Sally Haslanger’s work on the construction of gender and race (Haslanger 2012), or Elizabeth Barnes’s (2016) constructionist account of disability seem to essentially incorporate normative concepts. This connection, in turn, raises a number of further questions about why they are connected, and how we ought to understand their relationship.

One answer to these questions is simply that, once we understand the constructed nature of some category or phenomena, different normative conclusions will follow. For instance, some have emphasized that because constructionist explanations highlight the role of agents in the production or the sustenance of phenomena, they make those agents subject to moral evaluation (Kukla 2000; Mallon 2016, forthcoming).

A different approach might be that normative considerations ought to drive us towards certain metaphysical explanations. For instance, Esa Diaz-Leon (2015) has argued that constitutive constructionist explanations are politically better than causal constructionist ones, on the grounds that constitutive constructions are more tightly connected to our socio-conceptual practices:

revealing the constitutive connections between instantiating a certain category and standing in a certain relation to certain social practices, opens a clear path for social change: just change those social practices, and social change will automatically follow. (2015, 1145)

In contrast, Theresa Marques (2017) has argued that a focus on causal social construction is more relevant to projects of social justice. But if we see constructionism as a kind of explanation, then this debate can seem to put the cart before the horse. The correctness of an explanation is given by some facts in the world. Deciding what we would like those facts to be, given our aims, seems to fail to appreciate the reality of our socio-conceptual practices and their consequences.

More generally, while normative constructionist projects can be deeply engaged with our best scientific understanding, many naturalists will be tempted to attempt to distinguish descriptive and normative elements in order to engage them separately.

At the same time, ongoing naturalist work on human cooperation and coordination suggests the future possibility of more thoroughgoing naturalist approaches to construction that integrate naturalistic approaches to norms and normativity (e.g., Bicchieri 2016, Sripada 2006, and the entry on social norms ) with accounts of the human kinds that our socio-conceptual behaviors structure and shape.

The metaphor of “social construction” has proven remarkably supple in labeling and prompting a range of research across the social sciences and humanities, and the themes of personal and cultural causation taken up in this research are themselves of central concern. While most philosophical effort has gone towards the interpretation and refutation of provocative accounts of social construction arising especially out of studies in the history and sociology of science, social constructionist themes emerge across a host of other contexts, offering philosophical naturalists a range of alternate ways of engaging constructionist themes. Philosophical naturalists as well as working scientists have begun to take up this opportunity in ways that use the methods of philosophy and science to both state and evaluate social constructionist hypotheses (though not always under that label). Because of the powerful and central role culture plays in shaping human social environments, behaviors, identities and development, there is ample room for continuing and even expanding the pursuit of social constructionist themes within a naturalistic framework.

• Andreasen, R. O., 1998. “A New Perspective on the Race Debate,” British Journal of the Philosophy of Science , 49: 199–225.
• Appiah, K. A., 1996. “Race, Culture, Identity: Misunderstood Connections,” in Color Conscious: The Political Morality of Race , K. A. Appiah and A. Guttmann (eds.), Princeton, NJ: Princeton University Press. pp. 30–105.
• Armon-Jones, C., 1986. “The Thesis of Constructionism,” in The Social Construction of Emotion , R. Harrè (ed.), Oxford: Blackwell. pp. 32–55.
• Ásta, 2018. Categories We Live By , New York: Oxford University Press.
• Atran, S., 1998. “Folk Biology and the Anthropology of Science: Cognitive Universals and Cultural Particulars,” Behavioral and Brain Sciences , 21: 547–609.
• Averill, J., 1980a. “A constructivist view of emotion,” in Emotion: Theory, Research and Experience (Vol. 1: Theories of Emotion), R. Plutchik and H. Kellerman (eds.), New York: Academic Press. pp. 305–339.
• –––, 1980b. “Emotion and Anxiety: sociocultural, biological, and psychological determinants”, in Explaining Emotions , A. Rorty (ed.), Berkeley: University of California Press. pp. 37–72.
• Bach, T., 2012. “Gender Is a Natural Kind with a Historical Essence,” Ethics , 122(2): 231–272.
• Barnes, B., D. Bloor, et al., 1996. Scientific knowledge: a sociological analysis , London: Athlone.
• Barnes, E., 2016. The Minority Body: a theory of disability , New York: Oxford University Press.
• Barrett, L. F., 2017. How Emotions Are Made: The secret life of the brain , New York: Houghton Mifflin.
• Bicchieri, C., 2016. Norms in the wild : how to diagnose, measure, and change social norms , New York: Oxford University Press.
• –––, 2006. The Grammar of Society: the Nature and Dynamics of Social Norms , Cambridge: Cambridge University Press.
• Boghossian, P., 2006. Fear of Knowledge: Against Relativism and Constructivism , New York: Oxford University Press.
• Boyd, R., 1992. “Constructivism, Realism, and Philosophical Method,” in J. Earman (ed.), Inference, Explanation, and Other Frustrations: Essays in the Philosophy of Science , Berkeley and Los Angeles: University of California Press: 131–198.
• Boyd, R. and P. J. Richerson, 1985. Culture and the evolutionary process , Chicago: University of Chicago Press.
• –––, 2005. The origin and evolution of cultures , New York: Oxford University Press.
• Boyer, P., 1994. The naturalness of religious ideas: a cognitive theory of religion , Berkeley: University of California Press.
• –––, 2001. Religion explained: the human instincts that fashion gods, spirits and ancestors , London: Heinemann.
• Bruner, J., L. Postman, et al., 1951. “Expectation and the Perception of Color,” American Journal of Psychology , LXIV: 216–227.
• Butler, J., 1990. Gender Trouble , New York: Routledge.
• Campbell, D. T., 1960. “Blind Variation and Selective Retention in Creative Thought as in Other Knowledge Processes,” Psychological Review , 67: 380–400.
• Carruthers, P., 2006. The architecture of the mind: massive modularity and the flexibility of thought , Oxford: Oxford University Press.
• Cavalli-Sforza, L. L. and M. W. Feldman, 1981. Cultural transmission and evolution: a quantitative approach , Princeton, N.J.: Princeton University Press.
• Churchland, P., 1988. “Perceptual Plasticity and Theoretical Neutrality: A Reply to Jerry Fodor,” Philosophy of Science , 55(2): 167–187.
• Cooper, Rachel, 2004. “Why Hacking is wrong about human kinds,” British Journal for the Philosophy of Science , 55(1): 73–85.
• Diaz-Leon, E., 2015. “What Is Social Construction?” European Journal of Philosophy , 23(4): 1137–1152.
• Dawkins, R., 1976. The selfish gene , New York: Oxford University Press.
• Feldman, M. and L. Cavalli-Sforza, 1989. “On the theory of evolution under genetic and cultural transmission with application to the lactose absorption problem,” in Mathematical Evolutionary Theory , ed. M. Feldman. Princeton, NJ: Princeton University Press. pp. 145–173.
• Fine, A., 1996. “Science Made Up: Constructivist Sociology of Scientific Knowledge,” in The Disunity of Science: Boundaries, Contexts, and Power , ed. P. Galison and D. Stump. Stanford: Stanford University Press: 231–54.
• Fodor, J., 1984. “Observation Reconsidered,” Philosophy of Science , 51: 23–43.
• –––, 1988. “A Reply to Churchland’s ‘Perceptual Plasticity and Theoretical Neutrality’,” Philosophy of Science , 55(2): 188–198.
• –––, 1983. The Modularity of Mind , Cambridge, MA: MIT Press. A Bradford Book.
• Foucault, M., 1978. The History of Sexuality, Vol. I: An Introduction , New York: Pantheon.
• Griffith, Aaron M., forthcoming. “Social Construction and Grounding,” Philosophy and Phenomenological Research .
• Griffiths, P. E., 1997. What Emotions Really Are , Chicago: The University of Chicago Press.
• Guala, Francesco, 2016. Understanding Institutions: The Science and Philosophy of Living Together , Princeton: Princeton University Press.
• Hacking, I., 1986. “Making Up People,” in Reconstructing Individualism: Autonomy, Individuality, and the Self in Western Thought , T. C. Heller, M. Sosna and D. E. Wellbery. Stanford: Stanford University Press, 222–236.
• –––, 1992. “World Making by Kind Making: Child-Abuse for Example,” in How Classification Works: Nelson Goodman among the Social Sciences , ed. M. Douglas and D. Hull. Edinburgh: Edinburgh University Press: 180–238.
• –––, 1995a. Rewriting the Soul: Multiple Personality and the Sciences of Memory , Princeton, NJ: Princeton University Press.
• –––, 1995b. “The looping effects of human kinds,” in Causal Cognition: A Multidisciplinary Debate , ed. D. Sperber, D. Premack and A. J. Premack, New York: Clarendon Press: 351–394.
• –––, 1998. Mad Travelers: Reflections on the Reality of Transient Mental Illnesses , Charlottesville, VA: University Press of Virginia.
• –––, 1999. The Social Construction of What? Cambridge, MA: Harvard University Press.
• Halperin, D. M., 1990. One hundred years of homosexuality: and other essays on Greek love , New York: Routledge.
• Harré, R., 1986. “An outline of the social constructionist viewpoint,” in The Social Construction of Emotions , R. Harré, New York: Basil Blackwell: 2–14.
• –––, 1986. “Emotion talk across times: Accidie and melancholy in the psychological context,” in The Social Construction of Emotions , R. Harré. New York: Basil Blackwell: 220– 227.
• Haslanger, S., 1995. “Ontology and Social Construction,” Philosophical Topics , 23(2): 95–125.
• –––, 2003. “Social Construction: The ‘Debunking’ Project,” in Socializing Metaphysics: The Nature of Social Reality , F. Schmitt. Lanham, MD: Rowman and Littlefield: 301–325.
• –––, 2005. “What Are We Talking About? The Semantics and Politics of Social Kinds,” Hypatia , 20(4): 10–26.
• –––, 2012. Resisting reality : social construction and social critique , New York: Oxford University Press.
• Hirschfeld, L. A., 1996. Race in the Making: Cognition, Culture, and the Child’s Construction of Human Kinds , Cambridge, MA: MIT Press.
• Hull, D. L., 1988. Science as a process: an evolutionary account of the social and conceptual development of science , Chicago: University of Chicago Press.
• Kitcher, P., 1993. The advancement of science: science without legend, objectivity without illusions , New York: Oxford University Press.
• –––, 2001. Science, truth, and democracy , Oxford, New York: Oxford University Press.
• Kripke, S. A., 1980. Naming and Necessity , Cambridge, MA: Harvard University Press.
• Kuhn, T. S., 1962/1970. The Structure of Scientific Revolutions , Chicago: University of Chicago Press.
• Kukla, A., 2000. Social constructivism and the philosophy of science , London: Routledge.
• Kunda, Z., 1999. Social Cognition: Making Sense of People , Cambridge, MA: MIT Press.
• Kuorikoski, J. and S. Pöyhönen, 2012. “Looping Kinds and Social Mechanisms,” Sociological Theory , 30(3): 187–205.
• Laimann, Jessica, forthcoming. “Capricious Kinds,” British Journal for the Philosophy of Science , first online 14 March 2018. doi:10.1093/bjps/axy024
• Laqueur, T. W., 1990. Making sex: body and gender from the Greeks to Freud , Cambridge, MA: Harvard University Press.
• Laudan, L., 1981. “The pseudo-science of science?” Philosophy of the Social Sciences , 11: 173–98.
• Machery, E. (2014). “Social Ontology and the Objection from Reification,” in M. Gallotti & J. Michael (Eds.), Perspectives on Social Ontology and Social Cognition , Dordrecht: Springer, pp. 87–102.
• Machery, E. and L. Faucher, 2005. “Social Construction and the Concept of Race,” Philosophy of Science , 72: 1208–1219.
• Mallon, R., 2003. “Social Construction, Social Roles and Stability,” in Socializing Metaphysics , ed. F. Schmitt, Lanham, MD: Rowman and Littlefield: 327–353.
• –––, 2004. “Passing, Traveling, and Reality: Social Construction and the Metaphysics of Race,” Noûs 38(4): 644–673.
• –––, 2006. “‘Race’: Normative, Not Metaphysical or Semantic,” Ethics , 116(3): 525–551.
• –––, 2007a. “A Field Guide to Social Construction,” Philosophy Compass , 2(1): 93–108.
• –––, 2007b. “Arguments from Reference and the Worry About Dependence,” Midwest Studies in Philosophy , XXXI: 160–183.
• –––, 2007c. “Human Categories Beyond Non-Essentialism,” Journal of Political Philosophy , 15(2): 146–168.
• –––, 2016. The Construction of Human Kinds , New York: Oxford University Press.
• –––, 2017. “Social Roles and Reification,” Routledge Handbook of Social Mind , Julian Kiverstein (ed.), London: Routledge, pp. 431–446.
• Mallon, R., E. Machery, S. Nichols, and S. P. Stich, 2009. “Against Arguments from Reference” Philosophy and Phenomenological Research , 79(2): 332–356.
• Mallon, R. and S. P. Stich, 2000. “The Odd Couple: The Compatibility of Social Construction and Evolutionary Psychology,” Philosophy of Science , 67: 133–154.
• Mannheim, K., 1936. Ideology and utopia; an introduction to the sociology of knowledge , New York: Harcourt, Brace.
• Marques, Teresa, 2017. “The relevance of causal social construction,” Journal of Social Ontology , 3(1): 1–25.
• McElreath, R., Boyd, R., & Richerson, P., 2003. “Shared norms and the evolution of ethnic markers,” Current Anthropology , 44(1), 122–129.
• Mills, C., 1998. Blackness Visible: Essays on Philosophy and Race , Ithaca: Cornell University Press.
• Murphy, D., 2006. Psychiatry in the scientific image , Cambridge, MA: MIT Press.
• Nelson, A., 1994. “How Could Scientific Facts be Socially Constructed?” Studies in the History and Philosophy of Science , 25(4): 535–547.
• Nichols, S., 2002. “On the Genealogy of Norms: A Case for the Role of Emotion in Cultural Evolution,” Philosophy of Science , 69: 234–255.
• Nisbett, R. E. and L. Ross, 1980. Human inference: Strategies and shortcomings of social judgment , Englewood Cliffs, NJ: Prentice-Hall.
• O’Connor, Cailin, 2017. The cultural Red King effect, The Journal of Mathematical Sociology , 41:3, 155–171, DOI: 10.1080/0022250X.2017.1335723
• Outlaw, L., 1990. “Toward a Critical Theory of ‘Race’,” in Anatomy of Racism , ed. D. T. Goldberg, Minneapolis: University of Minnesota Press, 58–82.
• Outlaw, L., 1995. “On W.E.B. Du Bois’s ‘The Conservation of Races’,” in Overcoming Racism and Sexism , L. A. Bell and D. Blumenfeld (eds.), Lanham, MD: Rowman and Littlefield: 79–102.
• Pickering, A., 1984. Constructing quarks: a sociological history of particle physics , Edinburgh: Edinburgh University Press.
• Pinker, S., 2002. The Blank Slate: The Modern Denial of Human Nature , New York: Viking.
• Prinz, J. J., 2004. Gut reactions : a perceptual theory of emotion , New York ; Oxford, Oxford University Press.
• –––, 2007. The emotional construction of morals , Oxford, New York: Oxford University Press.
• Putnam, H., 1975. “The meaning of ‘meaning’,” Mind, Language and Reality: Philosophical Papers (Volume 2), New York: Cambridge University Press, pp. 215–271.
• Quine, W. V. O., 1953. “Two Dogmas of Empiricism,” in From a Logical Point of View , Cambridge, MA: Harvard University Press: 20–46.
• Rosen, G., 1994. “Objectivity and Modern Idealism: What is the Question?” in Philosophy in Mind: The Place of Philosophy in the Study of Mind , ed. M. Michael and J. O’Leary-Hawthorne, Boston: Kluwer Academic Publishers, 277–319.
• Schaffer, J., 2017. “Social construction as grounding; or: fundamentality for feminists, a reply to Barnes and Mikkola,” Philosophical Studies , 174(10): 2449–2465.
• Scheff, T., 1984. Being Mentally Ill: A Sociological Theory , New York: Aldine Publishing Co.
• Searle, J., 1995. The Construction of Social Reality , New York: The Free Press.
• Showalter, E., 1996. Hystories: Hysterical Epidemics and Modern Media , New York: Columbia University Press.
• Sperber, D., 1996. Explaining culture: a naturalistic approach , Cambridge, MA: Blackwell.
• Sripada, C., 2006. “Punishment and the strategic structure of moral systems,” Biology and Philosophy , 20: 767–789.
• Stein, E., 1999. The mismeasure of desire: the science, theory and ethics of sexual orientation , Oxford, New York: Oxford University Press.
• Sterelny, K., 2003. Thought in a hostile world: the evolution of human cognition , Malden, MA: Blackwell.
• Sundstrom, R., 2002. “Race as a Human Kind,” Philosophy and Social Criticism , 28(1): 91–115.
• Taylor, P. C., 2013. Race: a philosophical introduction , Malden, MA: Polity Press.
• Thomasson, A., 2003. “Realism and Human Kinds,” Philosophy and Phenomenological Research , 67(3): 580–609.
• Tversky, A. and D. Kahneman, 1974. “Judgment under uncertainty: Heuristics and biases,” Science , 185: 1124–1131.

How to cite this entry . Preview the PDF version of this entry at the Friends of the SEP Society . Look up topics and thinkers related to this entry at the Internet Philosophy Ontology Project (InPhO). Enhanced bibliography for this entry at PhilPapers , with links to its database.

• Boghossian, Paul, “ What is Social Construction? ”, undated, online manuscript.
• Flores, Laura, “ What is Social Construction ”, online, undated manuscript.

epistemology: social | evolution: cultural | feminist philosophy, topics: perspectives on sex and gender | innateness: and language | naturalism | Nietzsche, Friedrich: moral and political philosophy | psychology: evolutionary | race | scientific knowledge: social dimensions of | social norms | social ontology

Copyright © 2019 by Ron Mallon < rmallon @ wustl . edu >

• Accessibility

Support SEP

Mirror sites.

View this site from another server:

• Info about mirror sites

The Stanford Encyclopedia of Philosophy is copyright © 2023 by The Metaphysics Research Lab , Department of Philosophy, Stanford University

Library of Congress Catalog Data: ISSN 1095-5054

What is Qualitative in Qualitative Research

• Open access
• Published: 27 February 2019
• Volume 42 , pages 139–160, ( 2019 )

Cite this article

You have full access to this open access article

• Patrik Aspers 1 , 2 &
• Ugo Corte 3  

584k Accesses

283 Citations

24 Altmetric

Explore all metrics

What is qualitative research? If we look for a precise definition of qualitative research, and specifically for one that addresses its distinctive feature of being “qualitative,” the literature is meager. In this article we systematically search, identify and analyze a sample of 89 sources using or attempting to define the term “qualitative.” Then, drawing on ideas we find scattered across existing work, and based on Becker’s classic study of marijuana consumption, we formulate and illustrate a definition that tries to capture its core elements. We define qualitative research as an iterative process in which improved understanding to the scientific community is achieved by making new significant distinctions resulting from getting closer to the phenomenon studied. This formulation is developed as a tool to help improve research designs while stressing that a qualitative dimension is present in quantitative work as well. Additionally, it can facilitate teaching, communication between researchers, diminish the gap between qualitative and quantitative researchers, help to address critiques of qualitative methods, and be used as a standard of evaluation of qualitative research.

Similar content being viewed by others

What is Qualitative in Research

Patrik Aspers & Ugo Corte

Unsettling Definitions of Qualitative Research

Japonica Brown-Saracino

What is “Qualitative” in Qualitative Research? Why the Answer Does not Matter but the Question is Important

Mario L. Small

Avoid common mistakes on your manuscript.

If we assume that there is something called qualitative research, what exactly is this qualitative feature? And how could we evaluate qualitative research as good or not? Is it fundamentally different from quantitative research? In practice, most active qualitative researchers working with empirical material intuitively know what is involved in doing qualitative research, yet perhaps surprisingly, a clear definition addressing its key feature is still missing.

To address the question of what is qualitative we turn to the accounts of “qualitative research” in textbooks and also in empirical work. In his classic, explorative, interview study of deviance Howard Becker ( 1963 ) asks ‘How does one become a marijuana user?’ In contrast to pre-dispositional and psychological-individualistic theories of deviant behavior, Becker’s inherently social explanation contends that becoming a user of this substance is the result of a three-phase sequential learning process. First, potential users need to learn how to smoke it properly to produce the “correct” effects. If not, they are likely to stop experimenting with it. Second, they need to discover the effects associated with it; in other words, to get “high,” individuals not only have to experience what the drug does, but also to become aware that those sensations are related to using it. Third, they require learning to savor the feelings related to its consumption – to develop an acquired taste. Becker, who played music himself, gets close to the phenomenon by observing, taking part, and by talking to people consuming the drug: “half of the fifty interviews were conducted with musicians, the other half covered a wide range of people, including laborers, machinists, and people in the professions” (Becker 1963 :56).

Another central aspect derived through the common-to-all-research interplay between induction and deduction (Becker 2017 ), is that during the course of his research Becker adds scientifically meaningful new distinctions in the form of three phases—distinctions, or findings if you will, that strongly affect the course of his research: its focus, the material that he collects, and which eventually impact his findings. Each phase typically unfolds through social interaction, and often with input from experienced users in “a sequence of social experiences during which the person acquires a conception of the meaning of the behavior, and perceptions and judgments of objects and situations, all of which make the activity possible and desirable” (Becker 1963 :235). In this study the increased understanding of smoking dope is a result of a combination of the meaning of the actors, and the conceptual distinctions that Becker introduces based on the views expressed by his respondents. Understanding is the result of research and is due to an iterative process in which data, concepts and evidence are connected with one another (Becker 2017 ).

