snowball sampling qualitative research pdf

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

Enhancing the sample diversity of snowball samples: Recommendations from a research project on anti-dam movements in Southeast Asia

Julian kirchherr.

1 Faculty of Geosciences, Utrecht University, Utrecht, the Netherlands

2 School of Geography and the Environment, University of Oxford, Oxford, United Kingdom

Katrina Charles

Associated data.

All data underlying the study are within the paper and its Supporting Information files.

Snowball sampling is a commonly employed sampling method in qualitative research; however, the diversity of samples generated via this method has repeatedly been questioned. Scholars have posited several anecdotally based recommendations for enhancing the diversity of snowball samples. In this study, we performed the first quantitative, medium- N analysis of snowball sampling to identify pathways to sample diversity, analysing 211 reach-outs conducted via snowball sampling, resulting in 81 interviews; these interviews were administered between April and August 2015 for a research project on anti-dam movements in Southeast Asia. Based upon this analysis, we were able to refine and enhance the previous recommendations (e.g., showcasing novel evidence on the value of multiple seeds or face-to-face interviews). This paper may thus be of particular interest to scholars employing or intending to employ snowball sampling.

Introduction

Snowball sampling is a commonly employed sampling method in qualitative research, used in medical science and in various social sciences, including sociology, political science, anthropology and human geography [ 1 – 3 ]. As is typical of terms adopted by a variety of fields, however, the phrase ‘snowball sampling’ is used inconsistently across disciplines [ 4 ]. The most frequently employed definition, suggested by Patton [ 5 ], Atkinson and Flint [ 6 ], Cohen and Arieli [ 7 ] and Bhattacherjee [ 8 ], is as a sampling method in which one interviewee gives the researcher the name of at least one more potential interviewee. That interviewee, in turn, provides the name of at least one more potential interviewee, and so on, with the sample growing like a rolling snowball if more than one referral per interviewee is provided.

This definition can initially seem self-explanatory, which may explain why snowball sampling is rarely discussed in most peer-reviewed papers that employ it. Various scholars use snowball sampling in their empirical work, but most provide only limited information on the method (see, e.g., [ 9 – 13 ]). Similarly, qualitative research textbooks often lack substantive discussion of snowball sampling (e.g., [ 8 , 14 – 19 ]). Bailey [ 14 ], for instance, devotes only a half-page of his 595-page book on social research methods to snowball sampling, acknowledging that ‘snowball sampling procedures have been rather loosely codified’ ([ 14 ], p. 96), an observation echoed by Penrod et al. [ 3 ].

This paper focuses on snowball sampling procedures, which we define as those actions undertaken to initiate, progress and terminate the snowball sample [ 1 , 20 ]. Despite the lack of substantive writing on snowball sampling as a method, several authors [ 2 , 3 , 21 ] have provided recommendations for enhancing a sample’s diversity in snowball sampling procedures (we discuss this further in Section 4). However, as this advice is not based on a quantitative analysis of evidence, but only on anecdotal evidence, there is a risk that these recommendations are based on coincidence. The aim of this paper is to provide advice on enhancing the sample diversity of a snowball sample. This advice is grounded in a medium- N analysis of relevant evidence, thus reducing the probability of positing advice that is based on coincidence [ 22 ]. A medium- N analysis is generally based on 10–100 cases, whereas anecdotal evidence is usually based only on a handful of cases [ 23 , 24 ]. At the core of our work, we provide descriptive analyses of various commonly prescribed strategies for enhancing the sample diversity of a snowball sample. These analyses are based on reach-outs to 211 individuals via snowball sampling for a research project on anti-dam movements in Southeast Asia, resulting in 81 interviews conducted between April and August 2015. As far as we are aware, ours is the first medium- N analysis to focus on enhancing the sample diversity of a snowball sample.

The remainder of this paper is organised as follows: in Section 2, we discuss snowball sampling as a method; in Section 3, we present the research project on anti-dam movements in Southeast Asia that served as the basis for our medium- N analysis on snowball sampling procedures; in Section 4, we present and discuss insights on snowball sampling procedures based upon this analysis as well as our resulting recommendations; finally, in Section 5, we summarise our argument.

Throughout this paper, we employ social science methodology terminology. We define key terms for this paper such as ‘snowball sampling’ or ‘sampling’, since these terms are not consistently codified in the scholarly literature. Due to limited space, however, we refrain from defining terms we have deemed common in this field of study, referring only to the relevant literature.

On snowball sampling

Traditional sampling methods are comprised of two elements [ 25 , 26 ]. First, a full set of data sources is defined, creating a list of the members of the population to be studied, known as a sampling frame. Second, a specific sample of data is collected from this sampling frame. Snowball sampling defies both elements, since it does not rely upon a sampling frame [ 27 ] (which may indicate that a different term for snowball sampling would be more accurate). Snowball sampling is often employed when no sampling frame can be constructed.

Researchers frequently cannot construct a sampling frame if a difficult-to-reach population is to be studied. Difficult-to-reach-populations are also referred to as ‘hard-to-reach-populations’ [ 28 ], ‘hidden populations’ [ 29 ] or ‘concealed populations’ [ 21 ] in the scholarly literature. Although not all scholars may agree that these terms are interchangeable, we deem them interchangeable for the purposes of this paper. For further discussion of this terminology, see [ 30 , 31 ].

A difficult-to-reach population does not wish to be found or contacted (e.g., illegal drug users, illegal migrants, prostitutes or homeless people [ 6 , 31 ]). Snowball sampling was originally used by researchers to study the structure of social networks [ 32 ]. The earliest empirical account of snowball sampling is from 1955 [ 33 ], with snowball sampling first described as a method in 1958 [ 34 ]. While it is still used to study the structure of social networks [ 35 ], over the last few decades, the method’s key purpose has largely transformed ‘into […] an expedient for locating members of a [difficult-to-reach] population’ ([ 36 ], p. 141).

Researchers grounded in quantitative thinking, such as Lijphart [ 37 ] and King et al. [ 38 ], tend to view the drawing of a random sample from a sampling frame as the gold standard of data collection. Even these researchers may nevertheless consider non-probability sampling methods, such as snowball sampling, a ‘necessary and irreplaceable sampling [method]’ ([ 39 ], p. 367) when confronted with difficult-to-reach populations, particularly if the dismissal of snowball sampling would mean that no research could be conducted at all. Ultimately, ‘an important topic is worth studying even if very little [access to] information is available’ ([ 38 ], p. 6). Still, some of those grounded in quantitative thinking call snowball sampling a method ‘at the margin of research practice’ ([ 6 ], p. 1), since the lack of a sampling frame means that, unlike individuals in a random sample, individuals in a population of interest do not have the same probability of being included in the final sample. Findings from a snowball sample would therefore not be generalisable [ 40 ] (on generalisability, see [ 41 ]).