Indeed, there are many definitions of qualitative research, but if we look for a definition that addresses its distinctive feature of being “qualitative,” the literature across the broad field of social science is meager. The main reason behind this article lies in the paradox, which, to put it bluntly, is that researchers act as if they know what it is, but they cannot formulate a coherent definition. Sociologists and others will of course continue to conduct good studies that show the relevance and value of qualitative research addressing scientific and practical problems in society. However, our paper is grounded in the idea that providing a clear definition will help us improve the work that we do. Among researchers who practice qualitative research there is clearly much knowledge. We suggest that a definition makes this knowledge more explicit. If the first rationale for writing this paper refers to the “internal” aim of improving qualitative research, the second refers to the increased “external” pressure that especially many qualitative researchers feel; pressure that comes both from society as well as from other scientific approaches. There is a strong core in qualitative research, and leading researchers tend to agree on what it is and how it is done. Our critique is not directed at the practice of qualitative research, but we do claim that the type of systematic work we do has not yet been done, and that it is useful to improve the field and its status in relation to quantitative research.

The literature on the “internal” aim of improving, or at least clarifying qualitative research is large, and we do not claim to be the first to notice the vagueness of the term “qualitative” (Strauss and Corbin 1998 ). Also, others have noted that there is no single definition of it (Long and Godfrey 2004 :182), that there are many different views on qualitative research (Denzin and Lincoln 2003 :11; Jovanović 2011 :3), and that more generally, we need to define its meaning (Best 2004 :54). Strauss and Corbin ( 1998 ), for example, as well as Nelson et al. (1992:2 cited in Denzin and Lincoln 2003 :11), and Flick ( 2007 :ix–x), have recognized that the term is problematic: “Actually, the term ‘qualitative research’ is confusing because it can mean different things to different people” (Strauss and Corbin 1998 :10–11). Hammersley has discussed the possibility of addressing the problem, but states that “the task of providing an account of the distinctive features of qualitative research is far from straightforward” ( 2013 :2). This confusion, as he has recently further argued (Hammersley 2018 ), is also salient in relation to ethnography where different philosophical and methodological approaches lead to a lack of agreement about what it means.

Others (e.g. Hammersley 2018 ; Fine and Hancock 2017 ) have also identified the treat to qualitative research that comes from external forces, seen from the point of view of “qualitative research.” This threat can be further divided into that which comes from inside academia, such as the critique voiced by “quantitative research” and outside of academia, including, for example, New Public Management. Hammersley ( 2018 ), zooming in on one type of qualitative research, ethnography, has argued that it is under treat. Similarly to Fine ( 2003 ), and before him Gans ( 1999 ), he writes that ethnography’ has acquired a range of meanings, and comes in many different versions, these often reflecting sharply divergent epistemological orientations. And already more than twenty years ago while reviewing Denzin and Lincoln’ s Handbook of Qualitative Methods Fine argued:

While this increasing centrality [of qualitative research] might lead one to believe that consensual standards have developed, this belief would be misleading. As the methodology becomes more widely accepted, querulous challengers have raised fundamental questions that collectively have undercut the traditional models of how qualitative research is to be fashioned and presented (1995:417).

According to Hammersley, there are today “serious treats to the practice of ethnographic work, on almost any definition” ( 2018 :1). He lists five external treats: (1) that social research must be accountable and able to show its impact on society; (2) the current emphasis on “big data” and the emphasis on quantitative data and evidence; (3) the labor market pressure in academia that leaves less time for fieldwork (see also Fine and Hancock 2017 ); (4) problems of access to fields; and (5) the increased ethical scrutiny of projects, to which ethnography is particularly exposed. Hammersley discusses some more or less insufficient existing definitions of ethnography.

The current situation, as Hammersley and others note—and in relation not only to ethnography but also qualitative research in general, and as our empirical study shows—is not just unsatisfactory, it may even be harmful for the entire field of qualitative research, and does not help social science at large. We suggest that the lack of clarity of qualitative research is a real problem that must be addressed.

Towards a Definition of Qualitative Research

Seen in an historical light, what is today called qualitative, or sometimes ethnographic, interpretative research – or a number of other terms – has more or less always existed. At the time the founders of sociology – Simmel, Weber, Durkheim and, before them, Marx – were writing, and during the era of the Methodenstreit (“dispute about methods”) in which the German historical school emphasized scientific methods (cf. Swedberg 1990 ), we can at least speak of qualitative forerunners.

Perhaps the most extended discussion of what later became known as qualitative methods in a classic work is Bronisław Malinowski’s ( 1922 ) Argonauts in the Western Pacific , although even this study does not explicitly address the meaning of “qualitative.” In Weber’s ([1921–-22] 1978) work we find a tension between scientific explanations that are based on observation and quantification and interpretative research (see also Lazarsfeld and Barton 1982 ).

If we look through major sociology journals like the American Sociological Review , American Journal of Sociology , or Social Forces we will not find the term qualitative sociology before the 1970s. And certainly before then much of what we consider qualitative classics in sociology, like Becker’ study ( 1963 ), had already been produced. Indeed, the Chicago School often combined qualitative and quantitative data within the same study (Fine 1995 ). Our point being that before a disciplinary self-awareness the term quantitative preceded qualitative, and the articulation of the former was a political move to claim scientific status (Denzin and Lincoln 2005 ). In the US the World War II seem to have sparked a critique of sociological work, including “qualitative work,” that did not follow the scientific canon (Rawls 2018 ), which was underpinned by a scientifically oriented and value free philosophy of science. As a result the attempts and practice of integrating qualitative and quantitative sociology at Chicago lost ground to sociology that was more oriented to surveys and quantitative work at Columbia under Merton-Lazarsfeld. The quantitative tradition was also able to present textbooks (Lundberg 1951 ) that facilitated the use this approach and its “methods.” The practices of the qualitative tradition, by and large, remained tacit or was part of the mentoring transferred from the renowned masters to their students.

This glimpse into history leads us back to the lack of a coherent account condensed in a definition of qualitative research. Many of the attempts to define the term do not meet the requirements of a proper definition: A definition should be clear, avoid tautology, demarcate its domain in relation to the environment, and ideally only use words in its definiens that themselves are not in need of definition (Hempel 1966 ). A definition can enhance precision and thus clarity by identifying the core of the phenomenon. Preferably, a definition should be short. The typical definition we have found, however, is an ostensive definition, which indicates what qualitative research is about without informing us about what it actually is :

Qualitative research is multimethod in focus, involving an interpretative, naturalistic approach to its subject matter. This means that qualitative researchers study things in their natural settings, attempting to make sense of, or interpret, phenomena in terms of the meanings people bring to them. Qualitative research involves the studied use and collection of a variety of empirical materials – case study, personal experience, introspective, life story, interview, observational, historical, interactional, and visual texts – that describe routine and problematic moments and meanings in individuals’ lives. (Denzin and Lincoln 2005 :2)

Flick claims that the label “qualitative research” is indeed used as an umbrella for a number of approaches ( 2007 :2–4; 2002 :6), and it is not difficult to identify research fitting this designation. Moreover, whatever it is, it has grown dramatically over the past five decades. In addition, courses have been developed, methods have flourished, arguments about its future have been advanced (for example, Denzin and Lincoln 1994) and criticized (for example, Snow and Morrill 1995 ), and dedicated journals and books have mushroomed. Most social scientists have a clear idea of research and how it differs from journalism, politics and other activities. But the question of what is qualitative in qualitative research is either eluded or eschewed.

We maintain that this lacuna hinders systematic knowledge production based on qualitative research. Paul Lazarsfeld noted the lack of “codification” as early as 1955 when he reviewed 100 qualitative studies in order to offer a codification of the practices (Lazarsfeld and Barton 1982 :239). Since then many texts on “qualitative research” and its methods have been published, including recent attempts (Goertz and Mahoney 2012 ) similar to Lazarsfeld’s. These studies have tried to extract what is qualitative by looking at the large number of empirical “qualitative” studies. Our novel strategy complements these endeavors by taking another approach and looking at the attempts to codify these practices in the form of a definition, as well as to a minor extent take Becker’s study as an exemplar of what qualitative researchers actually do, and what the characteristic of being ‘qualitative’ denotes and implies. We claim that qualitative researchers, if there is such a thing as “qualitative research,” should be able to codify their practices in a condensed, yet general way expressed in language.

Lingering problems of “generalizability” and “how many cases do I need” (Small 2009 ) are blocking advancement – in this line of work qualitative approaches are said to differ considerably from quantitative ones, while some of the former unsuccessfully mimic principles related to the latter (Small 2009 ). Additionally, quantitative researchers sometimes unfairly criticize the first based on their own quality criteria. Scholars like Goertz and Mahoney ( 2012 ) have successfully focused on the different norms and practices beyond what they argue are essentially two different cultures: those working with either qualitative or quantitative methods. Instead, similarly to Becker ( 2017 ) who has recently questioned the usefulness of the distinction between qualitative and quantitative research, we focus on similarities.

The current situation also impedes both students and researchers in focusing their studies and understanding each other’s work (Lazarsfeld and Barton 1982 :239). A third consequence is providing an opening for critiques by scholars operating within different traditions (Valsiner 2000 :101). A fourth issue is that the “implicit use of methods in qualitative research makes the field far less standardized than the quantitative paradigm” (Goertz and Mahoney 2012 :9). Relatedly, the National Science Foundation in the US organized two workshops in 2004 and 2005 to address the scientific foundations of qualitative research involving strategies to improve it and to develop standards of evaluation in qualitative research. However, a specific focus on its distinguishing feature of being “qualitative” while being implicitly acknowledged, was discussed only briefly (for example, Best 2004 ).

In 2014 a theme issue was published in this journal on “Methods, Materials, and Meanings: Designing Cultural Analysis,” discussing central issues in (cultural) qualitative research (Berezin 2014 ; Biernacki 2014 ; Glaeser 2014 ; Lamont and Swidler 2014 ; Spillman 2014). We agree with many of the arguments put forward, such as the risk of methodological tribalism, and that we should not waste energy on debating methods separated from research questions. Nonetheless, a clarification of the relation to what is called “quantitative research” is of outmost importance to avoid misunderstandings and misguided debates between “qualitative” and “quantitative” researchers. Our strategy means that researchers, “qualitative” or “quantitative” they may be, in their actual practice may combine qualitative work and quantitative work.

In this article we accomplish three tasks. First, we systematically survey the literature for meanings of qualitative research by looking at how researchers have defined it. Drawing upon existing knowledge we find that the different meanings and ideas of qualitative research are not yet coherently integrated into one satisfactory definition. Next, we advance our contribution by offering a definition of qualitative research and illustrate its meaning and use partially by expanding on the brief example introduced earlier related to Becker’s work ( 1963 ). We offer a systematic analysis of central themes of what researchers consider to be the core of “qualitative,” regardless of style of work. These themes – which we summarize in terms of four keywords: distinction, process, closeness, improved understanding – constitute part of our literature review, in which each one appears, sometimes with others, but never all in the same definition. They serve as the foundation of our contribution. Our categories are overlapping. Their use is primarily to organize the large amount of definitions we have identified and analyzed, and not necessarily to draw a clear distinction between them. Finally, we continue the elaboration discussed above on the advantages of a clear definition of qualitative research.

In a hermeneutic fashion we propose that there is something meaningful that deserves to be labelled “qualitative research” (Gadamer 1990 ). To approach the question “What is qualitative in qualitative research?” we have surveyed the literature. In conducting our survey we first traced the word’s etymology in dictionaries, encyclopedias, handbooks of the social sciences and of methods and textbooks, mainly in English, which is common to methodology courses. It should be noted that we have zoomed in on sociology and its literature. This discipline has been the site of the largest debate and development of methods that can be called “qualitative,” which suggests that this field should be examined in great detail.

In an ideal situation we should expect that one good definition, or at least some common ideas, would have emerged over the years. This common core of qualitative research should be so accepted that it would appear in at least some textbooks. Since this is not what we found, we decided to pursue an inductive approach to capture maximal variation in the field of qualitative research; we searched in a selection of handbooks, textbooks, book chapters, and books, to which we added the analysis of journal articles. Our sample comprises a total of 89 references.

In practice we focused on the discipline that has had a clear discussion of methods, namely sociology. We also conducted a broad search in the JSTOR database to identify scholarly sociology articles published between 1998 and 2017 in English with a focus on defining or explaining qualitative research. We specifically zoom in on this time frame because we would have expect that this more mature period would have produced clear discussions on the meaning of qualitative research. To find these articles we combined a number of keywords to search the content and/or the title: qualitative (which was always included), definition, empirical, research, methodology, studies, fieldwork, interview and observation .

As a second phase of our research we searched within nine major sociological journals ( American Journal of Sociology , Sociological Theory , American Sociological Review , Contemporary Sociology , Sociological Forum , Sociological Theory , Qualitative Research , Qualitative Sociology and Qualitative Sociology Review ) for articles also published during the past 19 years (1998–2017) that had the term “qualitative” in the title and attempted to define qualitative research.

Lastly we picked two additional journals, Qualitative Research and Qualitative Sociology , in which we could expect to find texts addressing the notion of “qualitative.” From Qualitative Research we chose Volume 14, Issue 6, December 2014, and from Qualitative Sociology we chose Volume 36, Issue 2, June 2017. Within each of these we selected the first article; then we picked the second article of three prior issues. Again we went back another three issues and investigated article number three. Finally we went back another three issues and perused article number four. This selection criteria was used to get a manageable sample for the analysis.

The coding process of the 89 references we gathered in our selected review began soon after the first round of material was gathered, and we reduced the complexity created by our maximum variation sampling (Snow and Anderson 1993 :22) to four different categories within which questions on the nature and properties of qualitative research were discussed. We call them: Qualitative and Quantitative Research, Qualitative Research, Fieldwork, and Grounded Theory. This – which may appear as an illogical grouping – merely reflects the “context” in which the matter of “qualitative” is discussed. If the selection process of the material – books and articles – was informed by pre-knowledge, we used an inductive strategy to code the material. When studying our material, we identified four central notions related to “qualitative” that appear in various combinations in the literature which indicate what is the core of qualitative research. We have labeled them: “distinctions”, “process,” “closeness,” and “improved understanding.” During the research process the categories and notions were improved, refined, changed, and reordered. The coding ended when a sense of saturation in the material arose. In the presentation below all quotations and references come from our empirical material of texts on qualitative research.

Analysis – What is Qualitative Research?

In this section we describe the four categories we identified in the coding, how they differently discuss qualitative research, as well as their overall content. Some salient quotations are selected to represent the type of text sorted under each of the four categories. What we present are examples from the literature.

Qualitative and Quantitative

This analytic category comprises quotations comparing qualitative and quantitative research, a distinction that is frequently used (Brown 2010 :231); in effect this is a conceptual pair that structures the discussion and that may be associated with opposing interests. While the general goal of quantitative and qualitative research is the same – to understand the world better – their methodologies and focus in certain respects differ substantially (Becker 1966 :55). Quantity refers to that property of something that can be determined by measurement. In a dictionary of Statistics and Methodology we find that “(a) When referring to *variables, ‘qualitative’ is another term for *categorical or *nominal. (b) When speaking of kinds of research, ‘qualitative’ refers to studies of subjects that are hard to quantify, such as art history. Qualitative research tends to be a residual category for almost any kind of non-quantitative research” (Stiles 1998:183). But it should be obvious that one could employ a quantitative approach when studying, for example, art history.

The same dictionary states that quantitative is “said of variables or research that can be handled numerically, usually (too sharply) contrasted with *qualitative variables and research” (Stiles 1998:184). From a qualitative perspective “quantitative research” is about numbers and counting, and from a quantitative perspective qualitative research is everything that is not about numbers. But this does not say much about what is “qualitative.” If we turn to encyclopedias we find that in the 1932 edition of the Encyclopedia of the Social Sciences there is no mention of “qualitative.” In the Encyclopedia from 1968 we can read:

Qualitative Analysis. For methods of obtaining, analyzing, and describing data, see [the various entries:] CONTENT ANALYSIS; COUNTED DATA; EVALUATION RESEARCH, FIELD WORK; GRAPHIC PRESENTATION; HISTORIOGRAPHY, especially the article on THE RHETORIC OF HISTORY; INTERVIEWING; OBSERVATION; PERSONALITY MEASUREMENT; PROJECTIVE METHODS; PSYCHOANALYSIS, article on EXPERIMENTAL METHODS; SURVEY ANALYSIS, TABULAR PRESENTATION; TYPOLOGIES. (Vol. 13:225)

Some, like Alford, divide researchers into methodologists or, in his words, “quantitative and qualitative specialists” (Alford 1998 :12). Qualitative research uses a variety of methods, such as intensive interviews or in-depth analysis of historical materials, and it is concerned with a comprehensive account of some event or unit (King et al. 1994 :4). Like quantitative research it can be utilized to study a variety of issues, but it tends to focus on meanings and motivations that underlie cultural symbols, personal experiences, phenomena and detailed understanding of processes in the social world. In short, qualitative research centers on understanding processes, experiences, and the meanings people assign to things (Kalof et al. 2008 :79).

Others simply say that qualitative methods are inherently unscientific (Jovanović 2011 :19). Hood, for instance, argues that words are intrinsically less precise than numbers, and that they are therefore more prone to subjective analysis, leading to biased results (Hood 2006 :219). Qualitative methodologies have raised concerns over the limitations of quantitative templates (Brady et al. 2004 :4). Scholars such as King et al. ( 1994 ), for instance, argue that non-statistical research can produce more reliable results if researchers pay attention to the rules of scientific inference commonly stated in quantitative research. Also, researchers such as Becker ( 1966 :59; 1970 :42–43) have asserted that, if conducted properly, qualitative research and in particular ethnographic field methods, can lead to more accurate results than quantitative studies, in particular, survey research and laboratory experiments.

Some researchers, such as Kalof, Dan, and Dietz ( 2008 :79) claim that the boundaries between the two approaches are becoming blurred, and Small ( 2009 ) argues that currently much qualitative research (especially in North America) tries unsuccessfully and unnecessarily to emulate quantitative standards. For others, qualitative research tends to be more humanistic and discursive (King et al. 1994 :4). Ragin ( 1994 ), and similarly also Becker, ( 1996 :53), Marchel and Owens ( 2007 :303) think that the main distinction between the two styles is overstated and does not rest on the simple dichotomy of “numbers versus words” (Ragin 1994 :xii). Some claim that quantitative data can be utilized to discover associations, but in order to unveil cause and effect a complex research design involving the use of qualitative approaches needs to be devised (Gilbert 2009 :35). Consequently, qualitative data are useful for understanding the nuances lying beyond those processes as they unfold (Gilbert 2009 :35). Others contend that qualitative research is particularly well suited both to identify causality and to uncover fine descriptive distinctions (Fine and Hallett 2014 ; Lichterman and Isaac Reed 2014 ; Katz 2015 ).

There are other ways to separate these two traditions, including normative statements about what qualitative research should be (that is, better or worse than quantitative approaches, concerned with scientific approaches to societal change or vice versa; Snow and Morrill 1995 ; Denzin and Lincoln 2005 ), or whether it should develop falsifiable statements; Best 2004 ).

We propose that quantitative research is largely concerned with pre-determined variables (Small 2008 ); the analysis concerns the relations between variables. These categories are primarily not questioned in the study, only their frequency or degree, or the correlations between them (cf. Franzosi 2016 ). If a researcher studies wage differences between women and men, he or she works with given categories: x number of men are compared with y number of women, with a certain wage attributed to each person. The idea is not to move beyond the given categories of wage, men and women; they are the starting point as well as the end point, and undergo no “qualitative change.” Qualitative research, in contrast, investigates relations between categories that are themselves subject to change in the research process. Returning to Becker’s study ( 1963 ), we see that he questioned pre-dispositional theories of deviant behavior working with pre-determined variables such as an individual’s combination of personal qualities or emotional problems. His take, in contrast, was to understand marijuana consumption by developing “variables” as part of the investigation. Thereby he presented new variables, or as we would say today, theoretical concepts, but which are grounded in the empirical material.

Qualitative Research

This category contains quotations that refer to descriptions of qualitative research without making comparisons with quantitative research. Researchers such as Denzin and Lincoln, who have written a series of influential handbooks on qualitative methods (1994; Denzin and Lincoln 2003 ; 2005 ), citing Nelson et al. (1992:4), argue that because qualitative research is “interdisciplinary, transdisciplinary, and sometimes counterdisciplinary” it is difficult to derive one single definition of it (Jovanović 2011 :3). According to them, in fact, “the field” is “many things at the same time,” involving contradictions, tensions over its focus, methods, and how to derive interpretations and findings ( 2003 : 11). Similarly, others, such as Flick ( 2007 :ix–x) contend that agreeing on an accepted definition has increasingly become problematic, and that qualitative research has possibly matured different identities. However, Best holds that “the proliferation of many sorts of activities under the label of qualitative sociology threatens to confuse our discussions” ( 2004 :54). Atkinson’s position is more definite: “the current state of qualitative research and research methods is confused” ( 2005 :3–4).

Qualitative research is about interpretation (Blumer 1969 ; Strauss and Corbin 1998 ; Denzin and Lincoln 2003 ), or Verstehen [understanding] (Frankfort-Nachmias and Nachmias 1996 ). It is “multi-method,” involving the collection and use of a variety of empirical materials (Denzin and Lincoln 1998; Silverman 2013 ) and approaches (Silverman 2005 ; Flick 2007 ). It focuses not only on the objective nature of behavior but also on its subjective meanings: individuals’ own accounts of their attitudes, motivations, behavior (McIntyre 2005 :127; Creswell 2009 ), events and situations (Bryman 1989) – what people say and do in specific places and institutions (Goodwin and Horowitz 2002 :35–36) in social and temporal contexts (Morrill and Fine 1997). For this reason, following Weber ([1921-22] 1978), it can be described as an interpretative science (McIntyre 2005 :127). But could quantitative research also be concerned with these questions? Also, as pointed out below, does all qualitative research focus on subjective meaning, as some scholars suggest?

Others also distinguish qualitative research by claiming that it collects data using a naturalistic approach (Denzin and Lincoln 2005 :2; Creswell 2009 ), focusing on the meaning actors ascribe to their actions. But again, does all qualitative research need to be collected in situ? And does qualitative research have to be inherently concerned with meaning? Flick ( 2007 ), referring to Denzin and Lincoln ( 2005 ), mentions conversation analysis as an example of qualitative research that is not concerned with the meanings people bring to a situation, but rather with the formal organization of talk. Still others, such as Ragin ( 1994 :85), note that qualitative research is often (especially early on in the project, we would add) less structured than other kinds of social research – a characteristic connected to its flexibility and that can lead both to potentially better, but also worse results. But is this not a feature of this type of research, rather than a defining description of its essence? Wouldn’t this comment also apply, albeit to varying degrees, to quantitative research?

In addition, Strauss ( 2003 ), along with others, such as Alvesson and Kärreman ( 2011 :10–76), argue that qualitative researchers struggle to capture and represent complex phenomena partially because they tend to collect a large amount of data. While his analysis is correct at some points – “It is necessary to do detailed, intensive, microscopic examination of the data in order to bring out the amazing complexity of what lies in, behind, and beyond those data” (Strauss 2003 :10) – much of his analysis concerns the supposed focus of qualitative research and its challenges, rather than exactly what it is about. But even in this instance we would make a weak case arguing that these are strictly the defining features of qualitative research. Some researchers seem to focus on the approach or the methods used, or even on the way material is analyzed. Several researchers stress the naturalistic assumption of investigating the world, suggesting that meaning and interpretation appear to be a core matter of qualitative research.

We can also see that in this category there is no consensus about specific qualitative methods nor about qualitative data. Many emphasize interpretation, but quantitative research, too, involves interpretation; the results of a regression analysis, for example, certainly have to be interpreted, and the form of meta-analysis that factor analysis provides indeed requires interpretation However, there is no interpretation of quantitative raw data, i.e., numbers in tables. One common thread is that qualitative researchers have to get to grips with their data in order to understand what is being studied in great detail, irrespective of the type of empirical material that is being analyzed. This observation is connected to the fact that qualitative researchers routinely make several adjustments of focus and research design as their studies progress, in many cases until the very end of the project (Kalof et al. 2008 ). If you, like Becker, do not start out with a detailed theory, adjustments such as the emergence and refinement of research questions will occur during the research process. We have thus found a number of useful reflections about qualitative research scattered across different sources, but none of them effectively describe the defining characteristics of this approach.

Although qualitative research does not appear to be defined in terms of a specific method, it is certainly common that fieldwork, i.e., research that entails that the researcher spends considerable time in the field that is studied and use the knowledge gained as data, is seen as emblematic of or even identical to qualitative research. But because we understand that fieldwork tends to focus primarily on the collection and analysis of qualitative data, we expected to find within it discussions on the meaning of “qualitative.” But, again, this was not the case.

Instead, we found material on the history of this approach (for example, Frankfort-Nachmias and Nachmias 1996 ; Atkinson et al. 2001), including how it has changed; for example, by adopting a more self-reflexive practice (Heyl 2001), as well as the different nomenclature that has been adopted, such as fieldwork, ethnography, qualitative research, naturalistic research, participant observation and so on (for example, Lofland et al. 2006 ; Gans 1999 ).

We retrieved definitions of ethnography, such as “the study of people acting in the natural courses of their daily lives,” involving a “resocialization of the researcher” (Emerson 1988 :1) through intense immersion in others’ social worlds (see also examples in Hammersley 2018 ). This may be accomplished by direct observation and also participation (Neuman 2007 :276), although others, such as Denzin ( 1970 :185), have long recognized other types of observation, including non-participant (“fly on the wall”). In this category we have also isolated claims and opposing views, arguing that this type of research is distinguished primarily by where it is conducted (natural settings) (Hughes 1971:496), and how it is carried out (a variety of methods are applied) or, for some most importantly, by involving an active, empathetic immersion in those being studied (Emerson 1988 :2). We also retrieved descriptions of the goals it attends in relation to how it is taught (understanding subjective meanings of the people studied, primarily develop theory, or contribute to social change) (see for example, Corte and Irwin 2017 ; Frankfort-Nachmias and Nachmias 1996 :281; Trier-Bieniek 2012 :639) by collecting the richest possible data (Lofland et al. 2006 ) to derive “thick descriptions” (Geertz 1973 ), and/or to aim at theoretical statements of general scope and applicability (for example, Emerson 1988 ; Fine 2003 ). We have identified guidelines on how to evaluate it (for example Becker 1996 ; Lamont 2004 ) and have retrieved instructions on how it should be conducted (for example, Lofland et al. 2006 ). For instance, analysis should take place while the data gathering unfolds (Emerson 1988 ; Hammersley and Atkinson 2007 ; Lofland et al. 2006 ), observations should be of long duration (Becker 1970 :54; Goffman 1989 ), and data should be of high quantity (Becker 1970 :52–53), as well as other questionable distinctions between fieldwork and other methods:

Field studies differ from other methods of research in that the researcher performs the task of selecting topics, decides what questions to ask, and forges interest in the course of the research itself . This is in sharp contrast to many ‘theory-driven’ and ‘hypothesis-testing’ methods. (Lofland and Lofland 1995 :5)

But could not, for example, a strictly interview-based study be carried out with the same amount of flexibility, such as sequential interviewing (for example, Small 2009 )? Once again, are quantitative approaches really as inflexible as some qualitative researchers think? Moreover, this category stresses the role of the actors’ meaning, which requires knowledge and close interaction with people, their practices and their lifeworld.

It is clear that field studies – which are seen by some as the “gold standard” of qualitative research – are nonetheless only one way of doing qualitative research. There are other methods, but it is not clear why some are more qualitative than others, or why they are better or worse. Fieldwork is characterized by interaction with the field (the material) and understanding of the phenomenon that is being studied. In Becker’s case, he had general experience from fields in which marihuana was used, based on which he did interviews with actual users in several fields.

Grounded Theory

Another major category we identified in our sample is Grounded Theory. We found descriptions of it most clearly in Glaser and Strauss’ ([1967] 2010 ) original articulation, Strauss and Corbin ( 1998 ) and Charmaz ( 2006 ), as well as many other accounts of what it is for: generating and testing theory (Strauss 2003 :xi). We identified explanations of how this task can be accomplished – such as through two main procedures: constant comparison and theoretical sampling (Emerson 1998:96), and how using it has helped researchers to “think differently” (for example, Strauss and Corbin 1998 :1). We also read descriptions of its main traits, what it entails and fosters – for instance, an exceptional flexibility, an inductive approach (Strauss and Corbin 1998 :31–33; 1990; Esterberg 2002 :7), an ability to step back and critically analyze situations, recognize tendencies towards bias, think abstractly and be open to criticism, enhance sensitivity towards the words and actions of respondents, and develop a sense of absorption and devotion to the research process (Strauss and Corbin 1998 :5–6). Accordingly, we identified discussions of the value of triangulating different methods (both using and not using grounded theory), including quantitative ones, and theories to achieve theoretical development (most comprehensively in Denzin 1970 ; Strauss and Corbin 1998 ; Timmermans and Tavory 2012 ). We have also located arguments about how its practice helps to systematize data collection, analysis and presentation of results (Glaser and Strauss [1967] 2010 :16).

Grounded theory offers a systematic approach which requires researchers to get close to the field; closeness is a requirement of identifying questions and developing new concepts or making further distinctions with regard to old concepts. In contrast to other qualitative approaches, grounded theory emphasizes the detailed coding process, and the numerous fine-tuned distinctions that the researcher makes during the process. Within this category, too, we could not find a satisfying discussion of the meaning of qualitative research.

Defining Qualitative Research

In sum, our analysis shows that some notions reappear in the discussion of qualitative research, such as understanding, interpretation, “getting close” and making distinctions. These notions capture aspects of what we think is “qualitative.” However, a comprehensive definition that is useful and that can further develop the field is lacking, and not even a clear picture of its essential elements appears. In other words no definition emerges from our data, and in our research process we have moved back and forth between our empirical data and the attempt to present a definition. Our concrete strategy, as stated above, is to relate qualitative and quantitative research, or more specifically, qualitative and quantitative work. We use an ideal-typical notion of quantitative research which relies on taken for granted and numbered variables. This means that the data consists of variables on different scales, such as ordinal, but frequently ratio and absolute scales, and the representation of the numbers to the variables, i.e. the justification of the assignment of numbers to object or phenomenon, are not questioned, though the validity may be questioned. In this section we return to the notion of quality and try to clarify it while presenting our contribution.

Broadly, research refers to the activity performed by people trained to obtain knowledge through systematic procedures. Notions such as “objectivity” and “reflexivity,” “systematic,” “theory,” “evidence” and “openness” are here taken for granted in any type of research. Next, building on our empirical analysis we explain the four notions that we have identified as central to qualitative work: distinctions, process, closeness, and improved understanding. In discussing them, ultimately in relation to one another, we make their meaning even more precise. Our idea, in short, is that only when these ideas that we present separately for analytic purposes are brought together can we speak of qualitative research.

Distinctions

We believe that the possibility of making new distinctions is one the defining characteristics of qualitative research. It clearly sets it apart from quantitative analysis which works with taken-for-granted variables, albeit as mentioned, meta-analyses, for example, factor analysis may result in new variables. “Quality” refers essentially to distinctions, as already pointed out by Aristotle. He discusses the term “qualitative” commenting: “By a quality I mean that in virtue of which things are said to be qualified somehow” (Aristotle 1984:14). Quality is about what something is or has, which means that the distinction from its environment is crucial. We see qualitative research as a process in which significant new distinctions are made to the scholarly community; to make distinctions is a key aspect of obtaining new knowledge; a point, as we will see, that also has implications for “quantitative research.” The notion of being “significant” is paramount. New distinctions by themselves are not enough; just adding concepts only increases complexity without furthering our knowledge. The significance of new distinctions is judged against the communal knowledge of the research community. To enable this discussion and judgements central elements of rational discussion are required (cf. Habermas [1981] 1987 ; Davidsson [ 1988 ] 2001) to identify what is new and relevant scientific knowledge. Relatedly, Ragin alludes to the idea of new and useful knowledge at a more concrete level: “Qualitative methods are appropriate for in-depth examination of cases because they aid the identification of key features of cases. Most qualitative methods enhance data” (1994:79). When Becker ( 1963 ) studied deviant behavior and investigated how people became marihuana smokers, he made distinctions between the ways in which people learned how to smoke. This is a classic example of how the strategy of “getting close” to the material, for example the text, people or pictures that are subject to analysis, may enable researchers to obtain deeper insight and new knowledge by making distinctions – in this instance on the initial notion of learning how to smoke. Others have stressed the making of distinctions in relation to coding or theorizing. Emerson et al. ( 1995 ), for example, hold that “qualitative coding is a way of opening up avenues of inquiry,” meaning that the researcher identifies and develops concepts and analytic insights through close examination of and reflection on data (Emerson et al. 1995 :151). Goodwin and Horowitz highlight making distinctions in relation to theory-building writing: “Close engagement with their cases typically requires qualitative researchers to adapt existing theories or to make new conceptual distinctions or theoretical arguments to accommodate new data” ( 2002 : 37). In the ideal-typical quantitative research only existing and so to speak, given, variables would be used. If this is the case no new distinction are made. But, would not also many “quantitative” researchers make new distinctions?

Process does not merely suggest that research takes time. It mainly implies that qualitative new knowledge results from a process that involves several phases, and above all iteration. Qualitative research is about oscillation between theory and evidence, analysis and generating material, between first- and second -order constructs (Schütz 1962 :59), between getting in contact with something, finding sources, becoming deeply familiar with a topic, and then distilling and communicating some of its essential features. The main point is that the categories that the researcher uses, and perhaps takes for granted at the beginning of the research process, usually undergo qualitative changes resulting from what is found. Becker describes how he tested hypotheses and let the jargon of the users develop into theoretical concepts. This happens over time while the study is being conducted, exemplifying what we mean by process.

In the research process, a pilot-study may be used to get a first glance of, for example, the field, how to approach it, and what methods can be used, after which the method and theory are chosen or refined before the main study begins. Thus, the empirical material is often central from the start of the project and frequently leads to adjustments by the researcher. Likewise, during the main study categories are not fixed; the empirical material is seen in light of the theory used, but it is also given the opportunity to kick back, thereby resisting attempts to apply theoretical straightjackets (Becker 1970 :43). In this process, coding and analysis are interwoven, and thus are often important steps for getting closer to the phenomenon and deciding what to focus on next. Becker began his research by interviewing musicians close to him, then asking them to refer him to other musicians, and later on doubling his original sample of about 25 to include individuals in other professions (Becker 1973:46). Additionally, he made use of some participant observation, documents, and interviews with opiate users made available to him by colleagues. As his inductive theory of deviance evolved, Becker expanded his sample in order to fine tune it, and test the accuracy and generality of his hypotheses. In addition, he introduced a negative case and discussed the null hypothesis ( 1963 :44). His phasic career model is thus based on a research design that embraces processual work. Typically, process means to move between “theory” and “material” but also to deal with negative cases, and Becker ( 1998 ) describes how discovering these negative cases impacted his research design and ultimately its findings.

Obviously, all research is process-oriented to some degree. The point is that the ideal-typical quantitative process does not imply change of the data, and iteration between data, evidence, hypotheses, empirical work, and theory. The data, quantified variables, are, in most cases fixed. Merging of data, which of course can be done in a quantitative research process, does not mean new data. New hypotheses are frequently tested, but the “raw data is often the “the same.” Obviously, over time new datasets are made available and put into use.

Another characteristic that is emphasized in our sample is that qualitative researchers – and in particular ethnographers – can, or as Goffman put it, ought to ( 1989 ), get closer to the phenomenon being studied and their data than quantitative researchers (for example, Silverman 2009 :85). Put differently, essentially because of their methods qualitative researchers get into direct close contact with those being investigated and/or the material, such as texts, being analyzed. Becker started out his interview study, as we noted, by talking to those he knew in the field of music to get closer to the phenomenon he was studying. By conducting interviews he got even closer. Had he done more observations, he would undoubtedly have got even closer to the field.

Additionally, ethnographers’ design enables researchers to follow the field over time, and the research they do is almost by definition longitudinal, though the time in the field is studied obviously differs between studies. The general characteristic of closeness over time maximizes the chances of unexpected events, new data (related, for example, to archival research as additional sources, and for ethnography for situations not necessarily previously thought of as instrumental – what Mannay and Morgan ( 2015 ) term the “waiting field”), serendipity (Merton and Barber 2004 ; Åkerström 2013 ), and possibly reactivity, as well as the opportunity to observe disrupted patterns that translate into exemplars of negative cases. Two classic examples of this are Becker’s finding of what medical students call “crocks” (Becker et al. 1961 :317), and Geertz’s ( 1973 ) study of “deep play” in Balinese society.

By getting and staying so close to their data – be it pictures, text or humans interacting (Becker was himself a musician) – for a long time, as the research progressively focuses, qualitative researchers are prompted to continually test their hunches, presuppositions and hypotheses. They test them against a reality that often (but certainly not always), and practically, as well as metaphorically, talks back, whether by validating them, or disqualifying their premises – correctly, as well as incorrectly (Fine 2003 ; Becker 1970 ). This testing nonetheless often leads to new directions for the research. Becker, for example, says that he was initially reading psychological theories, but when facing the data he develops a theory that looks at, you may say, everything but psychological dispositions to explain the use of marihuana. Especially researchers involved with ethnographic methods have a fairly unique opportunity to dig up and then test (in a circular, continuous and temporal way) new research questions and findings as the research progresses, and thereby to derive previously unimagined and uncharted distinctions by getting closer to the phenomenon under study.

Let us stress that getting close is by no means restricted to ethnography. The notion of hermeneutic circle and hermeneutics as a general way of understanding implies that we must get close to the details in order to get the big picture. This also means that qualitative researchers can literally also make use of details of pictures as evidence (cf. Harper 2002). Thus, researchers may get closer both when generating the material or when analyzing it.

Quantitative research, we maintain, in the ideal-typical representation cannot get closer to the data. The data is essentially numbers in tables making up the variables (Franzosi 2016 :138). The data may originally have been “qualitative,” but once reduced to numbers there can only be a type of “hermeneutics” about what the number may stand for. The numbers themselves, however, are non-ambiguous. Thus, in quantitative research, interpretation, if done, is not about the data itself—the numbers—but what the numbers stand for. It follows that the interpretation is essentially done in a more “speculative” mode without direct empirical evidence (cf. Becker 2017 ).

Improved Understanding

While distinction, process and getting closer refer to the qualitative work of the researcher, improved understanding refers to its conditions and outcome of this work. Understanding cuts deeper than explanation, which to some may mean a causally verified correlation between variables. The notion of explanation presupposes the notion of understanding since explanation does not include an idea of how knowledge is gained (Manicas 2006 : 15). Understanding, we argue, is the core concept of what we call the outcome of the process when research has made use of all the other elements that were integrated in the research. Understanding, then, has a special status in qualitative research since it refers both to the conditions of knowledge and the outcome of the process. Understanding can to some extent be seen as the condition of explanation and occurs in a process of interpretation, which naturally refers to meaning (Gadamer 1990 ). It is fundamentally connected to knowing, and to the knowing of how to do things (Heidegger [1927] 2001 ). Conceptually the term hermeneutics is used to account for this process. Heidegger ties hermeneutics to human being and not possible to separate from the understanding of being ( 1988 ). Here we use it in a broader sense, and more connected to method in general (cf. Seiffert 1992 ). The abovementioned aspects – for example, “objectivity” and “reflexivity” – of the approach are conditions of scientific understanding. Understanding is the result of a circular process and means that the parts are understood in light of the whole, and vice versa. Understanding presupposes pre-understanding, or in other words, some knowledge of the phenomenon studied. The pre-understanding, even in the form of prejudices, are in qualitative research process, which we see as iterative, questioned, which gradually or suddenly change due to the iteration of data, evidence and concepts. However, qualitative research generates understanding in the iterative process when the researcher gets closer to the data, e.g., by going back and forth between field and analysis in a process that generates new data that changes the evidence, and, ultimately, the findings. Questioning, to ask questions, and put what one assumes—prejudices and presumption—in question, is central to understand something (Heidegger [1927] 2001 ; Gadamer 1990 :368–384). We propose that this iterative process in which the process of understanding occurs is characteristic of qualitative research.

Improved understanding means that we obtain scientific knowledge of something that we as a scholarly community did not know before, or that we get to know something better. It means that we understand more about how parts are related to one another, and to other things we already understand (see also Fine and Hallett 2014 ). Understanding is an important condition for qualitative research. It is not enough to identify correlations, make distinctions, and work in a process in which one gets close to the field or phenomena. Understanding is accomplished when the elements are integrated in an iterative process.

It is, moreover, possible to understand many things, and researchers, just like children, may come to understand new things every day as they engage with the world. This subjective condition of understanding – namely, that a person gains a better understanding of something –is easily met. To be qualified as “scientific,” the understanding must be general and useful to many; it must be public. But even this generally accessible understanding is not enough in order to speak of “scientific understanding.” Though we as a collective can increase understanding of everything in virtually all potential directions as a result also of qualitative work, we refrain from this “objective” way of understanding, which has no means of discriminating between what we gain in understanding. Scientific understanding means that it is deemed relevant from the scientific horizon (compare Schütz 1962 : 35–38, 46, 63), and that it rests on the pre-understanding that the scientists have and must have in order to understand. In other words, the understanding gained must be deemed useful by other researchers, so that they can build on it. We thus see understanding from a pragmatic, rather than a subjective or objective perspective. Improved understanding is related to the question(s) at hand. Understanding, in order to represent an improvement, must be an improvement in relation to the existing body of knowledge of the scientific community (James [ 1907 ] 1955). Scientific understanding is, by definition, collective, as expressed in Weber’s famous note on objectivity, namely that scientific work aims at truths “which … can claim, even for a Chinese, the validity appropriate to an empirical analysis” ([1904] 1949 :59). By qualifying “improved understanding” we argue that it is a general defining characteristic of qualitative research. Becker‘s ( 1966 ) study and other research of deviant behavior increased our understanding of the social learning processes of how individuals start a behavior. And it also added new knowledge about the labeling of deviant behavior as a social process. Few studies, of course, make the same large contribution as Becker’s, but are nonetheless qualitative research.

Understanding in the phenomenological sense, which is a hallmark of qualitative research, we argue, requires meaning and this meaning is derived from the context, and above all the data being analyzed. The ideal-typical quantitative research operates with given variables with different numbers. This type of material is not enough to establish meaning at the level that truly justifies understanding. In other words, many social science explanations offer ideas about correlations or even causal relations, but this does not mean that the meaning at the level of the data analyzed, is understood. This leads us to say that there are indeed many explanations that meet the criteria of understanding, for example the explanation of how one becomes a marihuana smoker presented by Becker. However, we may also understand a phenomenon without explaining it, and we may have potential explanations, or better correlations, that are not really understood.

We may speak more generally of quantitative research and its data to clarify what we see as an important distinction. The “raw data” that quantitative research—as an idealtypical activity, refers to is not available for further analysis; the numbers, once created, are not to be questioned (Franzosi 2016 : 138). If the researcher is to do “more” or “change” something, this will be done by conjectures based on theoretical knowledge or based on the researcher’s lifeworld. Both qualitative and quantitative research is based on the lifeworld, and all researchers use prejudices and pre-understanding in the research process. This idea is present in the works of Heidegger ( 2001 ) and Heisenberg (cited in Franzosi 2010 :619). Qualitative research, as we argued, involves the interaction and questioning of concepts (theory), data, and evidence.