Several qualitative scholars rebut such criticism. Creswell, for instance, notes that ‘the intent [of qualitative research] is not to generalise to a population, but to develop an in-depth [and contextualised] exploration of a central phenomenon’ ([ 42 ], p. 203). Others [ 1 , 39 ] specifically oppose quantitative scholars’ negative framing of snowball sampling, arguing that this method would ‘generate a unique type of social knowledge’ ([ 1 ], p. 327). Due to the diversity of perspectives gathered, this knowledge would be particularly valuable for an in-depth and contextualised exploration of a central phenomenon. We therefore define the diversity of a sample as a measure of the range of viewpoints that have been gathered on a central phenomenon.

Researchers critical of snowball sampling respond to this defence by arguing that the method is unable to ensure sample diversity, which is a necessary condition for valid research findings. Indeed, some scholars have stated that snowball samples underrepresent and may even exclude those least keen to cooperate, since referrals may not materialise in an interview if a potential interviewee is only somewhat keen or not at all keen to be interviewed [ 3 , 43 ]. Similarly, potential interviewees with smaller networks may be underrepresented, as they are less likely to be referred for an interview [ 31 , 44 ]. Those with smaller networks may also be in a specific network whose different perspectives may be of interest but are excluded in the final sample. Meanwhile, snowball sampling is said to over represent those interviewees (and their respective networks) that the interviewer spoke with first; the relevant literature refers to this as ‘anchoring’ [ 20 , 39 ].

We do not aim to argue the ‘validity’ of the method, but rather to inform snowball sampling methodologies in order to promote sample diversity. From a qualitative perspective, ‘validity’ can be defined as ‘the correctness or credibility of a description, conclusion, explanation, interpretation or other sort of account’ ([ 45 ], p. 87), while quantitative researchers frequently use the terms ‘generalisability’ and ‘(external) validity’ interchangeably [ 46 , 47 ]. The term ‘validity’ is contested among qualitative researchers, and some qualitative researchers entirely reject the concept for qualitative work [ 48 , 49 ]. We do not aim to resolve this debate via this paper; instead, we focus on the (seemingly less-contested) term ‘sample diversity’. While we acknowledge that this term is not codified in qualitative textbooks such as the SAGE Encyclopedia of Qualitative Research Methods , sample diversity is considered desirable by the various qualitative scholars we reviewed. Boulton and Fitzpatrick demand, for instance, that qualitative researchers ‘ensure that the full diversity of individuals […] is included [in their sample]’ ([ 50 ], p. 84), a mandate echoed by other scholars [ 16 , 51 – 53 ].

In order to operationalise the concept of sample diversity, we used five key methodological recommendations to inform our research. In this paper, we use quantitative analyses from our experiences with snowball sampling to further reflect on these recommendations, which are briefly described below.

Prior personal contacts of the researcher are required

Patton ([ 5 ], p. 176) notes that snowball sampling ‘begins by asking well-situated people: “Who knows a lot about ____? Who should I talk to?”‘. In the absence of a sampling frame for the population of interest, however, the researcher must retain at least some prior personal or professional contacts in the population of interest which can serve as the seeds of the snowball sample [ 2 , 54 ]. Waters contends that building a diverse snowball sample ‘depend[s] almost exclusively on the researcher’s [prior personal or professional] contacts’ ([ 39 ], p. 372).

Sample seed diversity is important

Morgan [ 21 ] has claimed that the ‘best defence’ against a lack of sample diversity is to begin the sample with seeds that are as diverse as possible. Others echo this advice [ 3 , 39 , 55 ], arguing that it is ‘compulsory for the researcher to ensure that the initial set of respondents is sufficiently varied’ ([ 55 ], p. 55). The term ‘chain referral sampling’ has been used for snowball samples that are strategically built via multiple varying seeds [ 3 ].

Technology means face-to-face interviews are no longer required

Some researchers have argued that face-to-face interviews are obsolete. For instance, over 25 years ago, it was claimed there were ‘no remarkable differences’ ([ 56 ], p. 211) between information collected via telephone and information collected via face-to-face interviews. The increasing use of telecommunications in recent years is likely to have further reduced barriers to remote interviewing, and various scholars [ 57 , 58 ] continue to claim that ‘evidence is lacking that [telephone interviews] produce lower quality data’ ([ 59 ], p. 391). In particular, they have highlighted the benefits of using Skype for semi-structured interviews [ 57 ].

However, for snowball sampling, face-to-face interviews help to generate the trust that scholars claim is required in order to gain referrals [ 1 , 31 , 39 , 60 ]. Noy argues that ‘the quality of the referring process is naturally related to the quality of the interaction: […] if the researcher did not win the informant’s trust […], the chances the latter will supply the former referrals decrease’ ([ 1 ], p. 334).

Persistence is necessary to secure interviews

Although the value of persistence may be considered self-evident by some scholars, it is seen by multiple academics [ 61 – 63 ] as a central virtue of qualitative researchers. Many young career scholars who embrace snowball sampling are likely to hear such advice as, ‘If you cannot interview your envisaged interviewees initially, don’t give up!’. A ‘helpful hint’ for qualitative researchers seeking informants is, ‘Persevere–repeat contact’ [ 64 ].

More waves of sampling are required to access more reluctant interviewees

As a remedy for snowball sampling’s previously discussed bias towards excluding those least keen to be interviewed, multiple scholars suggest pursuing a snowball sample for multiple waves (with a new sampling wave reached once an interviewee introduces the interviewer to one or more potential interviewees) [ 65 – 68 ]. Those suggesting this remedy assume that pursuing more waves increases the likelihood of being referred to an interviewee from a particularly difficult-to-reach population who is at least somewhat keen to be interviewed.

Approval for this study was granted by the Central University Research Ethics Committee (CUREC) of the University of Oxford. Our population of interest for our research project were stakeholders in Southeast Asia’s dam industry. Since ‘the most dramatic conflicts over how to pursue sustainable development’ ([ 69 ], p. 83) have occurred over the construction of large dams, we see this industry as a conflict environment with widely varying viewpoints. A conflict environment is one in which people perceive their goals and interests to be contradicted by the goals or interests of the opposing side [ 70 ]. The major conflicting parties in the dam industry tend to be local and international non-governmental organisations (NGOs) and academics (usually keen not to construct a particular dam) versus international donors, the private sector and governments (usually keen to construct a particular dam) [ 71 , 72 ]. Each sub-population operating in a conflict environment can be considered difficult to reach since fear and mistrust are often pervasive [ 7 ]. Snowball sampling is a suitable research method in conflict environments because the introductions through trusted social networks that are at the core of this method can help interviewees to overcome fear and mistrust, which, in turn, ensures access [ 7 ]. This access is needed to gather the widely varying viewpoints in the hydropower industry, in particular viewpoints with regards to what constitutes just resettlement [ 73 , 74 ]. Based on this rationale, we chose snowball sampling as the main method for our research.