Ragin ( 2004 :22) points out that “a good definition of qualitative research should be inclusive and should emphasize its key strengths and features, not what it lacks (for example, the use of sophisticated quantitative techniques).” We define qualitative research as an iterative process in which improved understanding to the scientific community is achieved by making new significant distinctions resulting from getting closer to the phenomenon studied. Qualitative research, as defined here, is consequently a combination of two criteria: (i) how to do things –namely, generating and analyzing empirical material, in an iterative process in which one gets closer by making distinctions, and (ii) the outcome –improved understanding novel to the scholarly community. Is our definition applicable to our own study? In this study we have closely read the empirical material that we generated, and the novel distinction of the notion “qualitative research” is the outcome of an iterative process in which both deduction and induction were involved, in which we identified the categories that we analyzed. We thus claim to meet the first criteria, “how to do things.” The second criteria cannot be judged but in a partial way by us, namely that the “outcome” —in concrete form the definition-improves our understanding to others in the scientific community.

We have defined qualitative research, or qualitative scientific work, in relation to quantitative scientific work. Given this definition, qualitative research is about questioning the pre-given (taken for granted) variables, but it is thus also about making new distinctions of any type of phenomenon, for example, by coining new concepts, including the identification of new variables. This process, as we have discussed, is carried out in relation to empirical material, previous research, and thus in relation to theory. Theory and previous research cannot be escaped or bracketed. According to hermeneutic principles all scientific work is grounded in the lifeworld, and as social scientists we can thus never fully bracket our pre-understanding.

We have proposed that quantitative research, as an idealtype, is concerned with pre-determined variables (Small 2008 ). Variables are epistemically fixed, but can vary in terms of dimensions, such as frequency or number. Age is an example; as a variable it can take on different numbers. In relation to quantitative research, qualitative research does not reduce its material to number and variables. If this is done the process of comes to a halt, the researcher gets more distanced from her data, and it makes it no longer possible to make new distinctions that increase our understanding. We have above discussed the components of our definition in relation to quantitative research. Our conclusion is that in the research that is called quantitative there are frequent and necessary qualitative elements.

Further, comparative empirical research on researchers primarily working with ”quantitative” approaches and those working with ”qualitative” approaches, we propose, would perhaps show that there are many similarities in practices of these two approaches. This is not to deny dissimilarities, or the different epistemic and ontic presuppositions that may be more or less strongly associated with the two different strands (see Goertz and Mahoney 2012 ). Our point is nonetheless that prejudices and preconceptions about researchers are unproductive, and that as other researchers have argued, differences may be exaggerated (e.g., Becker 1996 : 53, 2017 ; Marchel and Owens 2007 :303; Ragin 1994 ), and that a qualitative dimension is present in both kinds of work.

Several things follow from our findings. The most important result is the relation to quantitative research. In our analysis we have separated qualitative research from quantitative research. The point is not to label individual researchers, methods, projects, or works as either “quantitative” or “qualitative.” By analyzing, i.e., taking apart, the notions of quantitative and qualitative, we hope to have shown the elements of qualitative research. Our definition captures the elements, and how they, when combined in practice, generate understanding. As many of the quotations we have used suggest, one conclusion of our study holds that qualitative approaches are not inherently connected with a specific method. Put differently, none of the methods that are frequently labelled “qualitative,” such as interviews or participant observation, are inherently “qualitative.” What matters, given our definition, is whether one works qualitatively or quantitatively in the research process, until the results are produced. Consequently, our analysis also suggests that those researchers working with what in the literature and in jargon is often called “quantitative research” are almost bound to make use of what we have identified as qualitative elements in any research project. Our findings also suggest that many” quantitative” researchers, at least to some extent, are engaged with qualitative work, such as when research questions are developed, variables are constructed and combined, and hypotheses are formulated. Furthermore, a research project may hover between “qualitative” and “quantitative” or start out as “qualitative” and later move into a “quantitative” (a distinct strategy that is not similar to “mixed methods” or just simply combining induction and deduction). More generally speaking, the categories of “qualitative” and “quantitative,” unfortunately, often cover up practices, and it may lead to “camps” of researchers opposing one another. For example, regardless of the researcher is primarily oriented to “quantitative” or “qualitative” research, the role of theory is neglected (cf. Swedberg 2017 ). Our results open up for an interaction not characterized by differences, but by different emphasis, and similarities.

Let us take two examples to briefly indicate how qualitative elements can fruitfully be combined with quantitative. Franzosi ( 2010 ) has discussed the relations between quantitative and qualitative approaches, and more specifically the relation between words and numbers. He analyzes texts and argues that scientific meaning cannot be reduced to numbers. Put differently, the meaning of the numbers is to be understood by what is taken for granted, and what is part of the lifeworld (Schütz 1962 ). Franzosi shows how one can go about using qualitative and quantitative methods and data to address scientific questions analyzing violence in Italy at the time when fascism was rising (1919–1922). Aspers ( 2006 ) studied the meaning of fashion photographers. He uses an empirical phenomenological approach, and establishes meaning at the level of actors. In a second step this meaning, and the different ideal-typical photographers constructed as a result of participant observation and interviews, are tested using quantitative data from a database; in the first phase to verify the different ideal-types, in the second phase to use these types to establish new knowledge about the types. In both of these cases—and more examples can be found—authors move from qualitative data and try to keep the meaning established when using the quantitative data.

A second main result of our study is that a definition, and we provided one, offers a way for research to clarify, and even evaluate, what is done. Hence, our definition can guide researchers and students, informing them on how to think about concrete research problems they face, and to show what it means to get closer in a process in which new distinctions are made. The definition can also be used to evaluate the results, given that it is a standard of evaluation (cf. Hammersley 2007 ), to see whether new distinctions are made and whether this improves our understanding of what is researched, in addition to the evaluation of how the research was conducted. By making what is qualitative research explicit it becomes easier to communicate findings, and it is thereby much harder to fly under the radar with substandard research since there are standards of evaluation which make it easier to separate “good” from “not so good” qualitative research.

To conclude, our analysis, which ends with a definition of qualitative research can thus both address the “internal” issues of what is qualitative research, and the “external” critiques that make it harder to do qualitative research, to which both pressure from quantitative methods and general changes in society contribute.

Åkerström, Malin. 2013. Curiosity and serendipity in qualitative research. Qualitative Sociology Review 9 (2): 10–18.

Google Scholar  

Alford, Robert R. 1998. The craft of inquiry. Theories, methods, evidence . Oxford: Oxford University Press.

Alvesson, Mats, and Dan Kärreman. 2011. Qualitative research and theory development. Mystery as method . London: SAGE Publications.

Book   Google Scholar  

Aspers, Patrik. 2006. Markets in Fashion, A Phenomenological Approach. London Routledge.

Atkinson, Paul. 2005. Qualitative research. Unity and diversity. Forum: Qualitative Social Research 6 (3): 1–15.

Becker, Howard S. 1963. Outsiders. Studies in the sociology of deviance . New York: The Free Press.

Becker, Howard S. 1966. Whose side are we on? Social Problems 14 (3): 239–247.

Article   Google Scholar  

Becker, Howard S. 1970. Sociological work. Method and substance . New Brunswick: Transaction Books.

Becker, Howard S. 1996. The epistemology of qualitative research. In Ethnography and human development. Context and meaning in social inquiry , ed. Jessor Richard, Colby Anne, and Richard A. Shweder, 53–71. Chicago: University of Chicago Press.

Becker, Howard S. 1998. Tricks of the trade. How to think about your research while you're doing it . Chicago: University of Chicago Press.

Becker, Howard S. 2017. Evidence . Chigaco: University of Chicago Press.

Becker, Howard, Blanche Geer, Everett Hughes, and Anselm Strauss. 1961. Boys in White, student culture in medical school . New Brunswick: Transaction Publishers.

Berezin, Mabel. 2014. How do we know what we mean? Epistemological dilemmas in cultural sociology. Qualitative Sociology 37 (2): 141–151.

Best, Joel. 2004. Defining qualitative research. In Workshop on Scientific Foundations of Qualitative Research , eds . Charles, Ragin, Joanne, Nagel, and Patricia White, 53-54. http://www.nsf.gov/pubs/2004/nsf04219/nsf04219.pdf .

Biernacki, Richard. 2014. Humanist interpretation versus coding text samples. Qualitative Sociology 37 (2): 173–188.

Blumer, Herbert. 1969. Symbolic interactionism: Perspective and method . Berkeley: University of California Press.

Brady, Henry, David Collier, and Jason Seawright. 2004. Refocusing the discussion of methodology. In Rethinking social inquiry. Diverse tools, shared standards , ed. Brady Henry and Collier David, 3–22. Lanham: Rowman and Littlefield.

Brown, Allison P. 2010. Qualitative method and compromise in applied social research. Qualitative Research 10 (2): 229–248.

Charmaz, Kathy. 2006. Constructing grounded theory . London: Sage.

Corte, Ugo, and Katherine Irwin. 2017. “The Form and Flow of Teaching Ethnographic Knowledge: Hands-on Approaches for Learning Epistemology” Teaching Sociology 45(3): 209-219.

Creswell, John W. 2009. Research design. Qualitative, quantitative, and mixed method approaches . 3rd ed. Thousand Oaks: SAGE Publications.

Davidsson, David. 1988. 2001. The myth of the subjective. In Subjective, intersubjective, objective , ed. David Davidsson, 39–52. Oxford: Oxford University Press.

Denzin, Norman K. 1970. The research act: A theoretical introduction to Ssociological methods . Chicago: Aldine Publishing Company Publishers.

Denzin, Norman K., and Yvonna S. Lincoln. 2003. Introduction. The discipline and practice of qualitative research. In Collecting and interpreting qualitative materials , ed. Norman K. Denzin and Yvonna S. Lincoln, 1–45. Thousand Oaks: SAGE Publications.

Denzin, Norman K., and Yvonna S. Lincoln. 2005. Introduction. The discipline and practice of qualitative research. In The Sage handbook of qualitative research , ed. Norman K. Denzin and Yvonna S. Lincoln, 1–32. Thousand Oaks: SAGE Publications.

Emerson, Robert M., ed. 1988. Contemporary field research. A collection of readings . Prospect Heights: Waveland Press.

Emerson, Robert M., Rachel I. Fretz, and Linda L. Shaw. 1995. Writing ethnographic fieldnotes . Chicago: University of Chicago Press.

Esterberg, Kristin G. 2002. Qualitative methods in social research . Boston: McGraw-Hill.

Fine, Gary Alan. 1995. Review of “handbook of qualitative research.” Contemporary Sociology 24 (3): 416–418.

Fine, Gary Alan. 2003. “ Toward a Peopled Ethnography: Developing Theory from Group Life.” Ethnography . 4(1):41-60.

Fine, Gary Alan, and Black Hawk Hancock. 2017. The new ethnographer at work. Qualitative Research 17 (2): 260–268.

Fine, Gary Alan, and Timothy Hallett. 2014. Stranger and stranger: Creating theory through ethnographic distance and authority. Journal of Organizational Ethnography 3 (2): 188–203.

Flick, Uwe. 2002. Qualitative research. State of the art. Social Science Information 41 (1): 5–24.

Flick, Uwe. 2007. Designing qualitative research . London: SAGE Publications.

Frankfort-Nachmias, Chava, and David Nachmias. 1996. Research methods in the social sciences . 5th ed. London: Edward Arnold.

Franzosi, Roberto. 2010. Sociology, narrative, and the quality versus quantity debate (Goethe versus Newton): Can computer-assisted story grammars help us understand the rise of Italian fascism (1919- 1922)? Theory and Society 39 (6): 593–629.

Franzosi, Roberto. 2016. From method and measurement to narrative and number. International journal of social research methodology 19 (1): 137–141.

Gadamer, Hans-Georg. 1990. Wahrheit und Methode, Grundzüge einer philosophischen Hermeneutik . Band 1, Hermeneutik. Tübingen: J.C.B. Mohr.

Gans, Herbert. 1999. Participant Observation in an Age of “Ethnography”. Journal of Contemporary Ethnography 28 (5): 540–548.

Geertz, Clifford. 1973. The interpretation of cultures . New York: Basic Books.

Gilbert, Nigel. 2009. Researching social life . 3rd ed. London: SAGE Publications.

Glaeser, Andreas. 2014. Hermeneutic institutionalism: Towards a new synthesis. Qualitative Sociology 37: 207–241.

Glaser, Barney G., and Anselm L. Strauss. [1967] 2010. The discovery of grounded theory. Strategies for qualitative research. Hawthorne: Aldine.

Goertz, Gary, and James Mahoney. 2012. A tale of two cultures: Qualitative and quantitative research in the social sciences . Princeton: Princeton University Press.

Goffman, Erving. 1989. On fieldwork. Journal of Contemporary Ethnography 18 (2): 123–132.

Goodwin, Jeff, and Ruth Horowitz. 2002. Introduction. The methodological strengths and dilemmas of qualitative sociology. Qualitative Sociology 25 (1): 33–47.

Habermas, Jürgen. [1981] 1987. The theory of communicative action . Oxford: Polity Press.

Hammersley, Martyn. 2007. The issue of quality in qualitative research. International Journal of Research & Method in Education 30 (3): 287–305.

Hammersley, Martyn. 2013. What is qualitative research? Bloomsbury Publishing.

Hammersley, Martyn. 2018. What is ethnography? Can it survive should it? Ethnography and Education 13 (1): 1–17.

Hammersley, Martyn, and Paul Atkinson. 2007. Ethnography. Principles in practice . London: Tavistock Publications.

Heidegger, Martin. [1927] 2001. Sein und Zeit . Tübingen: Max Niemeyer Verlag.

Heidegger, Martin. 1988. 1923. Ontologie. Hermeneutik der Faktizität, Gesamtausgabe II. Abteilung: Vorlesungen 1919-1944, Band 63, Frankfurt am Main: Vittorio Klostermann.

Hempel, Carl G. 1966. Philosophy of the natural sciences . Upper Saddle River: Prentice Hall.

Hood, Jane C. 2006. Teaching against the text. The case of qualitative methods. Teaching Sociology 34 (3): 207–223.

James, William. 1907. 1955. Pragmatism . New York: Meredian Books.

Jovanović, Gordana. 2011. Toward a social history of qualitative research. History of the Human Sciences 24 (2): 1–27.

Kalof, Linda, Amy Dan, and Thomas Dietz. 2008. Essentials of social research . London: Open University Press.

Katz, Jack. 2015. Situational evidence: Strategies for causal reasoning from observational field notes. Sociological Methods & Research 44 (1): 108–144.

King, Gary, Robert O. Keohane, S. Sidney, and S. Verba. 1994. Designing social inquiry. In Scientific inference in qualitative research . Princeton: Princeton University Press.

Chapter   Google Scholar  

Lamont, Michelle. 2004. Evaluating qualitative research: Some empirical findings and an agenda. In Report from workshop on interdisciplinary standards for systematic qualitative research , ed. M. Lamont and P. White, 91–95. Washington, DC: National Science Foundation.

Lamont, Michèle, and Ann Swidler. 2014. Methodological pluralism and the possibilities and limits of interviewing. Qualitative Sociology 37 (2): 153–171.

Lazarsfeld, Paul, and Alan Barton. 1982. Some functions of qualitative analysis in social research. In The varied sociology of Paul Lazarsfeld , ed. Patricia Kendall, 239–285. New York: Columbia University Press.

Lichterman, Paul, and Isaac Reed I (2014), Theory and Contrastive Explanation in Ethnography. Sociological methods and research. Prepublished 27 October 2014; https://doi.org/10.1177/0049124114554458 .

Lofland, John, and Lyn Lofland. 1995. Analyzing social settings. A guide to qualitative observation and analysis . 3rd ed. Belmont: Wadsworth.

Lofland, John, David A. Snow, Leon Anderson, and Lyn H. Lofland. 2006. Analyzing social settings. A guide to qualitative observation and analysis . 4th ed. Belmont: Wadsworth/Thomson Learning.

Long, Adrew F., and Mary Godfrey. 2004. An evaluation tool to assess the quality of qualitative research studies. International Journal of Social Research Methodology 7 (2): 181–196.

Lundberg, George. 1951. Social research: A study in methods of gathering data . New York: Longmans, Green and Co..

Malinowski, Bronislaw. 1922. Argonauts of the Western Pacific: An account of native Enterprise and adventure in the archipelagoes of Melanesian New Guinea . London: Routledge.

Manicas, Peter. 2006. A realist philosophy of science: Explanation and understanding . Cambridge: Cambridge University Press.

Marchel, Carol, and Stephanie Owens. 2007. Qualitative research in psychology. Could William James get a job? History of Psychology 10 (4): 301–324.

McIntyre, Lisa J. 2005. Need to know. Social science research methods . Boston: McGraw-Hill.

Merton, Robert K., and Elinor Barber. 2004. The travels and adventures of serendipity. A Study in Sociological Semantics and the Sociology of Science . Princeton: Princeton University Press.

Mannay, Dawn, and Melanie Morgan. 2015. Doing ethnography or applying a qualitative technique? Reflections from the ‘waiting field‘. Qualitative Research 15 (2): 166–182.

Neuman, Lawrence W. 2007. Basics of social research. Qualitative and quantitative approaches . 2nd ed. Boston: Pearson Education.

Ragin, Charles C. 1994. Constructing social research. The unity and diversity of method . Thousand Oaks: Pine Forge Press.

Ragin, Charles C. 2004. Introduction to session 1: Defining qualitative research. In Workshop on Scientific Foundations of Qualitative Research , 22, ed. Charles C. Ragin, Joane Nagel, Patricia White. http://www.nsf.gov/pubs/2004/nsf04219/nsf04219.pdf

Rawls, Anne. 2018. The Wartime narrative in US sociology, 1940–7: Stigmatizing qualitative sociology in the name of ‘science,’ European Journal of Social Theory (Online first).

Schütz, Alfred. 1962. Collected papers I: The problem of social reality . The Hague: Nijhoff.

Seiffert, Helmut. 1992. Einführung in die Hermeneutik . Tübingen: Franke.

Silverman, David. 2005. Doing qualitative research. A practical handbook . 2nd ed. London: SAGE Publications.

Silverman, David. 2009. A very short, fairly interesting and reasonably cheap book about qualitative research . London: SAGE Publications.

Silverman, David. 2013. What counts as qualitative research? Some cautionary comments. Qualitative Sociology Review 9 (2): 48–55.

Small, Mario L. 2009. “How many cases do I need?” on science and the logic of case selection in field-based research. Ethnography 10 (1): 5–38.

Small, Mario L 2008. Lost in translation: How not to make qualitative research more scientific. In Workshop on interdisciplinary standards for systematic qualitative research, ed in Michelle Lamont, and Patricia White, 165–171. Washington, DC: National Science Foundation.

Snow, David A., and Leon Anderson. 1993. Down on their luck: A study of homeless street people . Berkeley: University of California Press.

Snow, David A., and Calvin Morrill. 1995. New ethnographies: Review symposium: A revolutionary handbook or a handbook for revolution? Journal of Contemporary Ethnography 24 (3): 341–349.

Strauss, Anselm L. 2003. Qualitative analysis for social scientists . 14th ed. Chicago: Cambridge University Press.

Strauss, Anselm L., and Juliette M. Corbin. 1998. Basics of qualitative research. Techniques and procedures for developing grounded theory . 2nd ed. Thousand Oaks: Sage Publications.

Swedberg, Richard. 2017. Theorizing in sociological research: A new perspective, a new departure? Annual Review of Sociology 43: 189–206.

Swedberg, Richard. 1990. The new 'Battle of Methods'. Challenge January–February 3 (1): 33–38.

Timmermans, Stefan, and Iddo Tavory. 2012. Theory construction in qualitative research: From grounded theory to abductive analysis. Sociological Theory 30 (3): 167–186.

Trier-Bieniek, Adrienne. 2012. Framing the telephone interview as a participant-centred tool for qualitative research. A methodological discussion. Qualitative Research 12 (6): 630–644.

Valsiner, Jaan. 2000. Data as representations. Contextualizing qualitative and quantitative research strategies. Social Science Information 39 (1): 99–113.

Weber, Max. 1904. 1949. Objectivity’ in social Science and social policy. Ed. Edward A. Shils and Henry A. Finch, 49–112. New York: The Free Press.

Download references

Acknowledgements

Financial Support for this research is given by the European Research Council, CEV (263699). The authors are grateful to Susann Krieglsteiner for assistance in collecting the data. The paper has benefitted from the many useful comments by the three reviewers and the editor, comments by members of the Uppsala Laboratory of Economic Sociology, as well as Jukka Gronow, Sebastian Kohl, Marcin Serafin, Richard Swedberg, Anders Vassenden and Turid Rødne.

Author information

Authors and affiliations.

Department of Sociology, Uppsala University, Uppsala, Sweden

Patrik Aspers

Seminar for Sociology, Universität St. Gallen, St. Gallen, Switzerland

Department of Media and Social Sciences, University of Stavanger, Stavanger, Norway

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Patrik Aspers .

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Reprints and permissions

About this article

Aspers, P., Corte, U. What is Qualitative in Qualitative Research. Qual Sociol 42 , 139–160 (2019). https://doi.org/10.1007/s11133-019-9413-7

Download citation

Published : 27 February 2019

Issue Date : 01 June 2019

DOI : https://doi.org/10.1007/s11133-019-9413-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Qualitative research
• Epistemology
• Philosophy of science
• Phenomenology
• Find a journal
• Publish with us
• Track your research

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 21 April 2024

Modeling children’s moral development in postwar Taiwan through naturalistic observations preserved in historical texts

• Zhining Sui 1 , 2 ,
• Qinyan Wang 3 , 4 &
• Jing Xu 5  

Scientific Reports volume  14 , Article number:  9140 ( 2024 ) Cite this article

Metrics details

• Human behaviour

A core issue in the interdisciplinary study of human morality is its ontogeny in diverse cultures, but systematic, naturalistic data in specific cultural contexts are rare to find. This study conducts a novel analysis of 213 children’s socio-moral behavior in a historical, non-Western, rural setting, based on a unique dataset of naturalistic observations from the first field research on Han Chinese children. Using multilevel multinomial modeling, we examined a range of proactive behaviors in 0-to-12-year-old children’s peer cooperation and conflict in an entire community in postwar Taiwan. We modeled the effects of age, sex, kinship, and behavioral roles, and revealed complex interactions between these four variables in shaping children’s moral development. We discovered linkages between coercive and non-coercive behaviors as children strategically negotiated leadership dynamics. We identified connections between prosocial and aggressive behaviors, illuminating the nuances of morality in real life. Our analysis also revealed gendered patterns and age-related trends that deviated from cultural norms and contradicted popular assumptions about Chinese family values. These findings highlight the importance of naturalistic observations in cultural contexts for understanding how we become moral persons. This re-analysis of historically significant fieldnotes also enriches the interdisciplinary study of child development across societies.

Introduction

The fundamental question of how we become moral persons has intrigued scientists and humanists for centuries. Childhood provides a unique window into human morality and its formation 1 , 2 , 3 . Despite recent progress in tracing the ontogeny of human moral sensibilities, thanks to interdisciplinary dialogues between psychology and anthropology 4 , 5 , researchers advocate the urgency to broaden our horizons and examine child development in diverse cultural contexts 6 . One reason is that Western-centered sampling biases still persist in developmental science 7 . Another problem is conceptual and methodological biases rooted in different disciplinary traditions: Psychologists approach children as “stubborn autodidacts” 8 and prioritize standardized experiments over studying the complexity and richness of children’s social life in natural contexts 9 . Cultural anthropologists, on the other hand, tend to view children as “passive assimilators” in their environment 10 , 11 and over-emphasize parenting and socialization, rather than children’s active learning 12 .

A promising direction to address these problems is systematic, naturalistic observations in cultural contexts because this approach can produce ecologically valid data on human behavior 13 . Existing observational research has mainly focused on school settings or parent-child interactions in Western, urban communities, therefore studying peer interactions in communal settings in rural, non-Western contexts is imperative 14 . Moreover, examining historical documents can inform the study of human cognition in the past and present 15 , 16 . Our research is a rare attempt that uses a Bayesian multilevel multinomial logistic model to analyze a significant set of historical texts that documented children’s socio-moral behavior in their everyday lives. These texts are part of what we call “the Wolf Archive,” ethnographic fieldnotes left behind by the renowned anthropologist Arthur Wolf, collected during his first fieldwork in Taiwan (1958–1960). Together with his then wife Margery Wolf, Arthur Wolf conducted the first anthropological research on culturally Chinese children in a village near Taipei at the height of Taiwan’s Martial-law era. Wolf’s original research replicated the Six Cultures Study of Child Socialization (”SCS” thereafter) 17 , 18 , a landmark project in the history of cross-cultural research 19 . The SCS teamed together anthropologists and psychologists, used a mixed-methods design in fieldwork among communities across six societies, and it has inspired the revival of cross-cultural developmental research today 20 . In particular, the systematic, naturalistic observation called Child Observation remains the SCS’s most enduring legacy 19 . Child Observation in Wolf’s research is of unique value as its methodology improved from the SCS in several aspects: excellent local research assistants, much longer fieldwork, complete household demographic information of the entire village, as well as its observation protocol (see “ Methods ” section).

We coded these fieldnotes and analyzed a diverse range of behaviors of 213 children below age 13 (calculated by the end of Wolf’s fieldwork), compared to 23 children (ages 3–11) per field site in the SCS. We designed a new behavioral coding system that took inspiration from but also differed from the SCS guide 18 . We combined deductive, top-down and inductive, bottom-up perspectives to better capture the complexity of children’s moral experience in their cultural contexts: Using a top-down approach, we included focal themes in existing literature, e.g., typical prosocial behaviors such as helping (instrumental help), sharing (resources), and comforting (emotional support) and aggressive behaviors such as physical aggression, verbal aggression, etc. Using a bottom-up approach, we added salient themes in local contexts, such as leading, dominating, scolding, tattling, giving a dirty look, requesting for comfort/help/sharing, requesting for access (to play), etc. We targeted this broad list of behaviors (Supplementary Table   S1 ) as recent theories suggest that human morality consists of multiple types of solutions to problems of cooperation recurrent in human social life, including reducing and resolving conflicts 21 .

Our study aims to understand how demographic factors influence children’s moral development in the cultural contexts of ethnic Han society. We measured individual differences in social behaviors and modeled the effects of age, sex, behavioral roles, and initiator-recipient kinship. We added analysis of recipients, whereas both the SCS and recent research on prosocial or aggressive behavior predominantly focused on initiators 22 , 23 , 24 . We also analyzed the binary variable of initiator-recipient co-residence, as children in this close-knit and high-fertility village often interacted with both siblings and other peers. Notably, most children lived with their biological siblings, in contrast to their parents’ generation when adoption was more common 25 . Finally, we modeled the interactions between different kinds of behaviors, e.g., cooperative and conflictual behaviors, as previous research identified co-development of these behaviors 26 .

The Wolf Archive provides a rare opportunity to examine moral development from infancy to middle childhood (0–12) in an entire community. Recent studies have found that various moral inclinations emerge in early childhood, some in infancy 5 , 27 , earlier than what classic theories characterized 2 , 28 . Middle-childhood is also an important phase, as previous research have identified cross-cultural variations 29 , 30 , more strategic motivations underlying prosocial behavior 31 , and increasing sensitivity to social norms 32 , 33 . However, age-related changes in prosocial behavior are complex: although a meta-analysis suggests that prosocial behaviors increase as children get older, the results depend on specific study designs and analyses 23 . Our study considers both initiators’ and recipients’ ages in naturally-occurring prosocial behaviors. Studies of aggression prioritize adolescents and elementary school children, as they enter a larger social world and develop more varied aggressive strategies 34 . However, recent studies have shown that physical aggression emerges early in infancy 35 , 36 . Our study captures the nuances by examining various types of aggression and their age-related trends from infancy to age 12. Also, in this rural community with dense social ties, we consider aggression together with prosocial and other types of behaviors, i.e., dominance, leading, etc., as children’s rich repertoire of strategies to regulate conflicts, facilitate cooperation, and negotiate social statuses 37 .

Gender/sex is another important factor, often examined together with age. We borrowed the terminology in the SCS and our original data to pay respect, using “sex” to refer to children’s biological sex, although we do not presume biological causes of sex-differences in behaviors. Although studies from Western samples showed a general trend of girls being more prosocial than boys 23 and that such sex differences grew more consistent with age 38 , recent cross-cultural research did not find uniform differences in prosocial behaviors between boys and girls or consistent patterns of gender-age interaction 39 . Research on aggression also revealed complexity: While boys tend to exhibit physical aggression more often than girls 24 , 40 , 41 , 42 , 43 , gender difference in indirect and relational forms of aggression showed mixed results 40 , including in cross-cultural work 41 , 43 . Ethnographic observations further complicate these patterns, especially considering age-sex interaction 22 . Honoring the SCS’ legacies, our study examined age-sex interaction in prosocial, aggressive, and other behaviors. But going beyond the SCS’ era, we can apply advanced statistical modeling methods to ethnographic data.

This is the first study to systematically examine the social behavior of an entire community of culturally Chinese children. Arthur and Margery Wolf’s previous research from this community helped establish the foundations for studying the traditional Chinese family 25 , 44 , 45 , 46 . However, children are not a focus in the study of Chinese families, despite the fact that childhood experience is critical in shaping core features of traditional Chinese families, such as gender biases and inequality 47 . Even Margery Wolf’s famous article on child-training 45 prioritized socialization values, and children existed passively, as an object of childrearing ideologies and in the shadow of parent-child ties. Therefore, our re-analysis not only can bring to light the obscured world of children’s social life, especially their peer interactions, but also examine the relationship between cultural values and behavioral reality.

In this regard, we will address three focal questions after comprehensively modeling all the behavioral data we coded: First, children’s leadership dynamics: How do children mobilize themselves into group activities, enact norms, impart moral knowledge, or establish authority in the process? The Wolfs’ works hardly ever examined this topic, but we found it ubiquitous in children’s everyday play, therefore including behaviors such as leading (non-coercive) and scolding (moral criticism) into our coding scheme, in addition to dominating (coercive) examined in the scarce observational research on Chinese children 48 . Second, age-related trends: Margery Wolf noted that, in this traditional Chinese community where age is an important factor in social hierarchy, caregivers used harsher discipline on older children but a more lenient approach towards younger children, as younger children were assumed to have little capacity for moral reasoning 45 . Do age-related behavioral trends conform to entrenched cultural expectations in Chinese societies from the past to the present, that older children become role models for younger children and yield to them during conflicts? 45 , 49 Finally, sex-differences: Are sex-differences in behaviors aligned with ascribed gender roles according to moral precepts, i.e., submissive girls? Margery Wolf’s classic ethnography based on this community has highlighted adult women’s agency despite their structurally subordinate position in the patrilineal, patriarchal Han Chinese family 50 . However, girls seemed docile and passive in her limited exploration of children’s world. Systematically examining behaviors in conflict situations might reveal patterns that diverge from these impressionistic observations. Taken together, our study highlights children’s agency in their self-organized social world, in contrast to the “passive-child” imagery rooted in earlier paradigms of the Chinese family.

Six decades later, the once village is now part of New Taipei City and it is impossible to replicate such systematic observations of children’s communal life 51 . Childhood in Taiwan, China, and East Asia more broadly has experienced profound changes as a result of rapid economic development, urbanization, and industrialization, together with fertility decline and transformation of family structures and values 52 , 53 . This re-analysis of old fieldnotes provides a rare reference to compare and contrast with contemporary East Asian childhoods, enriching the interdisciplinary study of child development in cultural contexts.

Descriptive statistics

Demographic variables in our study include the initiator’s age and sex, the recipient’s age and sex, and household numbers of initiators and recipients. Our sample includes 213 children from 70 households: 102 girls (mean age at the study’s outset = 4.23 years, SD = 3.01), 98 boys (mean age at the study’s outset = 4.55 years, SD = 2.72), plus 13 infants born during the 2-year fieldwork period (7 girls and 6 boys). On average, each child contributed 61.10 behavioral occurrences (SD = 52.08), each household 185.91 occurrences (SD = 144.14). Boys participated more in observed behaviors than girls, both as initiators and recipients (Supplementary Fig.   S1 ). The number of children per household varied from 1 to 9 (mean = 3.04 children, SD = 1.80): 32 households had more girls, 27 had more boys, and 11 had an equal distribution (Supplementary Fig.   S2 ).

Overall, leading emerged as the predominant behavior across children, irrespective of household status, and behaviors happened more frequently among children from the same household, except for comforting and requesting for comfort (Supplementary Fig.   S1 ). Comforting behavior typically occurs when a younger sibling seeks comfort from an older sibling through crying or whining. The unexpectedly higher number of observations of comforting behavior between children from different households arose from a few outlier dyads. Upon accounting for the number of unique pairs displaying each behavior, it became evident that the average number of observations per dyad (except for dirty looks ) was consistently higher when both children in the interaction were from the same household (Fig.   1 b). Furthermore, same-sex dyads interacted more frequently than cross-sex dyads, except for helping , tattling , and ownership assertion (Fig.   1 a). we also discovered initiator-recipient variations based on sex and household (Fig.   1 c).

Average number of observations per pair of children for each behavior (i.e., total number of observations/total number of unique pairs). ( a ) Comparison between children pairs from the same or different households. ( b ) Comparison between children pairs with the same or different sexes. ( c ) Comparison between children pairs from the same or different households, and of the same or different sexes. Exact cross-sex variations with sexes specified are shown in Supplementary Fig.   S3 .

Finally, we discovered significant variations in the average age of initiators and recipients for behaviors like comforting and requesting for comfort (Supplementary Fig.   S4 ), which can be explained by the typical sibling interactions. However, for behaviors such as requesting for sharing and supporting opinions , the difference in average age-at-observation between initiators and recipients was relatively small as these behaviors often happened between same-age peers.

Comparing models

We fitted four Multilevel Multinomial Behavioral Models 54 . Each of the models, Model_i, Model_iF, Model_ih, and Model_ihF, has different components (see “ Methods ” section). Model_ihF, including all the components, showed the highest level of support based on the Watanabe-Akaike information criterion (WAIC) comparison 55 (Table 1 ). The probability that this model will make the best predictions on new data relative to the other three models is 74.9%. Model_iF had 25.1% of the model weight, indicating that the inclusion of fixed effects improved the predictive performance. The inclusion of household-level random effects has a limited impact on the model’s performance. This finding is consistent with the overlapping standard deviation observed for Model_i and Model_ih and for Model_iF and Model_ihF (Supplementary Fig.   S5 ). Therefore we will mainly focus on the discussion of Model_i and Model_iF. Results from Model_ih and Model_ihF are included in Supplementary Information.

Individual variations in behaviors

Model_i includes only the intercept and individual random effects for initiators and recipients. We did not focus on the intercept coefficients because the predicted probabilities closely matched the corresponding percentages from the empirical data (Supplementary Fig.   S6 ). The extent of individual variation in exhibiting each behavior differed between initiators and recipients (Table 2 ). Several behaviors had relatively low variances in the initiator’s random effects, such as requesting for sharing and supporting opinions , suggesting that the probabilities of initiating these behaviors did not vary greatly among children. On the other hand, requesting for comfort had a notably higher variance in the initiator’s random effect, suggesting that a subset of children, especially the younger ones, were more likely to initiate this behavior. Moreover, there was a distinctively high variance in the recipient’s random effect for comforting . This implies that some children were more likely to receive voluntary comfort from their peers.

Model_iF included additional fixed effects from sex, age, and kinship of initiators and recipients based on Model_i, giving smaller variance estimates of the initiators’ random effects as compared to the variance estimates obtained in Model_i for almost all the behaviors (Table  2 ). This implies that the predictor variables accounted for the substantial individual-level variance of initiators in all behaviors, except for helping . Similarly, the recipients’ individual variations discovered in Model_i can be substantially explained by these predictor variables in all behaviors other than leading and taking .

Correlations between different behaviors

Model_i revealed correlations among individual random effects across behaviors for both initiators and recipients (Supplementary Table   S2 ), indicating how different behaviors are linked together for the same children. Note that all the probabilities reported in the following paragraphs are the relative probabilities as compared to the reference behavior, Ownership Assertion .

For initiators, correlations among seven behaviors were statistically significant. We partitioned them into two groups (Fig. 2 ), discovering positive correlations within each group but negative correlations across the two: one group consists of dominating , leading , scolding , and the other includes physical aggression , requesting for comfort , requesting for help , and verbal aggression . Within the first group of behaviors, children who were more likely to initiate leading behaviors also showed a propensity for dominating ( \(\rho _{4,2}=0.438\) ) and scolding ( \(\rho _{11,4}=0.451\) ), and dominating others was positively correlated with scolding ( \(\rho _{11,2}=0.358\) ). For behaviors in the second group, children prone to initiating physical aggression also resort to verbal insults when provoked ( \(\rho _{18,6}=0.456\) ). Notably, we also discovered a counter-intuitive positive correlation between initiating physical aggression and requesting comfort ( \(\rho _{8,6}=0.422\) ), highlighting the simultaneous occurrence of anti-social and prosocial behaviors among the same children. Besides, a positive correlation between requesting comfort and requesting help ( \(\rho _{9,8}=0.501\) ) among the same initiators reflects a connection between the expression of instrumental and emotional needs. Behaviors across the two groups were negatively correlated. For the same children, the probability of leading was negatively correlated with physical aggression ( \(\rho _{6,4}=-0.321\) ) and requesting comfort ( \(\rho _{8,4}=-0.426\) ). Scolding was negatively correlated with requesting comfort ( \(\rho _{11,8}=-0.402\) ) and initiating aggression, both physically ( \(\rho _{11,6}=-0.322\) ) and verbally ( \(\rho _{18,11}=-0.297\) ).

Significant correlation of initiator’s and recipient’s individual random effects across behaviors estimated by Model_i and Model_ih.

For the same recipients, correlations among three behaviors were statistically significant: comforting , dominating , and requesting access (Fig. 2 ). The probability of a child receiving requests for access was negatively correlated with that of being comforted ( \(\rho _{1,7}=-0.351\) ) or dominated ( \(\rho _{2,7}=-0.370\) ). On the other hand, children who were frequently targeted for dominance were also more likely to receive comfort from others ( \(\rho _{1,2}=0.393\) ).

Demographic effects

Model_iF included several predictor variables in addition to Model_i. Compared to the variance estimates obtained in Model_i, the variance estimates of the initiators’ random effects decreased for almost all the behaviors (Table  2 ), indicating that individual-level variance of initiators discovered in Model_i can be substantially explained by sex, age, and kinship for all behaviors except helping . Similarly, the predictor variables accounted for substantial individual-level variance among recipients in all behaviors other than leading and taking . We predicted the probabilities of each of the 19 behaviors between an “average” recipient and an “average” initiator based on demographic variables (age, sex, and household status/proxy for kinship), using the estimated coefficients obtained from Model_iF.

We investigated the effects of age on the predicted probabilities of 19 behaviors, as well as how such effects were moderated by sex, behavioral role (initiator/recipient), and kinship (household status). The predicted probabilities formed three patterns: they either consistently increased with age, consistently decreased with age, or initially increased up to a certain peaking age and then declined. We report the effects of age in the following order of modification factors: household-status, sex, and behavioral role.

First, the effect of age on behavioral probabilities did not show great variations by household status, namely, whether the initiator and recipient were from the same household or not. However, the effect of age was modified by sex for most behaviors, except for comforting , dominating , and leading , which displayed higher probabilities as the initiator’s age increased, regardless of sex. We examined the trends with increasing age and different sexes for the same initiator or recipient, as the age-sex interaction effect did not differ across behavioral roles. For simplicity, we only listed the trends for intra-household interactions, which occurred more frequently than inter-household interactions (Supplementary Fig.   S1 b), in Table 3 . For inter-household interactions, see Supplementary Table   S8 .

Age-sex interaction manifests in multiple ways. First, for certain behaviors, the initiator’s age had distinct and even contrasting effects between girls and boys. When initiators were girls, probabilities of dirty looks , verbal aggression , supporting opinions , and playful teasing were predicted to peak before age 5 and then sharply decreased. When initiators were boys, the probability declined consistently for dirty looks and increased for the other three behaviors. For recipients of these behaviors, however, the effect of their age did not differ by sex. Across both sexes, older children were more likely to receive dirty looks, face verbal aggression, and have their opinions supported. Children aged 7–10 years were most likely to receive playful teasing. Second, certain behaviors showed age-sex interaction only for recipients, not initiators. For girls, the probability of receiving physical aggression peaked at ages 2–3, while that of receiving aggressive teasing peaked at ages 4–5. Conversely, the probability for boys to receive physical aggression and aggressive teasing decreased with age. Girls aged 6–8 years were most likely to experience ownership assertion from others, whereas boys became increasingly likely targets of ownership assertion with age. 5–6-year-old girls 8–10-year-old boys were most likely to become targets of taking behaviors. It’s noteworthy that regardless of sex, toddlers (younger than two) were most likely to initiate physical aggression and assert ownership, and children aged 2–5 were most likely to initiate aggressive teasing and taking resources. Finally, for sharing and tattling , age-sex interaction existed among both initiators and recipients: girls were most likely to share at 5–6 and tattle at 6–7 years old, but were most likely to receive sharing and tattling when they were 1–3 years old; for boys, the probability of initiating both behaviors peaked before the age of 2 and that of receiving these behaviors increased with age.

Another factor is whether the effect of age on a given behavior differed across initiators and recipients. First, helping behavior is the only exception, the probability of which peaked among 3–8-year-old children for both initiators and recipients. Second, the initiator-recipient age difference affected some behaviors. For example, the probabilities of comforting and dominating were predicted to increase with the initiator’s age but to decrease with the recipient’s age. Besides, older children were less likely to initiate requests for sharing resources, comforting, helping, and access to play, but more likely to receive these requests. Finally, for certain behaviors, initiator’s age and recipient’s age had different but not opposite effects, such as leading and scolding .

We predicted the probabilities of 19 behaviors with different sexes of initiators and recipients while keeping their ages at the average of the sample (Fig. 3 , Supplementary Figs.   S10 , S11 ). When both the initiator and recipient resided in the same household, a proxy of biological sibling relation, sex did not have a statistically significant impact on the predicted probabilities of any behaviors. However, when the initiator and recipient resided in different households, the sex of the initiator had a significant impact on the predicted probabilities of aggressive behaviors, while the sex of the recipient did not affect any behaviors. Compared to boys, girls were more likely to scold, give dirty looks to, or tattle on others in different households. Boys were more likely to initiate aggressive teasing and verbal aggression toward children from other households. Boys were also more likely to initiate physical aggression, but only towards other boys from other households.

Predicted probabilities of response behaviors as a function of initiator’s sex. Plots are presented for behaviors that are significantly affected by the initiator’s sex. The plots for all behaviors are in Supplementary Fig.   S10 . All continuous covariates are held constant at the sample mean. The confidence intervals are the 95% percentile intervals, as calculated from the posterior samples of Model_iF. The coefficients of fixed effects used in the prediction are listed in Supplementary Table   S6 .

Initiator-recipient co-residence

We investigated if initiator-recipient co-residence had an impact on the predicted probability of 19 behaviors (Fig. 4 ), considering four different combinations of sexes, with the ages held constant at the sample mean. Helping behavior was more likely to happen among children from the same household. However, there were no significant differences in the other behaviors.