In order to ensure sample diversity for our research project on anti-dam movements in Southeast Asia, we aimed to gather perspectives mostly from six main sub-populations: (1) local NGOs, (2) international NGOs, (3) international donors, (4) academia, (5) the private sector and (6) the government. We hypothesized that ‘dam developers’, a main sub-category of the interviewee category ‘private sector’, would be the most significant challenge to ensuring the diversity of our sample. Early in our process, many of the scholars with whom we discussed our research project argued that it would be impossible to interview a dam developer from a Chinese institution; meanwhile, researchers from a comparable research project that ended approximately when our project started reported being unable to interview any dam developers from European institutions. We also initially failed to collect data from dam developers: for instance, a survey we initiated that was distributed by Aqua~Media (host of a major global dam developer conference) to more than 1,500 dam developers yielded just five responses, only one of which was complete. We considered this weak response rate to be due, at least in part, to the dam industry’s negative view of academicians since the publication of Ansar et al. [ 75 ], which Nombre ([ 76 ], p. 1), the president of the International Commission on Large Dams (ICOLD), called ‘[highly] misleading’.

None of our researchers had significant direct links to the dam industry upon the start of the project; however, we did retain a variety of indirect links. Our researchers had past links to a management consultancy that serves various dam industry players, (more limited) links to an international donor working in the hydropower sector and links to activists in Myanmar advocating against dam projects.

After a favourable ethics review of our study by the CUREC of the University of Oxford, we commenced semi-structured interviews in April 2015, mostly via cold calls (we include cold e-mails in the term ‘cold calls’ throughout this paper). Initially, we conducted research via telephone only. We then undertook field research in Singapore, Myanmar and Thailand from June to August 2015 and terminated our data collection in late August 2015.

In total, 81 semi-structured interviews were carried out during this period. From a qualitative perspective, this is a relatively large sample size (for instance, the average qualitative PhD dissertation is based on 31 interviews [ 77 ]); from a quantitative perspective, however, the sample size is quite small [ 78 ]. Of our 81 interviews, 48 (59%) were conducted via telephone, 26 (32%) face-to-face and 7 (9%) online, either via e-mail or an online survey. Most of our interviews (57%) were carried out in July in Myanmar. Of our 81 interviewees, only 24 (30%) were women. Researchers who employ snowball sampling frequently employ personal/professional contact seeds and cold call seeds to build their sample (e.g., [ 2 , 79 , 80 ] with a seed defined as the starting point of a sample [ 65 ]). Of the 81 interviews analysed, 53 (65%) were rooted in a personal or professional contact ( Fig 1 ) (i.e. the seed of the interview pathway was a contact we had already retained prior to the research project). The remaining 28 (35%) interviews were rooted in cold calls.

Given the sensitive nature of the interview topic, all interviewees were assured anonymity. Thus, all of the interviews are coded, with the first letter indicating the mode of interview ( T for telephone, F for face-to-face, O for online survey or e-mail), the second letter indicating the category of interviewee ( A for academia, G for government, I for international donor, NI for international NGO, NL for national NGO, P for private sector) and the sequence of numbers indicating the interview number within a particular mode. Researcher A is indicated by RA , Researcher B by RB ; CON represents a conference event. Bold type indicates that an interview was completed, while X that an interview was not completed.

As outlined in the previous section, snowball sampling is sometimes criticised for producing samples that lack sample diversity. To address this criticism, we reviewed the (scarce) literature on enhancing sample diversity via snowball sampling procedures prior to commencing our study. Upon reflection during our research, we chose to pursue our analysis retrospectively in order to challenge some of the recommendations provided in literature. Our analysis is structured alongside the five core pieces of advice found in this literature ( Table 1 ). Our results are based on a quantitative analysis of the 81 interviews we conducted. Although we endeavoured to include all interview attempts, some initial cold calls may have been overlooked in this retrospective approach. Therefore, some of our analysis, particularly in Section 4.4, may be too optimistic. Overall, we were able reconstruct 211 reach-out attempts.

Sample diversity is measured by representation from five identified sub-groups.

Results and discussion

On prior personal and professional contacts.

Our analysis provides evidence that sample diversity can be reached even if no prior personal or professional contacts to the population of interest have been retained. The seeds of the interviews are depicted in Fig 2 , with the left side of the figure depicting the 53 interviews based on a personal or professional contact and the right side depicting the 28 interviews that were based on cold calls. This figure shows two main points of interest: first, both types of seeds include interviews in each interview category; second, the interview sub-category ‘dam developer’, which we hypothesised would be the most difficult to include in the sample, is also covered by both types of seeds. We can therefore conclude that a diverse sample could have been built even if we had relied solely on cold calls.

It is acknowledged, however, that building a snowball sample from cold calls is particularly labour-intensive [ 39 ]: in our research, only 25% of our cold calls led to an interview, compared to 62% of the referrals. Significant differences in the value of referrals persist from one interviewee group to another ( Fig 3 ). We measure the value of referrals via a concept we call ‘network premium’. To gauge the network premium, we subtracted the cold call response rate (i.e., the number of interviews initiated via cold calls divided by the total number of cold calls) from the referral response rate (i.e. the number of interviews initiated via referrals divided by the total number of referrals). Referrals were the most valuable when contacting international donors and private sector players, with network premiums of 74% and 52%, respectively, indicating that these groups are particularly difficult-to-reach populations.

(1) Unable to retrace for 13 identified reach-outs if initiated via referral or cold call; four reach-outs coded as ‘Other’. (2) Unable to retrace for one interview carried out via referral coded as ‘Other’. (3) Including personal contacts and contacts via conferences. (4) Referral response rate–Cold call response rate.

The overall results from these analyses are encouraging for scholars interested in researching a population to which no personal or professional contacts are retained prior to the research project. While personal or professional contacts maintained to the research population of interest can accelerate the research endeavour, our results also showcase that (at least for our topic of interest) a diverse sample can be built from cold calls if a researcher is willing to invest some time in reach-outs.

On seed variation

Our research confirms the scholars’ advice that seed diversity is important. Fig 4 (a variation of Fig 2 ) depicts the completed interviews from a seed perspective, with RA, RB and cold calls as the three main seeds of the sample. The sample built via RA, who has a background in the private sector, is largely biased towards this sector, with 47% of all interviews seeded via RA private sector interviews. RB conducted 57% of interviews, whose background is closest to local NGOs, were with local NGOs. Meanwhile, the sample built via cold calls indicates no significant biases towards any interviewee category. Interviews based on the network of RB included one (TNL17) with a leading activist from a remote area of Myanmar who provided unique insights into the early days of an anti-dam campaign. This insight helped us to develop a narrative of the campaign that was not skewed to the later days of the campaign and the activists prominent in these later days. The sample diversity ensured via RB was thus central to the quality of our research.

It is noteworthy that the three different seeds in Fig 4 include interviews in all interviewee categories, including the sub-category ‘dam developer’ (the sole exception is the interviewee category ‘international NGO, which contains zero interviews for RB). This indicates that, at least for our topic of interest, a fairly diverse sample can be generated even if the researcher is unable to vary her or his seed, although the overall data suggest that seed variation can significantly enhance sample diversity. Fig 3 may therefore be viewed as a case for collaboration among researchers; if researchers with different backgrounds and different personal and professional contacts to the population of interest begin to collaborate, such collaborations are bound to contribute to sample diversity.

On face-to-face interviews

Our descriptive analysis provides evidence to further support the argument that face-to-face interviews are redundant, with our data indicating that face-to-face interviews can lead to more sought referrals than telephone interviews (perhaps since trust may be more readily established via face-to-face conversations than over the telephone). Fig 5 aims to quantify the value of face-to-face interviews. Overall, 30 (37%) of our interviews were initiated via prior face-to-face conversations, while prior telephone conversations and online contact each led to only eight interviews (10%). An examination shows that of the nine interviews conducted with dam developers, the interviewee sub-category deemed most difficult to access, seven (78%) were initiated via prior face-to-face interviews, while not a single telephone interview led to a referral to a dam developer. These interviews proved to be essential for our research. For instance, one Chinese dam developer challenged a claim from numerous NGOs that his company would not engage with NGOs, which, in turn, allowed us to present a more balanced portrayal of the interplay between Chinese dam developers and NGOs.

(1) Comprises interviews with those already retaining a personal or professional contact prior to the research project.

While our research did not investigate whether face-to-face interviews lead to lower-quality data than telephone interviews, our data provide tentative evidence that face-to-face interviews are not obsolete; they can still be helpful for those employing or intending to employ snowball sampling, since these interviews can lead to more sought referrals and thus enhanced sample diversity. We acknowledge, however, that this finding may not be true for all populations. For instance, studies on individuals with sexually transmitted diseases have found that these interviewees (particularly men) tend to report more truthfully in an audio-computer-assisted self-interview (ACASI) than in a face-to-face interview, since interviewees tend to be more comfortable reporting on sexually transmitted diseases to a computer than to a live person [ 81 , 82 ].

On persistence

Our data suggest that persistence can indeed enhance sample diversity, but we can also conclude that excessive persistence does not necessarily yield dividends. Instead of distributing a great many interview reminders during our study, we reached out to the majority of our proposed interview subjects only once. Nevertheless, the scarce data we collected regarding persistence indicates its value. We map this data in Fig 6 , with the left side depicting our success rate in relation to the number of reach-outs (either one, two or three) and the right side depicting a deep dive on success rates achieved with two reach-outs (distinguishing between reach-out attempts to unknown potential interviewees and those to whom we were referred by other interviewees). We sent one interview reminder to 28 of our proposed interviewees. This led to 10 additional interviews, a success rate of 36%, equalling 12% of the total interviews analysed for this paper. Reminders appear to be only somewhat more helpful when contacting referrals in comparison to their usefulness with cold calls–a single reminder led to an interview in 39% of our cases for the former group and 38% for the latter. One of the most valuable interviews for our research gained via a reminder was with the CEO of a Burmese dam developer. This interviewee compared Chinese and European dam developers in Myanmar, which helped us to further refine our narrative on social-safeguard policy adherence by Chinese dam developers in Myanmar.

(1) Number of reach-outs unknown for 32 reach-outs. Eight potential interviewees responded, but refused interview.

Excessive persistence, however, does not appear to be worthwhile. We sent three reminders to seven of our envisaged interviewees, but as Fig 6 shows, this did not lead to a single additional interview. While our data does not suggest that excessive persistence is helpful to researchers, it may also not be recommended for ethical reasons. A potential interviewee who does not respond to an interview request after two reach-outs may be indicating via this non-response that she or he is not interested in participating in the research. If a single request remains unanswered, the researcher may hypothesise that, for instance, the e-mail was overlooked, a hypothesis particularly likely when conducting interviews with time-pressed leaders of organisations. Indeed, all 10 interviews only carried out upon the second reach-out were interviews with interviewees in management positions.

Our data on persistence provide some evidence that those employing or intending to employ snowball sampling can enhance sample diversity if every reach-out is carefully tracked and followed by a reminder. We typically sent a reminder after one week if no response was obtained upon the first reach-out. This persistence may help to include those least keen to be interviewed for a research endeavour.

Our data show some evidence that, for our topic of study, pursuing interviews for even a few waves provided the perspectives of particularly difficult-to-reach populations and thus achieved sample diversity. More than 60% of our interviews were conducted in the zeroth or first wave ( Fig 7 ). These include seven of the nine interviews conducted with dam developers, the sub-category we deemed most challenging to interview. The remaining two interviews with dam developers were conducted in the second wave. However, not a single interview with a dam developer was carried out in the third wave and beyond, although a fifth of our total interviews were carried out in the third or later waves. Pursuing interviews for multiple waves nevertheless yielded novel insights. For instance, interview FNL12, which was conducted in the sixth wave, yielded insights on small dam construction in Myanmar–a topic of (some) interest to our research endeavour, but not covered in detail by previous interviews. Furthermore, we note that our finding regarding the limited value of multiple waves may also be specific to our population, with this finding perhaps indicating a low degree of network segmentation in the population in question [ 83 ]. Meanwhile, a high degree of network segmentation may impede the pursuance of multiple waves, since interviewees may lack the suitable contacts for a referral [ 84 ].

While additional waves can lead to novel insights, our overall data on waves provide some evidence that the number of waves pursued is not a definitive indicator for sample diversity. Even very few waves can yield access to particularly difficult-to-access populations.

Our quantitative analysis of pathways to delivering sample diversity in snowball samples yielded the following revisions to the literature’s recommendations:

• Prior personal contacts are not essential for achieving sample diversity but tend to be helpful, as generating new contacts during research can be labour-intensive.
• Sample seed diversity is important to achieving sample diversity.
• Face-to-face interviews build trust and can help to generate further referrals.
• Persistence (within reason) is helpful in securing interviews.
• Sample diversity is not necessarily enhanced if a seed is advanced over numerous waves.