Predicted probabilities of response behaviors as a function of the household status of the initiator and the recipient. Plots are presented for behaviors that are strongly affected by the household status. The plots for all behaviors are in Supplementary Fig.   S12 . All continuous covariates are held constant at the sample mean. The confidence intervals are the 95% percentile intervals, as calculated from the posterior samples of Model_iF. The coefficients of fixed effects used in the prediction are listed in Supplementary Table   S6 .

Our research is a rare study that uses modeling methods to examine naturalistic observations of children’s socio-moral behavior in a historical, non-Western, and rural context. The original dataset, to which we were granted unique access, was the fruit of “the first attempt ever to record in a systematic manner the behavior of Chinese children” 56 , and occupies a significant niche in the intellectual history of anthropology and cross-cultural research on child development. We developed a new coding system from these historical texts and used multilevel multinomial logistic regressions to analyze the effects of demographic factors on a variety of social behaviors. Previous research typically focused on one particular class of behavior (e.g., prosocial or aggression) and analyzed demographic variables such as age and/or sex. Based on the unique data, our study examined multiple prosocial, aggressive, and other locally salient behaviors in an organic community, and added two ecologically valid factors, behavioral role (initiator and recipient), as well as kinship (sibling relation). Our study not only revealed positive correlations within various prosocial behaviors and aggressive behaviors respectively, but also identified linkages between cooperative and non-cooperative behaviors, especially in leadership dynamics. Moreover, we found age, sex, and behavioral roles (initiator/recipient) as key predictors of these behaviors and their effects interacted with each other. Below we discuss the most important findings that generate novel insights on studying culture and moral development in naturalistic settings. We first focus on leadership strategies, as the findings exemplify the value of our approach. We then offer contextualized interpretations of the age-related trends and gendered patterns of other prosocial and aggressive behaviors. Taken together, through re-engaging intellectual history via a new approach, our study challenges classic views on Chinese childhood, illuminate how children’s actual behavior is shaped by but also diverges from cultural ideology, and open up new inquiries for future work.

Our study discovered intricate connections between cooperative and non-cooperative behaviors in the context of leadership dynamics, an important space for moral development and value transmission . Notably, leading emerges as the most frequent proactive behavior, highlighting children’s substantial investments in mobilizing and collaborating with their peers to achieve shared objectives. The prevalence of leading alongside correlations between distinct behavioral strategies within the same children compels us to closely examine leadership within naturalistic settings. For a given child, leading , defined as attempts to persuade another child through non-coercive means, was positively correlated with dominating , defined as attempts to coercively impose one’s will on others, and scolding , defined as criticizing another child for specific misbehavior. One plausible interpretation is that leading , together with dominating and scolding , may function as complementary strategies to establish and maintain leadership positions during peer interactions. Leaders might intend to use these strategies to directly benefit themselves, but in certain cases, such as in scolding , they may also have conferred benefits upon the other party and facilitated the transmission of moral values. Leading and scolding in peer interactions, although not a prominent topic in previous studies of moral development, were identified from a qualitative, bottom-up perspective as salient ethnographic topics, their significance and interconnectedness further illuminated via statistical modeling. This example shows the novelty and merit of our analytical approach.

Moreover, a closer look at these leadership behaviors revealed possible cultural influences as well as deviation from cultural expectations in real life. For the same initiators, negative correlation between leading and physical aggression suggests that children in leadership roles refrained from physically bullying others. We also found negative correlations between scolding and physical aggression , verbal aggression and requesting comfort . Children who scolded others may strive to become a moral exemplar or authority for their peers and siblings, therefore, were less inclined to engage in physical aggression or display emotional vulnerability due to concerns of shame, a key Chinese moral socialization value 57 . From the perspective of recipients, children targeted by domineering leaders were also the ones more likely to get emotional comfort. Meanwhile the same children were unlikely to receive requests for access to group activities, which means that they did not have leadership over play groups. This suggests that subordinate children did receive care from other children, the dark and the bright side of moral life intertwined. Regarding demographic patterns, all three behaviors, leading , dominating , and scolding , increased with the initiator’s age, which is consistent with local cultural norms that ascribe moral knowledge and authority to older children. We found no significant sex differences in either leading or dominating , suggesting that girls were as likely to assume leadership roles as boys. Furthermore, girls were more likely to initiate scolding than boys, a strategy to establish their authority via moral preaching. Mainstream anthropology scholarship, including the Wolfs’ own works 45 , 50 , emphasized gender socialization ideology in Chinese culture, i.e., girls submitting to boys, rather than actual experience in childhood, therefore assumed a passive role of girls. Aligned with that gender ideology, the scant observational research of contemporary urban Chinese children found that boys displayed more dominance than girls 48 . This contrast might have resulted from differences in study design, that our data are based on a much more extensive fieldwork with a larger sample size, or that our study employed a more rigorous and sophisticated statistical approach. It might also relate to differences in historical contexts and lifestyles: for example, rural girls in our study had more free-play and and their conduct was less restrained by adult monitoring compared to their urban counterparts; girls in multi-child families had more opportunities and experience in domineering others than urban, singleton children 12 . Regardless, our study offers a precious glimpse into young girls’ agency in leadership dynamics, a topic that has long been obscured in the study of traditional Han Chinese families, and opens up new inquiries about historical continuity and change.

The impacts of age on a variety of behaviors deserve contextualized interpretation, as both the initiator’s and the recipient’s age were important predictors of children’s behaviors. Traditional Chinese societies generally placed more moral demands on older children 45 . Parents from this community expected older children to act as role models for the younger ones and refrain from bullying them 12 , but we found mixed evidence in the actual behavior. Younger children requested prosocial favors from older children, while older children initiated both coercive (dominating) and prosocial behaviors. Older children took the lead in group activities, and offered comfort and assistance, but also asserted their authority by maintaining orders and bullying the children who disobeyed. Younger children tended to look up to older ones for guidance and made requests. They also challenged older ones’ authority through subtle expressions like resentful looks or playful teasing. We found an interesting contrast between physical aggression , which peaked among children aged 24–42 months and then declined, similar to what previous studies suggested 36 , and dominating , which became more likely as the initiator’s age increased, as shown earlier. As children grew older, they learned to restrict physically attacking others but resorted to other coercive means to impose their own preferences onto others. From a recipient perspective, with increasing age, they solidified their authority and were less likely to be dominated, led, or caught up in a fight. They also sent fewer emotional requests and received less comfort, perhaps the result of learning to control their emotions in accordance with Chinese cultural precepts. Additionally, resource-exchanges during leisure time, such as snacks, rubber bands, tiles, and cards, became petty and trivial for older children, resulting in a decreased likelihood of sharing, taking, and asserting ownership. The finding that older children, although being care-takers and role models, did not actually yield to younger ones in conflict situations, contradicts the important Chinese cultural norm of “older children yielding to younger ones,” a cultural norm that Margery Wolf observed from this community 45 . Hence our study has broader implications for comparing children’s actual behavior with cultural and moral ideologies.

Sex differences in behaviors, although only identified in non-sibling dyads, also shed new light on moral development in Chinese culture. First, we found no sex-differences in prosocial behaviors, which contradicts the female-prosociality bias found in contemporary survey research on Taiwan 58 but lends support to recent cross-cultural experimental findings 39 . More importantly, patterns in aggressive behaviors offer valuable insights into the relationship between culture, gender socialization and behavior. In non-sibling interactions, boys were more likely to initiate physical aggression compared to girls, but only when the recipients were also boys. Besides physical aggression, we found consistent patterns in other forms of aggression regardless of the recipient’s sex, among non-sibling dyads: boys initiated direct forms of aggression, such as verbally insulting and aggressively teasing others, and such tendencies increased with age; girls displayed subtler forms, including giving dirty looks and tattling, while their verbal aggression and aggressive teasing declined with age. As age increased, boys were less likely to tattle on others, but they were also more likely than girls to become the targets of tattling. Our study therefore reveals a more complex picture of sex differences in aggression than prior observational or self-report studies on Chinese children 48 , 58 . Girls’ tactics are especially interesting: Tattling can help mitigate conflicts while asserting oneself, through seeking help from external authority. Dirty looks provide a socially acceptable means of expressing discontent without escalating conflicts or drawing potential punishment. These strategies, together with the often ignored girl-to-girl physical aggression, defy stereotypes of docile young girls in traditional Han Chinese families in literary and ethnographic representation, including Margery Wolf’s own works 45 , 50 . Also, the Wolf’s writings, or previous literature more broadly, have rarely discussed how sex differences in aggression take shape in the specific historical and political contexts. Fine-grained ethnographic analysis of the Wolf Archive, however, shows that fighting and violence among boys might relate to the gangster tradition in the area and the policing culture in Taiwan’s Martial-Law era 12 . These statistical and ethnographic findings, from a typical patriarchal community known for its son-preference and daughter-discrimination, offer insights into both cultural patterning and individual agency and demonstrate historicized and contextualized understanding of gender and aggression .

Kinship, especially sibling relations, is an important factor in children’s social life in this rural community. Although children interacted with their siblings more frequently than with those from other households, this factor alone did not significantly predict their social behaviors. This might be related to children’s residential patterns: instead of segregated apartment buildings housing nuclear families in the cities, this village consists of clustered farmhouses often connected via extensive kinship ties, with ample space for communal life. Even when children were interacting with their siblings, other children of different ages were often present and mingling together in the same group activities, therefore sibling relation became a less important predictor than age. The only exception is helping , which more likely occurred between siblings than between non-siblings, probably because siblings looked out for each other when needed 59 . Also, household status did modify the effects of some other factors. Notably, sex differences in children’s behaviors that were significant in non-sibling dyads, e.g. aggression. Patterns of aggression that are aligned with well-established findings in previous works, i.e., male-bias in physical aggression, disappeared when the initiator and the recipient were from the same household. The ethnographic analysis did show that sibling fighting at home was a salient theme in this community 12 , which can partly explain why sibling relation overrides sex differences in predicting aggressive behaviors. Future work should pay more attention to sibling dynamics when examining gendered moral development.

The present study also has several limitations. First, the effects of kinship on children’s behavior remains an intriguing question that merits further analysis beyond the scope of this study. The current study used household number as a generic index for kinship, especially sibling relationships, without specifying birth order, family size, or the nature of family ties, e.g., biological or adoptive siblings. Because most families in this sample had more than two children, and some still adopted children–a local custom that declined during the Japanese-rule but was not eliminated by the 1960s 56 -future work can disentangle how these different factors impacted children’s behavior. Also, although our data came from 70 distinct households, many households were connected via kinship. A majority of the villagers descended from the same Chinese immigrants in southern Fujian who had arrived in the region during the eighteenth and nineteenth centuries. A large joint family lived in this village, as well as some nuclear or stem family households, as part of the Chinese family cycle 44 : For example, when two brothers get married, one large household may split into two smaller ones for their respective families. As a result, we may apply the ideas and techniques of Gaussian process regression to the multinomial model. Instead of considering discrete boundaries between households, we may employ a matrix of distances between pairs of observed behaviors. However, anthropologists may find such statistical models unnecessarily complex, and it remains an open question if this dataset from only one village is suitable for such increased statistical complexity in analysis.

Given its historical, naturalistic datasets, the Wolf Archive also affords researchers the opportunity for further reflections and analyses. First, our model only considered a limited number of individual-level demographic predictors, but other household-level variables may matter too. Household SES status might be a predictor of children’s behaviors if we combine the current behavioral data with SES data inferred from other fieldnotes left behind by Arthur Wolf. Moreover, as natural observations also include reactive behaviors, we may extend our analysis to reactive behaviors. Since some proactive and reactive behaviors are matched (for example, dominating vs. submitting ), we can explore the mechanisms and demographic patterns of behavioral contingencies, reciprocity, and social relations, via social network analyses. Also, besides dyadic interactions, quantifying and modeling multi-agent dynamics in the raw data is a promising next step. Moreover, given the importance of age in predicting children’s behaviors in this dataset, further analyses should closely examine developmental patterns of particular behaviors broken down to different age groups, from infancy to early adolescence, or model the longitudinal trajectories of behaviors across the two-year research time span, and interpret those behavioral patterns in the local context. Additionally, given the complexity of naturalistic observations, we can eventually integrate quantitative and ethnographic, qualitative approaches, to better understand human behavior in its socio-cultural contexts 12 . Finally, this dataset only captures children’s social life in one particular time-space, and future work should gather more diverse datasets of naturalistic observations from old and new ethnographic records to ensure enriched cross-cultural comparison and generalization.

Original data

The study is a secondary analysis of historical texts, field-notes collected by the late anthropologist Arthur Wolf in a Hokkien-speaking village in rural Taiwan, 1958–1960, as part of his dissertation research. His research was approved by Cornell University and conducted according to the relevant guidelines and regulations at that time in the U.S. Children’s social behavior was observed inside the village, at home, and at the elementary school outside the village. The first-hand witness of children’s social behavior was Arthur Wolf’s research assistant, an Taiwanese teenager girl who spoke the local language and was trusted by local children and their families because of her personality. She observed children’s naturalistic behaviors in meticulous detail documented them in systematic episodes, written in Chinese. On the same day, the research assistant then reported her observations to Margery Wolf, who was the anthropologist Arthur Wolf’s wife and performed the role of a “scribe” at that time 60 . Margery Wolf then translated these observations into English and typed them up. These typewritten notes, preserved in Arthur Wolf’s private library in Northern California, constitute the original data that our study is based on. All the observations were indexed by their event information (data, time, location) and by the ID of the participants, both initiator and recipient. Demographic information was also collected in a systematic manner, such as the age at observation, sex, and household number of all the people involved.

The data collection approach, the excellent local research assistant, and the prolonged fieldwork made the observational texts in the Wolf archive even richer and rarer, compared to the SCS materials. According to the SCS field guide 18 , Child Observation should focus on a predefined set of social situations. Wolf’s RA, in contrast, reported everything she saw the children doing and saying and how other people were involved or reacted, all in spontaneous episodes rather than waiting for a particular situation to occur. Also, while the SCS field guide designed CO as “short excerpts of behavior rather than extended interaction sequences,” Wolf’s RAs did much better than that, by violating the instructions and recording extended behavioral sequences faithfully 56 . Therefore, although Wolf intended to follow the SCS design and target children ages 3–12 (calculated at the beginning of study), it turned out that these observational records contained abundant information about a much larger sample of children which also includes those who were younger than 3 or older than 12.

Digitization and secondary analysis

Arthur Wolf’s original fieldnotes were housed in his residence. With unique permission to access and use them, we digitized these materials into machine-readable files. We obtained ethical approval from the Internal Review Board of the University of Washington for analyzing these fieldnotes. We did secondary analysis on de-identified data, including naturalistic observation texts and demographic and household information. In both types of texts all the participants were labeled by numbers. All analyses were carried out in accordance with relevant guidelines and regulations.

We assigned a unique ID to each of the 1677 observational episodes, and manually coded all the episodes according to a standardized behavioral coding protocol we designed. We designed this new behavior-coding protocol that includes about fifty social interaction themes. In this study, we focused on child-to child dyadic interactions, excluding child-adult interactions such as command-obey, as well as all behaviors that were not directed from an initiator to a recipient, such as agricultural work and housework. We also limited the target population to children younger than 13 years old at the end of the two-year fieldwork. Though there were 1677 observational episodes, we amalgamated responses over all observation episodes. The final dataset analyzed in this paper contains 19 categories of proactive, dyadic directional behaviors (Supplementary Table   S1 ), adding up to 6507 entries of behaviors over two years, and 18 categories of reactive 1-to-1 directional behaviors, adding up to 2344 entries. Since the proactive (e.g., dominating) and reactive behaviors (e.g., submitting) were coded based on different schemes, for clarity of analysis and convenience of interpretation in this study we focus on just the proactive behaviors. Note that sibling care , a proactive behavior, was excluded from the study because it can only be exhibited by children who have younger siblings, whereas other behaviors do not have this constraint. The effect of attending to siblings or not will conflate risk factors for those who could care but did not versus those who did not care because they had no siblings.

A sample episode

Here is a sample observational episode in this dataset: Observation ID: 28, Date: 08/03/1959, Location: Two logs near the big tree. Observer: MC. Observation content: 493, 157, 145, and 144 were sitting on the two logs. The others were nearby. MC asked them if they had eaten. 145: Let’s not answer her. No one answers her and they laugh. MC did not pay any attention. 144 answered her. He said: I ate two halves. (He meant to say that he had eaten, but mispronounced the words. Actually, he hadn’t eaten.) 157: Yeh, you’ve already eaten two halves of fruit. How big your stomach is? Oh! You’ve eaten two halves. You must be very full, etc. 157 kept yelling these comments over and over and everyone was laughing at 144’s mistake. 144 finally got angry and hit 157 lightly, saying: Quit it! Quit it! The children continued to laugh and 157 kept saying this. 144 started to tickle 157. They laugh. 131 came to tickle 157 also. 157 stood up and tried to catch 131, but he missed him. 157–144: Stand up and let’s fight. We’ll see who wins. 144 stands up smiling and they wrestle. The other children are still laughing because 157 continues to tease 144 about his big stomach, etc. as they wrestle. 144 is beginning to get angry and 157 sees this. 157 runs away from 144 and yells: Oh, your stomach is so big from eating two halves. 144–157: I’m going to knock you down on the ground. 157 runs away, sits down again. 157 comes near and 144 stands up. Then 157 runs away. This is repeated several times with 144 threatening to hit 157 and 157 teasing 144 about the two halves he has eaten. 145–157: I’m going to hit you, too. Quit it! Finally, 144 ignores him. 157 came and sat in front of 144 and started to sing a song. He changed the words to call 144 a “Big Forehead”. (144’s forehead protrudes a little) and soon all the children were singing this. 144–157: I’m going to hit you. He picks up his slingshot and says: I’m going to hit you with my slingshot. 157 continues to sing. 144 hit him with the slingshot, but 157 kept singing. 144 turned to 493: Why are you laughing at me? 493: Copulate with your mother. I’ll1 not laughing at you. I’m just singing a song. 144 and 493 swear at each other and punch each other with their shoulders (they are sitting next to each other). 157 is still singing. 144 is very angry now. 144: I’m going to hit you. 157 runs home with 144 after him. 157 goes into House 14. 144 sits in front and says he’s going to wait until 157 comes out. 157 says something to 144. 144 grabs a stick and runs into the house and hits him. 140 is in the kitchen and says: Are you two still fighting in there? P comes out and 157 comes out too. 157 begin to tease him again. 144 angrily chases him again. 157 hides behind a tree and sings the song. 144 still has the stick in his hands. 145 turns to 131: You No Teeth (131 is missing some teeth). 131–145: You Eleven Fingers (145 has 11 fingers). 131 hit 145 on the face. 145 ran to 144. 144–131: What are you laughing at? 144 threw a rock at 131 but misses. 131 keeps singing the song.144 starts to chase him but 131 runs home. Then 144 turned and chased 157 again. 157 continues to sing. 153 came out. 144–153: 157 is scolding me. 157: I’m not. I’m not. I’m just singing a song. He continues to sing. 153–144: What did he scold you about? 144: He calls me Big Forehead. 153–157: Why are you scolding anyone? Go find the ducks. 157 went to find the ducks and 144 went home. 149 called from the house: Quit fighting, 157, and go the ducks. They aren’t in the house now. 157 ignored her.

Statistical approach

To analyze the behavior data we fitted multilevel multinomial logistic regressions following Koster and McElreath 54 . This approach accounts for the multinomial character of the response variable while also accounting for children’s repeated observations across observation episodes. We denote the behavioral themes as \(1,2,\ldots ,19\) , with the probability of observing each behavior between initiator i and recipient j being \(\pi _{1,i,j},\ldots ,\pi _{19,i,j}\) .

Children can exhibit a set of behaviors due to various unobserved factors, resulting in the clustering, i.e. dependence, of the behavioral variables by individual. Behaviors may also cluster at the household level in the sense that the members of the same household tend to exhibit similar behaviors. Thus, we used multilevel modeling to account for this potential higher-level clustering. Our statistical model allowed the probabilities of exhibiting each behavior to vary across initiators, recipients, and households, even with the same age and sex, by adding random effects. Our model can be summarized by 18 equations that contrast the odds of exhibiting all the different behaviors instead of a reference behavior, behavioral theme 19. Multilevel multinomial logistic behavioral models that apply generalized linear mixed model principles are a good fit for the structure of observational data obtained using scan sampling techniques. Time-varying covariates could be included in these models (e.g., age), and the dependence between measurements made on the same child and the imbalance in the population sample was addressed by the addition of random factors. More notably, by displaying correlated random effects across the response categories, the models provided insights into the behavioral patterns. It is important to highlight that when there are few occurrences of a certain behavior, the posterior distribution can simply reflect the prior of the model for that rare behavior. In our study, though behavioral themes (1) Comforting , (5) Dirty Looks , and (8) Requesting for Comfort had relatively small numbers of occurrences, we did not combine these behaviors from the original coding scheme because we believed that these behaviors are salient in understanding children’s social life and also failed to come up with a reasonable scheme to combine them. Despite the relatively small numbers of these three behaviors, we expected that there would not be major problems fitting the model. In fact, the prior and posterior distribution of the parameters differed for these behaviors (Supplementary Figs.   S10 ,   S11 ).

We fitted four models named by the random and fixed effects included in each of them. Model_i and Model_iF include individual random effects from the initiator and recipient, while Model_ih and Model_ihF include random effects from both the individual and household of the initiator and recipient. Model_i and Model_ih did not include any fixed effect from predictor variables, whereas Model_iF and Model_ihF included predictor variables.

Model_i—individual-level random effects only.