We do not claim that these insights are comprehensive, but we believe that these interpretations of our data may serve as a starting point for future scholars using snowball sampling procedures. All of the analyses presented in this section are based only on descriptive statistics. This means, for instance, that we cannot control for confounds such as effort [ 85 ]. An experimental research design would yield the most robust insights on sampling procedures to enhance the sampling diversity of a snowball sample (with, for instance, one research project staffed with scholars with relevant personal or professional contacts and another staffed with scholars without relevant contacts).

Overall, this work aims to advance the literature on snowball sampling as a qualitative sampling approach. While snowball sampling procedures may qualify ‘as the least “sexy” facet of qualitative research’ ([ 1 ], p. 328), these procedures are ‘not self-evident or obvious’ ([ 20 ], p. 141), since the snowball sample does not ‘somehow magically’ ([ 20 ], p. 143) start, proceed and terminate when a scholar attempts to develop a diverse sample. Rather, continuous, deliberate effort by the researcher(s) is required. Our paper has attempted to provide some insights on this effort.

Unfortunately, we developed the idea to write this paper only during the course of our research project, and thus some of our data may be skewed. For instance, we may not have been able to trace all original reach-out attempts and our data on persistence may therefore be biased. Some of those scholars grounded in quantitative thinking may also claim that the insights outlined in Section 4 lack external validity since our sample size is relatively small from a quantitative methodological perspective. In addition, our population was very specific and thus may not be comparable to other difficult-to-reach populations, and we also did not adopt an experimental research design as described above. Hence, we encourage scholars to replicate our findings via their respective research projects that employ snowball sampling. With many scholars claiming to feel more pressed than ever to deliver research results with maximum efficiency, we hope that these initial descriptive analyses of snowball sampling procedures provide some valuable insights to those employing or intending to employ this method and aiming to improve their management of it.

Supporting information

Acknowledgments.

We wish to thank our reviewers at PLOS ONE who provided constructive thoughts on this piece of work. We also thank Ralf van Santen for his outstanding contributions to this work as a research assistant.

Funding Statement

The authors received no specific funding for this work.

Data Availability

Have a language expert improve your writing

Run a free plagiarism check in 10 minutes, generate accurate citations for free.

• Knowledge Base

Methodology

• What Is Snowball Sampling? | Definition & Examples

What Is Snowball Sampling? | Definition & Examples

Published on August 17, 2022 by Kassiani Nikolopoulou . Revised on June 22, 2023.

Snowball sampling is a non-probability sampling method where new units are recruited by other units to form part of the sample . Snowball sampling can be a useful way to conduct research about people with specific traits who might otherwise be difficult to identify (e.g., people with a rare disease).

Also known as chain sampling or network sampling , snowball sampling begins with one or more study participants. It then continues on the basis of referrals from those participants. This process continues until you reach the desired sample, or a saturation point.

A number of criteria are used for the selection:

• The couple must have been together for a period of at least five years.
• The couple must live together now.
• The couple must live within a certain geographic area.
• The couple must have examples of changes or challenges they have experienced together (e.g., long-distance, illness or loss of a loved one).

Table of contents

When to use snowball sampling, types of snowball sampling, advantages and disadvantages of snowball sampling, other interesting articles, frequently asked questions about snowball sampling.

Snowball sampling is a widely employed method in qualitative research , specifically when studying hard-to-reach populations .

These may include:

• Populations that are small relative to the general population
• Geographically dispersed populations
• Populations possessing a social stigma or particular shared characteristic of interest

In all these cases, accessing members of the population can be difficult for non-members, as there is no sampling frame available.

Research in the fields of public health (e.g., drug users), public policy (e.g., undocumented immigrants), or niche genres (e.g., buskers) often uses snowball sampling.

This sampling method is also used to study sensitive topics, or topics that people may prefer not to discuss publicly. This is usually due to a perceived risk associated with self-disclosure. Snowball sampling allows you to access these populations while considering ethical issues , such as protecting their privacy and ensuring confidentiality.

Prevent plagiarism. Run a free check.

Snowball sampling begins with a convenience sample of one or more initial participants. Multiple data collection points (or waves) follow. These initial participants, called “seeds,” are used to recruit the first wave’s participants.

Wave 1 participants recruit wave 2 participants, and the sample expands, wave by wave, like a snowball growing in size as it rolls down a hill.

Depending on your research objectives , there are three different types of snowball sampling methods to choose from:

Linear snowball sampling

Exponential non-discriminative snowball sampling, exponential discriminative snowball sampling.

Linear snowball sampling relies on one referral per participant. In other words, the researcher recruits only one participant, and this participant, in turn, recruits only one participant. This process goes on until you have included enough participants in the sample.

Linear snowball sampling works best when there are few restrictions (called inclusion and exclusion criteria ) as to who is included in the sample.

As you finish up the interview , you ask them if they can refer someone else who also owns a tiny house. They happen to know someone, and pass the contact details to you. You interview them as well. Towards the end of the interview, you ask them to introduce you to one more person.

If more than two names are mentioned, it is a good idea to ask the interviewee how well they know those people, and then interview the person who is least known to them.

In exponential non-discriminative snowball sampling , the first participant provides multiple referrals. In other words, the researcher recruits the first participant, and this participant in turn recruits several others. The researcher includes all referrals in the sample. This type of snowball sampling is best used when you want to reach a larger sample.

In this method, participants give multiple referrals. However, the researcher screens those referrals, choosing only those who meet specific criteria to participate in the sample. The key difference between this and exponential non-discriminative snowball sampling is that not all referrals are included in the sample.

Exponential discriminative snowball sampling is most used when screening participants according to specific criteria is vital to your research goals.

As you inquire with your acquaintances, you find someone who bought a tiny house a year ago. At the end of the interview, you ask them if they know of other owners. You do not specify that the purchase has to be in the past three years.

As it happens, they do know of two more people who bought tiny houses in the same area as they did. You contact both, and find out that one bought the house four years ago and the other eight months ago. Since the one who bought the house four years ago does not meet your criteria, you only interview the other.

Like all research methods , snowball sampling has distinct advantages and disadvantages. It is important to be aware of these in order to determine whether it’s the best approach for your research design .

Advantages of snowball sampling

Depending on your research goals, there are advantages to using snowball sampling.