The probabilities of all the distinct behaviors sum to one, so we have

For each observed behavior, the log-ratio comparing the odds of initiator i and recipient j exhibiting each of the 18 pivot behaviors instead of the reference behavior is assumed to be given by

where each \(\alpha _k\) is an intercept that contrasts the behavior k against the reference category and where \(\nu ^I_{k,i}\) and \(\nu ^R_{k,j}\) are person-specific effects for initiation and reception of behavior theme k in initiator i and recipient j , respectively. Across all the behavioral themes we assume the priors

These state that person-specific intercepts, \(\nu ^I_{1,i},\ldots ,\nu ^I_{18,i}\) and \(\nu ^R_{1,j},\ldots ,\nu ^R_{18,j}\) , both follow a multivariate normal distribution with mean zero and their own positive-definite \(18\times 18\) variance-covariance matrix, across behaviors other than the reference category. The off-diagonal elements of these matrices represent the individual-level pairwise covariance between random effects among behaviors 1–18, while the diagonal elements signify the individual-level variance of random effects within each behavior.

The intercept \(\alpha _k\) for a behavior k represents the log-odds of exhibiting that behavior relative to the reference behavior, assuming that all random effect terms are zero. In other words, \(\alpha _k\) is the mean log-odds over all subjects for exhibiting behavior k compared to the reference behavior. Random effects were introduced to account for varying occurrence probability of behavior k rather than the reference by different initiators and recipients. A positive individual-level random effect for the initiator i , \(\nu ^I_{k,i}>0\) , implies that initiator i is more likely to exhibit behavior k instead of reference behavior than the population average, and vice versa. Instead of the magnitude of random effect from each individual, we are interested in the variance of the individual random effects in each behavioral category, providing insights into the extent to which unobserved individual-level factors contribute to the observed variation in the occurrence of each behavior, compared to other sources of variation. A large variance of the individual-level random effects for the initiator for behavior k implies that the probability of exhibiting behavior k instead of reference behavior varies greatly among the initiators. We can also obtain the pairwise correlations across all behaviors (except for the reference) from the corresponding pairwise covariances (e.g., behavior k vs behavior l for initiators, \(\rho ^I_{k,l}=\sigma _{\nu ^I_{k,l}}/(\sigma _{\nu ^I_{k}}\sigma _{\nu ^I_{l}})\) ), which provided insights into the co-occurrence of different behaviors by initiators and recipients. When the initiator’s individual-level random effect has a positive correlation between two behaviors k and l , \(\rho ^I_{k,l}>0\) , an individual who is more (or less) likely to initiate behavior k is also more (or less) likely to initiate behavior l . Conversely, a negative correlation implies that initiators who are more (or less) likely to exhibit behavior k are less (or more) likely to exhibit behavior l . The interpretation of individual-level random effects and pairwise correlations for the recipient is similar. We chose behavioral theme 19, ownership assertion , as the reference behavior because correlations of random effects between ownership assertion and other behaviors were of less interest than correlations of random effects between the remaining peer-interactive behaviors.

Because of the considerable complexity of this model, we fit it using Bayesian methods, enabling the use of highly-flexible Markov Chain Monte Carlo (MCMC) algorithms. Specifically, we assigned standard normal priors to the intercepts, \(\alpha _k\) . In theory, the covariance structure of the individual-level random effects, \(\Sigma _{\nu \_I}\) or \(\Sigma _{\nu \_R}\) , can be decomposed into a correlation matrix and a vector of element-specific variances, or “scale” terms 61 . To improve the computational efficiency and arithmetic stability of MCMC simulations, we employed a non-centered parameterization of the random effects based on a Cholesky factorization of the correlation matrix. This decomposition represents the Hermitian positive-definite correlation matrix as a product of a lower triangular matrix and its conjugate transpose 62 , 63 . We set a Cholesky factorized prior with shape equal to 2 for the parameterized correlation matrix.

Model_iF—fixed effects from individual characteristics

To investigate if some demographic characteristics of children were highly influential factors in children’s behaviors, we included some predictor variables as fixed effects in our model. By interpreting the coefficient of the fixed effects in our models, we are able to improve our understanding of the various factors influencing children’s development and behaviors.

In addition to the individual-level random effects presented in Model_i, Model_iF includes individual-level variables for age and sex, as well as their interaction. To investigate how children from the same or different families influence the interaction between them, Model_iF also incorporates an indicator of whether the initiator and recipient were from the same household. We standardized the continuous predictors 64 , i.e., ages of initiators and recipients, to make sampling (from the posterior distribution as described later) more efficient. Specifically, the variables were shifted and rescaled to have a mean of zero and a standard deviation of one. Model_iF had the form

where \(x_{Im}\) and \(x_{Rm}\) are the fixed effect m from the initiators and the recipients, respectively, and \(x_{H}\) is a binary indicator variable that reflects whether the initiator and recipient in each behavior observation belonged to the same household or not. The individual-level effects \(\nu ^I_{k,i}\) and \(\nu ^R_{k,j}\) are assumed to follow the same multivariate normal distribution as in Model_i, and \(\pi _{k,i,j}\) always sum up to 1 for all i and j . Model_iF summed over all fixed effects included in the model.

In this mixed effects model, the intercept \(\alpha _k\) represents the log-odds of exhibiting behavior k relative to the reference behavior, when all predictors and random effects are zero. Specifically, it represents the average log-odds of behavior k that occurs between female initiators and recipients, who are of the same age as the sample average, and come from different households. The coefficient of a predictor, denoted as \(\beta _{k,Im}\) , \(\beta _{k,Rm}\) , or \(\beta _{k,H}\) , measures the effect of a one-unit increase in that predictor, \(x_{Im}\) , \(x_{Rm}\) , or \(x_{H}\) , on the average log-odds of exhibiting behavior k instead of the reference behavior. In other words, it tells us how much the average log-odds of behavior k change when we increase the corresponding predictor by one unit, holding all other predictors constant.

The prior distributions on the parameters in this model are the same as those in Model_i, with additional standard normal priors assigned to the independent fixed effects parameters \(\beta _{k,Im}\) , \(\beta _{k,Rm}\) , and \(\beta _{k,H}\) .

Model_ih—both individual and household-level random effects

In addition to the individual-level random effects presented in Model_i, Model_ih included random effects from both the initiator’s household [ i ] and the recipient’s household [ j ]. Model_ih had the form

where \(h^I_{k,[i]}\) and \(h^R_{k,[j]}\) are household-level random effects for initiator i and recipient j , respectively. The individual-level effects \(\nu ^I_{k,i}\) and \(\nu ^R_{k,j}\) follow the same multivariate normal distribution as in Model_i, and all \(\pi _k\) always sum up to 1. This extended model has nested random effects since each individual is uniquely associated with only one household. Since the personal random effects should be nested within households, the random terms, while not identical, are more likely to be similar within a household versus between households.

Including the household-level random effects changes the interpretation of the individual-level random effects. The individual-level random effects are interpreted as the deviation from the household-level average rather than the population average. A positive individual-level random effect for the initiator i , \(\nu ^I_{k,i}>0\) , now implies that the probability of exhibiting behavior k instead of the reference behavior is higher for initiator i than the average likelihood within the corresponding household, and vice versa. This means that the influence of unobserved individual-level factors on the likelihood of exhibiting a particular behavior is being measured relative to the average behavior of the household rather than the population as a whole. Furthermore, when a household-level random effect of the initiator is positive, \(h^I_{k,[i]}>0\) , individuals in that household [ i ] has an above-average chance of initiating behaviors k instead of the reference behavior, and vice versa. The interpretations are similar for recipients.

The prior distributions for the intercepts and the individual-level random effects in this model are the same as those in Model_i, with additional Cholesky factorized priors with a shape equal to 2 assigned to the parameterized correlation matrices of household-level random effects, \(\Sigma _{h\_I}\) and \(\Sigma _{h\_R}\) .

Model_ihF—fixed effects from individual characteristics

In addition to the individual-level random effects presented in Model_iF, Model_ihF included random effects from both initiator’s and recipient’s household. Model_ihF had the form

The prior distributions for the intercepts and independent fixed effects in this model are the same as those used in Model_iF. Similarly, the prior distributions for the covariance structures of random effects are the same as those employed in Model_ih.

Estimation of parameters

Since this multilevel multinomial logistic model is not implemented in standard software, we fitted a Bayesian version of the model using Markov Chain Monte Carlo (MCMC) estimation rather than the commonly used maximum likelihood method. In particular, the inference is based on the expectations of posterior quantities, such as posterior means and standard deviations of parameters.

We used R’s RStan package 61 to facilitate MCMC. RStan is the R interface to Stan , which is a state-of-the-art platform for statistical modeling and high-performance statistical computation. Users can specify log density functions in Stan’s probabilistic programming language and get full Bayesian statistical inference via Hamiltonian Monte Carlo sampling (HMC), which is a family of MCMC algorithms 65 . Stan is preferred over the older but widely used BUGS software due to its considerably higher efficiency and faster running speed 66 . We employed weakly informative priors for the parameters of fixed effects and variance-covariance matrices of random effects as described beside the statement of models. We performed prior sensitivity analysis and validated the use of the weakly informative priors. We ran each model on three chains, each with 10000 iterations and a warmup of 5000 iterations. We confirmed model convergence by examining the trajectory plot of the chains and the R-hat Gelman and Rubin convergence diagnostic.

Analysis of raw output

We utilized the rethinking package to prepare data, summarize the posterior, and plot model predictions 67 . We compared the predictive performances of the four models based on the Widely Applicable Information Criterion (WAIC) 55 . We estimated the Cholesky matrix using HMC chains and computed the correlations between the random effects across behaviors via recomposition from the lower triangular matrix and its conjugate transpose. This allowed us to determine if individuals who engage in more of the first behavior also tend to engage in more or less of the second behavior (relative to the reference category).

The coefficients of the fixed effects in Model_iF and Model_ihF are interpreted as the effect of a one-unit difference in one predictor on the log-odds of exhibiting behaviors 1 to 18 instead of the reference behavior after adjusting for other predictors. However, the interpretation of the coefficients is rather awkward. It would be much easier and straightforward to interpret the effects of a predictor on each behavior, rather than on a contrast between two behaviors. Besides, Retherford and Choe 68 noted that coefficients (or odds ratios) are not only difficult to interpret but may even be misleading because the sign of a coefficient may not reflect the direction of the effect of the predictor on either of the response probabilities being compared (i.e., \(\pi _k\) and \(\pi _{19}\) ). Thus, we calculated the predicted response probabilities for each of the 19 behaviors from the estimated coefficients of the fixed effects using a random effect value of zero, giving the predicted probabilities of behavior between an “average” recipient and an “average” initiator. We plotted the predicted probabilities with one predictor varying at a time while holding other predictors constant, together with 95% credible intervals incorporating uncertainty in the fixed effect parameters.

Data availability

The raw fieldnotes are part of a private historical archive not available to the public yet. All de-identified and processed data and associated R scripts are available at https://github.com/zhiningsui/children_behavior_multinomial_analysis .

Rousseau, J.-J. Emile (Phoemixx Classics Ebooks, 2021).

Piaget, J. The moral judgment of the child (Routledge, 2013).

Stafford, C. The roads of Chinese childhood: Learning and identification in Angang. (Cambridge University Press, 1995).

Tomasello, M. Becoming human: A theory of ontogeny (Harvard University Press, 2019).

Woo, B. M., Tan, E. & Hamlin, J. K. Human morality is based on an early-emerging moral core. Annu. Rev. Dev. Psychol. 4 , 41–61. https://doi.org/10.1146/annurev-devpsych-121020-023312 (2022).

Article   Google Scholar  

Barrett, H. C. Towards a cognitive science of the human: Cross-cultural approaches and their urgency. Trends Cogn. Sci. 24 , 620–638 (2020).

Article   PubMed   Google Scholar  

Nielsen, M., Haun, D., Kärtner, J. & Legare, C. H. The persistent sampling bias in developmental psychology: A call to action. J. Exp. Child Psychol. 162 , 31–38. https://doi.org/10.1016/j.jecp.2017.04.017 (2017).

Article   PubMed   PubMed Central   Google Scholar  

Harris, P. L. Trusting what you’re told: How children learn from others (Harvard University Press, 2012).

Xu, J. Learning “merit” in a chinese preschool: Bringing the anthropological perspective to understanding moral development. Am. Anthropol. 121 , 655–666. https://doi.org/10.1111/aman.13269 (2019).

Astuti, R. Taking people seriously. HAU: J. Ethnogr. Theory 7 , 105–122. https://doi.org/10.14318/hau7.1.012 (2017).

Hirschfeld, L. A. Why don’t anthropologists like children?. Am. Anthropol. 104 , 611–627 (2002).

Xu, J. "Unruly" children: Historical fieldnotes and learning morality in a Taiwan village (Cambridge University Press, 2024).

Bonetto, E., Guiller, T. & Adam-Troian, J. A lost idea in psychology: Observation as starting point for the scientific investigation of human behavior (2023). https://doi.org/10.31234/osf.io/9yk3n .

Lew-Levy, S. et al. Peer learning and cultural evolution. Child Dev. Perspect. (2023).

Muthukrishna, M., Henrich, J. & Slingerland, E. Psychology as a historical science. Annu. Rev. Psychol. 72 , 717–749. https://doi.org/10.1146/annurev-psych-082820-111436 (2021).

Atari, M. & Henrich, J. Historical psychology. Curr. Dir. Psychol. Sci. 32 , 176–183 (2023).

Whiting, B. B. Six cultures: Studies of child rearing. (Wiley, 1963).

Whiting, J. W. M., Child, I. L., Lambert, W. W. et al. Field guide for a study of socialization. (Wiley, 1966).

LeVine, R. A. The six cultures study: Prologue to a history of a landmark project. J. Cross Cult. Psychol. 41 , 513–521. https://doi.org/10.1177/0022022110362567 (2010).

Amir, D. & McAuliffe, K. Cross-cultural, developmental psychology: Integrating approaches and key insights. Evol. Hum. Behav. 41 , 430–444 (2020).

Curry, O. S., Mullins, D. A. & Whitehouse, H. Is it good to cooperate? testing the theory of morality-as-cooperation in 60 societies. Curr. Anthropol. 60 , 47–69. https://doi.org/10.1086/701478 (2019).

Whiting, B. & Edwards, C. P. A cross-cultural analysis of sex differences in the behavior of children aged three through 11. J. Soc. Psychol. 91 , 171–188. https://doi.org/10.1080/00224545.1973.9923040 (1973).

Fabes, R. A. & Eisenberg, N. Meta-analyses of age and sex differences in children’s and adolescents’ prosocial behavior. In W. Damon (ed.) Handbook of child psychology , vol. 3, 1–29 (John Wiley & Sons Chichester, UK, 1998).

Archer, J. Sex differences in aggression in real-world settings: A meta-analytic review. Rev. Gen. Psychol. 8 , 291–322 (2004).

Wolf, A. P. & Huang, C.-S. Marriage and Adoption in China, 1845–1945 (Stanford University Press, 1980).

Jambon, M., Madigan, S., Plamondon, A. & Jenkins, J. Developmental trajectories of physical aggression and prosocial behavior in early childhood: Family antecedents and psychological correlates. Dev. Psychol. 55 , 1211–1225. https://doi.org/10.1037/dev0000714 (2019).

Turiel, E. Moral development in the early years: When and how. Hum. Dev. 61 , 297–308 (2018).

Kohlberg, L. The development of modes of moral thinking and choice in the years 10 to 16 . Ph.D. thesis, The University of Chicago (1958).

Blake, P. R. et al. The ontogeny of fairness in seven societies. Nature 528 , 258–261 (2015).

Article   ADS   CAS   PubMed   Google Scholar  

House, B. R. et al. Ontogeny of prosocial behavior across diverse societies. Proc. Natl. Acad. Sci. 110 , 14586–14591 (2013).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Grueneisen, S. & Warneken, F. The development of prosocial behavior—from sympathy to strategy. Curr. Opin. Psychol. 43 , 323–328. https://doi.org/10.1016/j.copsyc.2021.08.005 (2022).

House, B. R. & Tomasello, M. Modeling social norms increasingly influences costly sharing in middle childhood. J. Exp. Child Psychol. 171 , 84–98. https://doi.org/10.1016/j.jecp.2017.12.014 (2018).

House, B. R. et al. Universal norm psychology leads to societal diversity in prosocial behaviour and development. Nat. Hum. Behav. 4 , 36–44. https://doi.org/10.1038/s41562-019-0734-z (2019).

Tremblay, R. E. The development of agressive behaviour during childhood: What have we learned in the past century?. Int. J. Behav. Dev. 24 , 129–141 (2000).

Alink, L. R. A. et al. The early childhood aggression curve: Development of physical aggression in 10- to 50-month-old children. Child Dev. 77 , 954–966. https://doi.org/10.1111/j.1467-8624.2006.00912.x (2006).

Tremblay, R. E. Decade of behavior distinguished lecture: Development of physical aggression during infancy. Infant Mental Health J. Off. Publ. World Assoc. Infant Mental Health 25 , 399–407 (2004).

Google Scholar  

Hawley, P. H. The duality of human nature. Curr. Dir. Psychol. Sci. 23 , 433–438. https://doi.org/10.1177/0963721414548417 (2014).

Rose, A. J. & Rudolph, K. D. A review of sex differences in peer relationship processes: Potential trade-offs for the emotional and behavioral development of girls and boys. Psychol. Bull. 132 , 98–131. https://doi.org/10.1037/0033-2909.132.1.98 (2006).

House, B., Silk, J. B. & McAuliffe, K. No strong evidence for universal gender differences in the development of cooperative behaviour across societies. Philos. Trans. R. Soc. B Biol. Sci. 378 . https://doi.org/10.1098/rstb.2021.0439 (2022).

Card, N. A., Stucky, B. D., Sawalani, G. M. & Little, T. D. Direct and indirect aggression during childhood and adolescence: A meta-analytic review of gender differences, intercorrelations, and relations to maladjustment. Child Dev. 79 , 1185–1229 (2008).

Lansford, J. E. et al. Boys’ and girls’ relational and physical aggression in nine countries. Aggressive Behav. 38 , 298–308. https://doi.org/10.1002/ab.21433 (2012).

Endendijk, J. J. et al. Gender differences in child aggression: Relations with gender-differentiated parenting and parents’ gender-role stereotypes. Child Dev. 88 , 299–316 (2017).

Björkqvist, K. Gender differences in aggression. Curr. Opin. Psychol. 19 , 39–42 (2018).

Wolf, M. The house of Lim: A study of a Chinese farm family (Appleton-Century-Crofts, 1968).

Wolf, M. Child training and the chinese family. Studies in Chinese society (1978).

Wolf, A. P. Sexual attraction and childhood association: A Chinese brief for Edward Westermarck (Stanford University Press, 1995).

Greenhalgh, S. Intergenerational contracts: Familial roots of sexual stratification in taiwan. chap. 2, 39–77 (Stanford University Press, Stanford, CA, 1988).

Jankowiak, W., Joiner, A. & Khatib, C. What observation studies can tell us about single child play patterns, gender, and changes in Chinese society. Cross-Cult. Res. 45 , 155–177 (2011).

Kajanus, A. A developmental perspective on social status: children’s understanding of hierarchy in Nanjing and London. J. Roy. Anthropol. Inst. https://doi.org/10.1111/1467-9655.14065 (2023).

Wolf, M. Women and the family in rural Taiwan (Stanford University Press, 1972).

Duryea, M. Changing lives and life changes on Taipei’s urban border, 1959-1994 . Ph.D. thesis (1999).

Xu, J. The good child: Moral development in a Chinese preschool (Stanford University Press, 2017).

Cheng, Y.-H.A. Ultra-low fertility in East Asia. Vienna Yearbook of Population Research 18 , 83–120 (2020).

Article   ADS   Google Scholar  

Koster, J. & McElreath, R. Multinomial analysis of behavior: Statistical methods. Behav. Ecol. Sociobiol. 71 , 138 (2017).

Lartillot, N. Identifying the best approximating model in Bayesian phylogenetics: Bayes factors, cross-validation or wAIC? (Syst, Biol, 2023).

Wolf, A. P. Chinese children and their mothers . unpublished manuscript.

Fung, H. Becoming a moral child: The socialization of shame among young Chinese children. Ethos 27 , 180–209 (1999).

Hsiao, Y., Cheng, C.-L. & Chiu, Y.-W. Gender network dynamics in prosocial and aggressive behavior of early adolescents. Soc. Netw. 58 , 12–23. https://doi.org/10.1016/j.socnet.2019.01.002 (2019).

Xu, J. Learning morality with siblings: The untold tale of a mid-twentieth century Taiwanese family. J. Chin. Hist. 6 , 337–363 (2022).

Wolf, M. Chinanotes: engendering anthropology. Fieldnotes: The makings of anthropology 343–355 (1990).

Stan Development Team. RStan: the R interface to Stan (2020). R package version 2.21.2.

Carpenter, B. et al. Stan : A probabilistic programming language. J. Stat. Softw. 76 (2017).

Wang, J.-Y. Variance reduction for multivariate monte carlo simulation. J. Deriv. 16 , 7–28 (2008).

Oka, M. Interpreting a standardized and normalized measure of neighborhood socioeconomic status for a better understanding of health differences. Arch. Public Health 79 (2021).

Betancourt, M. & Girolami, M. Hamiltonian monte carlo for hierarchical models. Current trends in Bayesian methodology with applications 79 , 2–4 (2015).

Monnahan, C. C., Thorson, J. T. & Branch, T. A. Faster estimation of bayesian models in ecology using hamiltonian monte carlo. Methods Ecol. Evol. 8 , 339–348 (2017).

Mcelreath, R. Statistical Rethinking: A Bayesian Course with Examples in R and STAN (Chapman and Hall/CRC, Boca Raton, 2020).

Book   Google Scholar  

Retherford, R. D. & Choe, M. K. Statistical models for causal analysis (Wiley-Interscience, Newy York, 2011).

Download references

Acknowledgements

This work was supported by a research grant from the Chiang Ching-kuo Foundation, a National Academy of Education/Spencer Postdoctoral Fellowship and a Wenner-Gren Hunt Postdoctoral Fellowship to JX. This work was primarily conducted at the University of Washington. Our deepest gratitude goes to Dr. Hill Gates, the owner of Arthur Wolf’s archive on Taiwan Village Children, for granting JX unique access to this archive. Thanks to Ken Rice for providing advice to ZS on statistical methods.

Author information

Authors and affiliations.

Department of Biostatistics, University of Washington, 1410 NE Campus Parkway, Seattle, WA, 98195, USA

Zhining Sui

Department of Biostatistics and Computational Biology, University of Rochester, 265 Crittenden Boulevard, Rochester, NY, 14642, USA

Department of Linguistics, University of Washington, 1410 NE Campus Parkway, Seattle, WA, 98195, USA

Qinyan Wang

Amazon.com, Inc., 400 9th Ave N, Seattle, WA, 98109, USA

Department of Anthropology and eScience Institute, University of Washington, 1410 NE Campus Parkway, Seattle, WA, 98195, USA

You can also search for this author in PubMed   Google Scholar

Contributions

JX conceived of the ideas in this paper and ZS designed the models and performed the modeling analysis. JX provided the raw data, designed the coding protocol, coded all the behavioral data, and supervised the project. ZS and QW developed the R code. JX drafted the “ Introduction ” section, ZS drafted the “ Results ” and the “ Methods ” sections and prepared all figures and tables, and all authors contributed to the “ Discussion ” section. JX oversaw the writing, revised all sections of the manuscript and all authors reviewed the final version of the manuscript.

Corresponding author

Correspondence to Jing Xu .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Supplementary information., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Sui, Z., Wang, Q. & Xu, J. Modeling children’s moral development in postwar Taiwan through naturalistic observations preserved in historical texts. Sci Rep 14 , 9140 (2024). https://doi.org/10.1038/s41598-024-59985-6

Download citation

Received : 16 November 2023

Accepted : 17 April 2024

Published : 21 April 2024

DOI : https://doi.org/10.1038/s41598-024-59985-6

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

ORIGINAL RESEARCH article

This article is part of the research topic.

Passive Brain-Computer Interfaces: Toward an “Out of the Lab” Employment

Consistent Spectro-Spatial Features of Human ECoG Successfully Decode Naturalistic Behavioral States Provisionally Accepted

• 1 Kuwait University, Kuwait
• 2 University of California, San Diego, United States

The final, formatted version of the article will be published soon.

Objective Understanding the neural correlates of naturalistic behavior is critical for extending and confirming the results obtained from trial-based experiments and designing generalizable brain-computer interfaces that can operate outside laboratory environments. In this study, we aimed to pinpoint consistent spectro-spatial features of neural activity in humans that can discriminate between naturalistic behavioral states.Approach We analyzed data from five participants using electrocorticography (ECoG) with broad spatial coverage. Spontaneous and naturalistic behaviors such as "Talking" and "Watching TV" were labeled from manually annotated videos. Linear discriminant analysis (LDA) was used to classify the two behavioral states. The parameters learned from the LDA were then used to determine whether the neural signatures driving classification performance are consistent across the participants.Main Results Spectro-spatial feature values were consistently discriminative between the two labeled behavioral states across participants. Mainly, θ, α, and low and high γ in the postcentral gyrus, precentral gyrus, and temporal lobe showed significant classification performance and feature consistency across participants. Subject-specific performance exceeded 70%. Combining neural activity from multiple cortical regions generally does not improve Alasfour et al.decoding performance, suggesting that information regarding the behavioral state is non-additive as a function of the cortical region.Significance To the best of our knowledge, this is the first attempt to identify specific spectro-spatial neural correlates that consistently decode naturalistic and active behavioral states. The aim of this work is to serve as an initial starting point for developing brain-computer interfaces that can be generalized in a realistic setting and to further our understanding of the neural correlates of naturalistic behavior in humans.

Keywords: Brain-Computer Interfaces, neural decoding, neural signal processing, Naturalistic behavior, ECoG

Received: 19 Feb 2024; Accepted: 19 Apr 2024.

Copyright: © 2024 Alasfour and Gilja. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Dr. Abdulwahab Alasfour, Kuwait University, Kuwait City, Kuwait

People also looked at

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• HHS Author Manuscripts

The What, How, and Why of Naturalistic Behavior

In the past few years, advances in machine learning have fueled an explosive growth of descriptive and generative behavior models of animal behavior. These new approaches offer higher levels of detail and granularity than has previously been possible, allowing for fine-grained segmentation of animals’ actions and precise quantitative mappings between an animal’s sensory environment and its behavior. How can these new methods help us understand the governing principles shaping complex and naturalistic behavior? In this review, we will recap ways in which our ability to detect and model behavior have improved in recent years, and consider how these techniques might be used to revisit classical normative theories of behavioral control.

Introduction

Animal behavior is shaped by an interaction of past experience, internal motivational state, and external environmental cues. Behaviors critical for survival, such as feeding and social interaction, are furthermore influenced by the genetically encoded wiring of species-specific neural circuitry. Given this diverse set of influences, the analysis of behavior informs scientific progress in fields spanning genetics, ecology, neuroscience, economics, and robotics. The theories and models of behavior generated by these fields vary in their objective, language, and level of mechanistic detail. They can be broadly divided into categories of descriptive models that identify what an agent did, generative models that tell us how an agent can be expected to behave given a set of conditions, and normative models that identify why a behavior occurred by defining the underlying principles that inform its structure.

Machine learning tools for animal posture and behavior analysis are becoming increasingly accessible to the research community, fueling new approaches to the quantification and study of animal behavior. Many of these newly developed methods are descriptive models that improve the throughput and granularity with which we can quantify behavior, bringing its analysis to the realm of big data. An open question is how these methods can give us insight into the organization and control of behavior itself, in the form of generative and normative models. In this review, we will briefly recount scientific applications of descriptive behavioral analysis, before turning to how detailed pose and behavior data might also be used to build generative models and test classical theories about the principles that shape behavioral organization.

Descriptive models: turning video files into data

In many applications, the use of computational methods for behavior analysis can be likened to automation of manufacturing: the things produced are not fundamentally different from what was around before, but now they can be produced with greater precision and at scale. Several recent reviews 1 – 8 have described the use of modern computer vision or machine learning methods to characterize the pose, kinematics, or actions of behaving agents. At the heart of many approaches is markerless pose estimation, which characterizes animal posture in terms of a set of experimenter-defined 2D or 3D “keypoints” 9 – 12 . Pose estimates or other features extracted from behavior video can be used to quantify postural dynamics of animals 13 – 15 , or paired with supervised (experimenter-defined) behavior detection 16 – 18 or unsupervised (data-defined) behavior discovery 19 – 23 methods to segment continuous posture data into actions ( Figure 1A ).

Unsupervised behavior analysis and its evaluation. A) A typical unsupervised behavior analysis pipeline takes in continuous data (eg pose estimates) and segments it into motifs. Usually data is first pre-processed to remove sources of variance that are not meaningful for behavior (such as the absolute positions of the animals) and extract some representation of animal movement. These processed features are then segmented into motifs using some form of clustering algorithm. B) Different design choices in an unsupervised analysis produce different motifs: there is therefore a pressing need for ways to evaluate the “usefulness” of unsupervised analysis methods. Three commonly used approaches are comparison of motifs to manual annotations, eg in terms of Precision and Recall (left), ability of motifs to predict neural activity (center), and ability to distinguish between animal species, strains, or treatment groups based on the occurrence or patterning of motifs (right).

The practical advantage of these methods is twofold. First, they accelerate data processing when applied at scale. By training algorithms to detect behaviors of interest, researchers can distill hundreds of hours of video into precise and detailed records of animals’ actions that are ready for downstream analysis. In addition to saving human effort, this approach can improve annotation quality: whereas human annotators show substantial within-individual and between-individual variability in labeling behavior start and stop times 17 , automated algorithms are self-consistent. And second, computational behavior analyses can produce high temporal resolution, highly granular descriptions of animals’ actions, surpassing what is feasible for a human annotator 19 – 21 .

The combined impact of automation and increased granularity is beginning to change the way behavioral experiments are designed and analyzed, with a growing number of labs opting to collect long-term, high-throughput recordings of animals in complex environments. Automating the frame-by-frame annotation of animal behavior has enabled high-throughput screening of hundreds of animals across strains, mutant lines, or experimental perturbations 15 , 23 – 30 . Differences between animal cohorts in these assays can most simply be detected by identifying a change in the proportion of time animals express one or more behaviors, or by a change in the transition probability between pairs of behaviors 24 , 25 . Difference in behavior can also be correlated with other within-cohort experimental measures, like gene expression 29 .

Beyond the level of individual behaviors, differences between cohorts may be identified by training a classifier to distinguish animal groups given either postural features or a histogram of expressed behaviors 24 , 25 , 27 . The magnitude of differences between cohorts can also be quantified and visualized to infer relationships between them 25 , 27 . Interestingly, some studies have used dimensionality reduction methods (like Principal Component Analysis or Fisher’s Linear Discriminant Analysis) to identify different behavioral modes across individuals, mutant lines, or drug treatments 25 , 26 , 29 . These analyses suggest a low-dimensional “behavior manifold” might be learned that can capture the variability in behavior expressions across individuals.

Generative model and forecasting: the Laplacian demon in the details

If we knew an animal’s entire sensory environment and history, could we predict what it would do next? How far into the future would our predictions hold? Generative models can help us better understand how animals integrate sensory cues and their own recent history to determine their ongoing behavior. A generative model captures the relationship between two probability distributions, specifically the probability distribution of one signal conditioned on the value of another. In terms of behavior, this might translate to learning a distribution over possible behavioral responses given an animal’s past actions or its sensory environment.

Generative models can be used as a means of categorizing behavior: for example, autoregressive Hidden Markov models capture the statistics of time series evolution, but their inferred hidden states can also be used for unsupervised behavioral segmentation 20 . But generative models have also been used to uncover how animals’ sensory environment, state, and history shape expression of behavior 31 – 34 . For example, fly courtship song type, previously thought to vary at random, can be predicted by postural cues from the courted female 31 , and furthermore this sensory-evoked behavior is better predicted when taking into account an estimate of the singing male’s internal state 33 .

Animals often show preparatory movements or gradual escalation of interactions from which we can predict of future movements—for example, mice dart and tail-rattle before initiating an attack. As their name implies, generative models can also be used to generate “realistic-looking” behavioral data, predicting an animal’s actions given initial conditions 32 , 34 ; in machine learning, this area of research is called imitation learning. Model-synthesized trajectories predict how an animal would respond given an environment, provided that environment is close to the conditions the model was trained on: for example, simulated flies follow walls and perform wing extensions when they encounter other simulated flies 34 . Trained models that can forecast behavior are often hierarchical, allowing them to capture structure in behavior on multiple timescales. One exciting promise of these models is that they provide continuous, behavior-related signals of different degrees of granularity, that can be contrasted with recorded neural dynamics to identify neural correlates of behavior or internal states 35 .

Normative models: what’s my motivation?

Why does behavior have the structure it does? A normative model is one that uses some measure of value or utility to derive predictions of how a system “ought to” behave if driven by that utility. At its highest level, we expect animal behavior to either directly or indirectly further the survival of that animal’s genetic material. While this maxim is quite broad, principles of survival have informed many more specific theories of naturalistic behavior control. We will review examples from single-agent and dyadic behaviors, and discuss how computational behavior analysis might enrich our experimental investigation of these models. There is also a rich literature on principles of collective behavior 36 , 37 , which will not be covered here.

Behavior of an agent in an environment.

Exploration of new environments is a fundamental behavioral drive, be it to establish territory, identify threats, or discover resources. Normative theories predict that exploring animals balance the reward of information gained with the cost of time passage 38 or risk of injury or predation. Foraging behavior, an extension of exploration, also encompasses an assortment of value-based decision-making tasks 39 – 41 . During foraging, an animal makes dietary choices as to what resources to pursue 42 , and also must efficiently use time and energy when collecting resources. In a model environment of food patches, the marginal value theorem determines the precise time to migrate to new food sources, given full knowledge of the underlying average reward rate of the environment 43 , 44 ( Figure 2A ). At the opposite extreme, if nothing is known about the structure of the environment, random Lévy flight motion is optimal 45 . Factors such as food source distribution, prey movement, and competition with conspecifics also impact strategy. Foraging under risk of predation introduces the further wrinkle of balancing the energetic cost of monitoring the environment with the survival cost of being caught by a predator 46 , 47 .

Normative models make predictions about the temporal organization of behavior. A) The Marginal Value Theorem predicts that animals foraging in a patchy environment should migrate to a new patch when their net rate of energy accumulation within a patch (blue line) matches the mean rate of energy accumulation across the environment (parallel red lines). Figure adapted from Charnov, 1976. B) In the resource holding potential model of aggressive escalation, animals integrate evidence to estimate the difference in fitness between fighters, and flee the encounter once their estimate passes a threshold. One possible formulation of this strategy is shown here, adapted from Enquist et al., 1990. The war of attrition model further hypothesizes that animals’ motivation to obtain resources can change their wilingness to flee, indicated here by the shift of the red line.

Behavior in pairs of interacting agents.

Dyadic behavior can include both symmetric interactions between members of a species and asymmetric interactions such as predator-prey behavior. Predatory imminence theory suggests a topology of defensive behavior, in which the type of behavior a prey species expresses is shaped by the immediacy and magnitude of predator threat 48 ( Figure 3A ). This escalating organization of defensive behaviors is built around a key conflict: that predators can be evaded either by reducing detectability (through freezing) or by overt escape actions that transiently increase detectability. Because escape behavior draws attention and interrupts ongoing foraging by the prey, an optimal strategy finds a balance between these two defensive behaviors.

Behavioral escalation in predator-prey and within-species aggression contexts. A) Predator imminence theory describes an organization of defensive strategies in which level of predatory threat shifts the behavior animals express. Figure adapted from Fanselow & Lester, 1988. B) Escalation of aggressive encounters between conspecifics shows a similar scaling of behavior, a strategy thought to minimize risk of injury. Escalation could serve to establish animals’ relative resource holding potentials, or could be a “war of attrition” process of animals communicating their level of commitment to a contested resource.

Outside of predator-prey conflict, violence in nature occurs predominantly between members of the same species, and rarely results in serious injury. The survival value of intra-species aggression has been questioned since Darwin, who hypothesized that aggression ensures that the fittest members of a species obtain greater territories and resources. Subsequent theories further posed that aggression could help balance distribution of animals across available environment 49 . In social species, establishment of dominance hierarchies is thought to stabilize groups and create networks that shape the flow of information among group members 50 , 51 . An extensive literature in game theory has explored the structure of aggression, and gave rise to the concept of “evolutionarily stable strategies”: behavioral strategies in a species that are stable to small perturbations in the form of mutant strategies 52 .

Normative models of aggression often focus on the observation that, like defensive behavior, aggressive behaviors vary in their intensity ( Figure 3B ). Escalation of aggression could serve several functions. In the “resource holding potential” hypothesis, escalation allows animals to exchange information about relative fitness while minimizing risk of injury, by only escalating if the relative fitness of the pair remains unclear ( Figure 2B ) 53 , 54 . This hypothesis predicts that animals more closely matched in size escalate aggression further than more asymmetric pairs, a prediction that holds to varying extents across species 55 , 56 . Alternatively, the “war of attrition” model 57 poses that escalation signals the cost an animal is willing to pay for a resource, explaining how manipulations such as food deprivation might motivate an animal to escalate further or faster 58 . Escalation may also simply reflect the animal’s relative levels of aggression and fear 59 . Models of aggressive escalation were extensively investigated in the 1970’s, 80’s, and 90’s, in species ranging from dung beetles to hermit crabs to red deer; this work is excellently reviewed in 60 .

Going forward.

Computational behavior analysis is experiencing a period of explosive growth, fueling and fueled by a push among neuroscientists to study more complex and naturalistic animal behaviors. Yet existing normative models of behavior largely precede the computer vision revolution in animal tracking. New behavior quantification methods will allow more rigorous testing of predictions made by classical normative models, whereas generative modeling provides methods for capturing behavioral control policies based on precise quantification of animals’ sensory environment and behavior. An open challenge for modern behavior analysis is to re-evaluate our normative theories of behavior in the light of more plentiful data, to determine how useful they are as descriptors of animals’ actions.

Towards this goal, the theoretical framework that has been established around exploration and foraging strategies provides rich ground for experimental exploration. Exploratory behavior and learning in neuroscience are often studied in conceptual tasks that do not require physical exploration, permitting neural correlates to be measured 61 , 62 . An exciting promise of computational behavior analysis is that it allows exploration to be studied in complex physical environments such as mazes 63 , 64 , where—in striking contrast to conceptual exploration—mice rapidly learn long sequences of actions, and can show one-shot learning of their home path out of the maze 64 . By tracking exploring animals’ posture, researchers can detect deliberative actions such as “vicarious trial and error”, when animals pause to investigate options at a choice point 65 , 66 . Quantification of deliberative actions may help study another normative theory of exploration, the principle of effort minimization, which predicts that as an environment or scenario becomes familiar animals may swap between decision strategies, balancing low-effort habit and high-effort planning and simulation 67 .

Analysis and generative modeling of animal actions and sensory environments during naturalistic foraging could also help determine how animals construct and update their internal model of food availability in their environment. For example, recent work has demonstrated that imitation learning can capture learned foraging behavior in a head-fixed task 62 . Elimination of the behavior annotation bottleneck could also allow for higher volume experiments testing the sensitivity of foraging behavior to combinations of environmental parameters such as degree of patch structure, presence of threat or conspecific cues, choices between food types, or changes in food availability.

Normative theories of dyadic interactions have often focused on the rules governing transitions between behaviors: either the switch from freezing to escape in predator imminence theory, or the gradual escalation of aggression in intra-species encounters. These progressions are challenging to study in the lab, as animals show rapid habituation and priming after a few presentations of a threating stimulus or conspecific. With automated analyses, classical assays like the resident-intruder paradigm could be expanded to longer timescales in enriched environments, so that animals have time and space for more naturalistic encounters to unfold 68 .

Finally, one exciting area for exploration is to develop theories and models for how competing motivational signals should interact to shape behavior. Animal behavior in the wild strikes a balance between competing drives: hunger and thirst, predator defense, and social (reproductive, parenting, and territorial) motivations. These drives can be thought of as low-dimensional intervening variables between sensation and action 69 : any single drive can be affected by multiple sensory cues- for example, both mating experience 70 and social isolation 71 make mice more aggressive, and both dry food and salt make mice thirsty. And drives can be linked to any learned behavior: both aggression and thirst can be used to motivate nose-poking or lever-pressing in operant tasks 72 , 73 .

We cannot measure an animal’s aggressive motivation directly, but we can apply experimental manipulations that we know will alter aggression—and also hunger, thirst, stress, reproductive state, or time of day—and measure resulting changes in behavior. How might these different state manipulations interact? Before we can generate hypotheses of how the brain is controlling behavior, it is good to stop and ask: how complex is behavior in the first place? If the annotation bottleneck can be resolved, we might begin to use high-throughput behavioral studies to ask whether intervening variables of hunger, thirst, fear, or aggression are indeed the one-dimensional signals we call them, or whether behavioral control policies and their underlying neural drives are more complex.

Future challenges for descriptive models of behavior

Performance..

A particular challenge facing unsupervised behavior discovery is deciding what constitutes a “good” representation of behavior 74 . Common metrics for evaluating the quality of an unsupervised behavior discovery method include agreement with human annotations, ability to account for variance in neural data, and utility in distinguishing between strains or conditions ( Figure 1B ), however many paper apply these only to in-house datasets, making it difficult to compare methods. Establishing benchmark datasets for evaluating behavior algorithms may help overcome this issue 75 .

Generalization.

Behavior classifiers and pose estimators trained in one lab rarely perform well out-of-the-box when used by another lab, unless the two groups standardize their data acquisition setup. Creating algorithms that can be tuned to new environments, or that are trained on larger datasets, would help ensure replicability of the results of computational analyses.

Interpretability.

A common complaint about machine learning methods is that they are a black box. However, automated methods could actually become more transparent in their choices than human annotators, who can struggle to communicate their decision process during annotation. Efforts to make human-interpretable behavior classification tools may help scientists have greater confidence in their results 76 .

Long timescales.

Behavior has structure over multiple scales: from the coordination of actions into sequences 77 , to effects of priming 78 , habituation 79 , satiety 80 , 81 , and time of day 82 , 83 . Long timescales of behavior can be captured intrinsically by sequence learning 84 or hierarchical models 34 , or may be recovered post-hoc by coarse-graining of action sequences 85 . But there is still ample space for exploration in this area.

Level of granularity.

Neural recording datasets introduce another challenge for behavior analysis: not just finding the representation of behavior that best captures observed postural variance, but finding the right level of granularity to account for observed neural activity. The optimal behavior representation in this sense will likely be different for different brain regions, with some regions better explained by fine-grained, sub-second action motifs, and other regions more correlated with an animal’s overall behavioral objective (such as aggression or reproduction.) How best to integrate quantitative behavior analysis with neural imaging datasets is an exciting area of ongoing research.

Interacting agents.

Multi-agent behavior poses unique challenges for analysis, as these datasets include times when animals are interacting and times when each individual is behaving independently. This affects unsupervised methods in particular, as the relative positioning of agents is sometimes critical for interpreting behavior and sometimes doesn’t matter at all. We hypothesize that one way to overcome this challenge is to base behavior discovery methods on forecasting models, which could be trained to incorporate postural information from an agent’s partner only when that information is relevant to predicting what the agent does next.

Acknowledgements

I am grateful to Jennifer J. Sun, Yisong Yue, Pietro Perona, and David J. Anderson for helpful discussions on the organization of behavior and the application of machine learning to behavior analysis, and to Brady Weissbourd for discussions on theories of social behavior. This manuscript was supported by the National Institute of Mental Health of the National Institutes of Health under Award Number R00MH117264. The content is solely the responsibility of the author and does not necessarily represent the official views of the National Institutes of Health.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Works cited