• Snowball sampling helps you research populations that you would not be able to access otherwise . Members of stigmatized groups (e.g., people experiencing homelessness) may hesitate to participate in a research study due to fear of exposure. Snowball sampling helps in this situation, as participants refer others whom they know and trust to the researcher.
• Since snowball sampling involves individuals recruiting other individuals, it is low-cost and easy to recruit a sample in this way.
• Unlike probability sampling , where you draw your sample following specific rules and some form of random selection , snowball sampling is flexible . All you need is to identify someone who is willing to participate and introduce you to others.

Disadvantages of convenience sampling

Snowball sampling has disadvantages, too, and is not a good fit for every research design.

• As the sample is not chosen through random selection , it is not representative of the population being studied. This means that you cannot make statistical inferences about the entire population and there is a high chance of research bias .
• The researcher has little or no control over the sampling process and relies mainly on referrals from already-identified participants. Since people refer others whom they know (and share traits with), this sampling method has a high potential for sampling bias .
• Relying on referrals may lead to difficulty reaching your sample . People may not want to cooperate with you, hesitate to reveal their identities, or mistrust researchers in general.

If you want to know more about statistics , methodology , or research bias , make sure to check out some of our other articles with explanations and examples.

• Student’s  t -distribution
• Normal distribution
• Null and Alternative Hypotheses
• Chi square tests
• Confidence interval
• Quartiles & Quantiles
• Cluster sampling
• Stratified sampling
• Data cleansing
• Reproducibility vs Replicability
• Peer review
• Prospective cohort study

Research bias

• Implicit bias
• Cognitive bias
• Placebo effect
• Hawthorne effect
• Hindsight bias
• Affect heuristic
• Social desirability bias

Snowball sampling is best used in the following cases:

• If there is no sampling frame available (e.g., people with a rare disease)
• If the population of interest is hard to access or locate (e.g., people experiencing homelessness)
• If the research focuses on a sensitive topic (e.g., extramarital affairs)

Snowball sampling relies on the use of referrals. Here, the researcher recruits one or more initial participants, who then recruit the next ones.

Participants share similar characteristics and/or know each other. Because of this, not every member of the population has an equal chance of being included in the sample, giving rise to sampling bias .

Snowball sampling is a non-probability sampling method , where there is not an equal chance for every member of the population to be included in the sample .

This means that you cannot use inferential statistics and make generalizations —often the goal of quantitative research . As such, a snowball sample is not representative of the target population and is usually a better fit for qualitative research .

Snowball sampling is a non-probability sampling method . Unlike probability sampling (which involves some form of random selection ), the initial individuals selected to be studied are the ones who recruit new participants.

Because not every member of the target population has an equal chance of being recruited into the sample, selection in snowball sampling is non-random.

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 Citation Generator.

Nikolopoulou, K. (2023, June 22). What Is Snowball Sampling? | Definition & Examples. Scribbr. Retrieved April 11, 2024, from https://www.scribbr.com/methodology/snowball-sampling/

Is this article helpful?

Kassiani Nikolopoulou

Other students also liked, what is non-probability sampling | types & examples, population vs. sample | definitions, differences & examples, sampling bias and how to avoid it | types & examples, unlimited academic ai-proofreading.

✔ Document error-free in 5minutes ✔ Unlimited document corrections ✔ Specialized in correcting academic texts

Snowball Sampling Method: Techniques & Examples

Julia Simkus

Editor at Simply Psychology

BA (Hons) Psychology, Princeton University

Julia Simkus is a graduate of Princeton University with a Bachelor of Arts in Psychology. She is currently studying for a Master's Degree in Counseling for Mental Health and Wellness in September 2023. Julia's research has been published in peer reviewed journals.

Learn about our Editorial Process

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.

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:

Snowball sampling, also known as chain-referral sampling, is a non-probability sampling method where currently enrolled research participants help recruit future subjects for a study.

Snowball sampling is often used in qualitative research when the population is hard-to-reach or hidden. It’s particularly useful when studying sensitive topics or when the members of a population are difficult to locate.

This sampling technique is called “snowball” because the sample group grows like a rolling snowball.

Non-probability sampling means that researchers, or other participants, choose the sample instead of randomly selecting it, so not all population members have an equal chance of being selected for the study.

Linear Snowball Sampling

• Linear snowball sampling depends on a straight-line referral sequence, beginning with only one subject. This individual subject will provide one new referral, which is then recruited into the sample group.
• This referral will provide another new referral, and this pattern continues until the ideal sample size is reached.

Exponential Non-Discriminative Snowball Sampling

• In exponential non-discriminative snowball sampling, the first subject recruited to the sample provides multiple referrals. Each new referral will then provide the researchers with more potential research subjects.
• This geometric chain sampling sequence continues until there are enough participants for the study.

Exponential Discriminative Snowball Sampling

• This type of snowball sampling is very similar to exponential non-discriminative snowball sampling in that each subject provides multiple referrals.
• However, in this case, only one subject is recruited from each referral. Researchers determine which referral to recruit based on the objectives and goals of the study.
• First, researchers will form an initial sample by drafting any potential subjects from a population (seeds).
• Even if only a couple of subjects are found at first, researchers will ask those subjects to recruit other individuals for the study. They recruit subjects by encouraging them to come forward on their own. Study participants will only provide specific names of recruited individuals if there is no risk of embarrassment or a violation of privacy. Otherwise, study participants do not identify any names of other potential participants.
• Current participants will continue to recruit others until the necessary sample size has been reached.

Snowball sampling requires special approval by an Institutional Review Board (IRB), whereby the researchers must provide a valid justification for using this method.

Researchers must also take precautions to protect the privacy of potential subjects, especially if the topic is sensitive or personal, such as studies of networks of drug users or prostitutes.

In addition, each respondent has the opportunity to participate or decline. Current participants in studies using this method do not receive any compensation for providing referrals, and study participants are not required to identify any names of other potential participants.

Example Situations

Snowball sampling is used when researchers have difficulty finding participants for their studies. This typically occurs in studies on hidden populations, such as criminals, drug dealers, or sex workers, as these individuals are difficult for researchers to access.

For example, a researcher studying the experiences of undocumented immigrants in a particular city. This population might be difficult to reach through traditional sampling methods due to fear of legal repercussions, lack of formal records, and other barriers.

The researcher might start by contacting a local organization that provides services to immigrants. Through this organization, the researcher could connect with a few willing individuals to participate in the study.

These initial participants (the “seeds”) would then be asked to refer the researcher to other undocumented immigrants they know who might also be willing to participate.

The new participants would then refer the researcher to others, and so on, creating a “snowball” effect where the number of participants grows as each person refers the researcher to others in their network.

The snowball sampling method is beneficial because current participants are likely to know others who share similar characteristics relevant to the study.

Members of these hidden populations tend to be closely connected as they share interests or are involved in the same groups, and they can inform others about the benefits of the study and reassure them of confidentiality.

Research Examples

• Researching non‐heterosexual women using social networks (Browne, 2002).
• Investigating lifestyles of heroin users (Kaplan, Korf, & Sterk, 1987).
• Identifying Argentinian immigrant entrepreneurs in Spain (Baltar & Brunet, 2012).
• Studying illegal drug users over the age of 40 (Waters, 2015).
• Assess the prevalence of irritable bowel syndrome in South China and its impact on health-related quality of life (Xiong, 2004).
• Obtaining samples of populations at risk for HIV (Kendall et al., 2008).

Enables access to hidden populations

Snowball sampling enables researchers to conduct studies when finding participants might otherwise be challenging. Concealed individuals, such as drug users or sex workers, are difficult for researchers to access, but snowball sampling helps researchers to connect to these hidden populations.

Avoids risk

Snowball sampling requires the approval of an Institutional Review Board to ensure the study is conducted ethically. In addition, each respondent has the opportunity to participate or to decline participation.

Saves money and time

Since current subjects are used to locate other participants, researchers will invest less money and time in planning and sampling.

Limitations

Difficult to determine sampling error.

Snowball sampling is a non-probability sampling method, so researchers cannot calculate the sampling error.

Bias is possible

Since current participants select other members for the sample, bias is likely. The initial participants will strongly impact the rest of the sample. In addition, an individual who is well-known and sociable is more strongly impacted by one who is more introverted.

Not always representative of the greater population

Because researchers are not selecting the participants themselves, they have little control over the sample. Researchers will thus have minimal knowledge as to whether the sample is representative of the target population.

• A sample is the participants you select from a target population (the group you are interested in) to make generalizations about. As an entire population tends to be too large to work with, a smaller group of participants must act as a representative sample.
• Representative means the extent to which a sample mirrors a researcher’s target population and reflects its characteristics (e.g. gender, ethnicity, socioeconomic level). In an attempt to select a representative sample and avoid sampling bias (the over-representation of one category of participant in the sample), psychologists utilize various sampling methods.
• Generalisability means the extent to which their findings can be applied to the larger population of which their sample was a part.

Felix-Medina, M. H., & Thompson, S. K. (2004). Combining link-tracing sampling and cluster sampling to estimate the size of hidden populations. JOURNAL OF OFFICIAL STATISTICS-STOCKHOLM- , 20 (1), 19-38.

Henderson, R. H., & Sundaresan, T. (1982). Cluster sampling to assess immunization coverage: a review of experience with a simplified sampling method. Bulletin of the World Health Organization , 60 (2), 253–260.

Malilay, J., Flanders, W. D., & Brogan, D. (1996). A modified cluster-sampling method for post-disaster rapid assessment of needs. Bulletin of the World Health Organization , 74 (4), 399–405.

Roesch, F. A. (1993). Adaptive cluster sampling for forest inventories. Forest Science , 39 (4), 655-669.

Smith, D. R., Conroy, M. J., & Brakhage, D. H. (1995). Efficiency of Adaptive Cluster Sampling for Estimating Density of Wintering Waterfowl. Biometrics , 51 (2), 777–788. https://doi.org/10.2307/2532964

Steven K. Thompson (1990) Adaptive Cluster Sampling, Journal of the American Statistical Association, 85:412,1050-1059, DOI: 10.1080/01621459.1990.10474975

Xiong, L. S., Chen, M. H., Chen, H. X., Xu, A. G., Wang, W. A., & Hu, P. J. (2004). A population‐based epidemiologic study of irritable bowel syndrome in South China: stratified randomized study by cluster sampling. Alimentary pharmacology & therapeutics , 19 (11), 1217-1224.

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

Enhancing the sample diversity of snowball samples: Recommendations from a research project on anti-dam movements in Southeast Asia

Contributed equally to this work with: Julian Kirchherr, Katrina Charles

Roles Conceptualization, Investigation, Visualization, Writing – original draft

* E-mail: [email protected]

Affiliations Faculty of Geosciences, Utrecht University, Utrecht, the Netherlands, School of Geography and the Environment, University of Oxford, Oxford, United Kingdom

Roles Supervision, Writing – review & editing

Affiliation School of Geography and the Environment, University of Oxford, Oxford, United Kingdom

• Julian Kirchherr, 
• Katrina Charles

• Published: August 22, 2018
• https://doi.org/10.1371/journal.pone.0201710
• Reader Comments

Snowball sampling is a commonly employed sampling method in qualitative research; however, the diversity of samples generated via this method has repeatedly been questioned. Scholars have posited several anecdotally based recommendations for enhancing the diversity of snowball samples. In this study, we performed the first quantitative, medium- N analysis of snowball sampling to identify pathways to sample diversity, analysing 211 reach-outs conducted via snowball sampling, resulting in 81 interviews; these interviews were administered between April and August 2015 for a research project on anti-dam movements in Southeast Asia. Based upon this analysis, we were able to refine and enhance the previous recommendations (e.g., showcasing novel evidence on the value of multiple seeds or face-to-face interviews). This paper may thus be of particular interest to scholars employing or intending to employ snowball sampling.

Citation: Kirchherr J, Charles K (2018) Enhancing the sample diversity of snowball samples: Recommendations from a research project on anti-dam movements in Southeast Asia. PLoS ONE 13(8): e0201710. https://doi.org/10.1371/journal.pone.0201710

Editor: Timothy C. Guetterman, University of Michigan, UNITED STATES

Received: September 13, 2017; Accepted: July 20, 2018; Published: August 22, 2018

Copyright: © 2018 Kirchherr, Charles. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All data underlying the study are within the paper and its Supporting Information files.

Funding: The authors received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Snowball sampling is a commonly employed sampling method in qualitative research, used in medical science and in various social sciences, including sociology, political science, anthropology and human geography [ 1 – 3 ]. As is typical of terms adopted by a variety of fields, however, the phrase ‘snowball sampling’ is used inconsistently across disciplines [ 4 ]. The most frequently employed definition, suggested by Patton [ 5 ], Atkinson and Flint [ 6 ], Cohen and Arieli [ 7 ] and Bhattacherjee [ 8 ], is as a sampling method in which one interviewee gives the researcher the name of at least one more potential interviewee. That interviewee, in turn, provides the name of at least one more potential interviewee, and so on, with the sample growing like a rolling snowball if more than one referral per interviewee is provided.

This definition can initially seem self-explanatory, which may explain why snowball sampling is rarely discussed in most peer-reviewed papers that employ it. Various scholars use snowball sampling in their empirical work, but most provide only limited information on the method (see, e.g., [ 9 – 13 ]). Similarly, qualitative research textbooks often lack substantive discussion of snowball sampling (e.g., [ 8 , 14 – 19 ]). Bailey [ 14 ], for instance, devotes only a half-page of his 595-page book on social research methods to snowball sampling, acknowledging that ‘snowball sampling procedures have been rather loosely codified’ ([ 14 ], p. 96), an observation echoed by Penrod et al. [ 3 ].

This paper focuses on snowball sampling procedures, which we define as those actions undertaken to initiate, progress and terminate the snowball sample [ 1 , 20 ]. Despite the lack of substantive writing on snowball sampling as a method, several authors [ 2 , 3 , 21 ] have provided recommendations for enhancing a sample’s diversity in snowball sampling procedures (we discuss this further in Section 4). However, as this advice is not based on a quantitative analysis of evidence, but only on anecdotal evidence, there is a risk that these recommendations are based on coincidence. The aim of this paper is to provide advice on enhancing the sample diversity of a snowball sample. This advice is grounded in a medium- N analysis of relevant evidence, thus reducing the probability of positing advice that is based on coincidence [ 22 ]. A medium- N analysis is generally based on 10–100 cases, whereas anecdotal evidence is usually based only on a handful of cases [ 23 , 24 ]. At the core of our work, we provide descriptive analyses of various commonly prescribed strategies for enhancing the sample diversity of a snowball sample. These analyses are based on reach-outs to 211 individuals via snowball sampling for a research project on anti-dam movements in Southeast Asia, resulting in 81 interviews conducted between April and August 2015. As far as we are aware, ours is the first medium- N analysis to focus on enhancing the sample diversity of a snowball sample.

The remainder of this paper is organised as follows: in Section 2, we discuss snowball sampling as a method; in Section 3, we present the research project on anti-dam movements in Southeast Asia that served as the basis for our medium- N analysis on snowball sampling procedures; in Section 4, we present and discuss insights on snowball sampling procedures based upon this analysis as well as our resulting recommendations; finally, in Section 5, we summarise our argument.

Throughout this paper, we employ social science methodology terminology. We define key terms for this paper such as ‘snowball sampling’ or ‘sampling’, since these terms are not consistently codified in the scholarly literature. Due to limited space, however, we refrain from defining terms we have deemed common in this field of study, referring only to the relevant literature.

On snowball sampling

Traditional sampling methods are comprised of two elements [ 25 , 26 ]. First, a full set of data sources is defined, creating a list of the members of the population to be studied, known as a sampling frame. Second, a specific sample of data is collected from this sampling frame. Snowball sampling defies both elements, since it does not rely upon a sampling frame [ 27 ] (which may indicate that a different term for snowball sampling would be more accurate). Snowball sampling is often employed when no sampling frame can be constructed.

Researchers frequently cannot construct a sampling frame if a difficult-to-reach population is to be studied. Difficult-to-reach-populations are also referred to as ‘hard-to-reach-populations’ [ 28 ], ‘hidden populations’ [ 29 ] or ‘concealed populations’ [ 21 ] in the scholarly literature. Although not all scholars may agree that these terms are interchangeable, we deem them interchangeable for the purposes of this paper. For further discussion of this terminology, see [ 30 , 31 ].

A difficult-to-reach population does not wish to be found or contacted (e.g., illegal drug users, illegal migrants, prostitutes or homeless people [ 6 , 31 ]). Snowball sampling was originally used by researchers to study the structure of social networks [ 32 ]. The earliest empirical account of snowball sampling is from 1955 [ 33 ], with snowball sampling first described as a method in 1958 [ 34 ]. While it is still used to study the structure of social networks [ 35 ], over the last few decades, the method’s key purpose has largely transformed ‘into […] an expedient for locating members of a [difficult-to-reach] population’ ([ 36 ], p. 141).

Researchers grounded in quantitative thinking, such as Lijphart [ 37 ] and King et al. [ 38 ], tend to view the drawing of a random sample from a sampling frame as the gold standard of data collection. Even these researchers may nevertheless consider non-probability sampling methods, such as snowball sampling, a ‘necessary and irreplaceable sampling [method]’ ([ 39 ], p. 367) when confronted with difficult-to-reach populations, particularly if the dismissal of snowball sampling would mean that no research could be conducted at all. Ultimately, ‘an important topic is worth studying even if very little [access to] information is available’ ([ 38 ], p. 6). Still, some of those grounded in quantitative thinking call snowball sampling a method ‘at the margin of research practice’ ([ 6 ], p. 1), since the lack of a sampling frame means that, unlike individuals in a random sample, individuals in a population of interest do not have the same probability of being included in the final sample. Findings from a snowball sample would therefore not be generalisable [ 40 ] (on generalisability, see [ 41 ]).

Several qualitative scholars rebut such criticism. Creswell, for instance, notes that ‘the intent [of qualitative research] is not to generalise to a population, but to develop an in-depth [and contextualised] exploration of a central phenomenon’ ([ 42 ], p. 203). Others [ 1 , 39 ] specifically oppose quantitative scholars’ negative framing of snowball sampling, arguing that this method would ‘generate a unique type of social knowledge’ ([ 1 ], p. 327). Due to the diversity of perspectives gathered, this knowledge would be particularly valuable for an in-depth and contextualised exploration of a central phenomenon. We therefore define the diversity of a sample as a measure of the range of viewpoints that have been gathered on a central phenomenon.

Researchers critical of snowball sampling respond to this defence by arguing that the method is unable to ensure sample diversity, which is a necessary condition for valid research findings. Indeed, some scholars have stated that snowball samples underrepresent and may even exclude those least keen to cooperate, since referrals may not materialise in an interview if a potential interviewee is only somewhat keen or not at all keen to be interviewed [ 3 , 43 ]. Similarly, potential interviewees with smaller networks may be underrepresented, as they are less likely to be referred for an interview [ 31 , 44 ]. Those with smaller networks may also be in a specific network whose different perspectives may be of interest but are excluded in the final sample. Meanwhile, snowball sampling is said to over represent those interviewees (and their respective networks) that the interviewer spoke with first; the relevant literature refers to this as ‘anchoring’ [ 20 , 39 ].

We do not aim to argue the ‘validity’ of the method, but rather to inform snowball sampling methodologies in order to promote sample diversity. From a qualitative perspective, ‘validity’ can be defined as ‘the correctness or credibility of a description, conclusion, explanation, interpretation or other sort of account’ ([ 45 ], p. 87), while quantitative researchers frequently use the terms ‘generalisability’ and ‘(external) validity’ interchangeably [ 46 , 47 ]. The term ‘validity’ is contested among qualitative researchers, and some qualitative researchers entirely reject the concept for qualitative work [ 48 , 49 ]. We do not aim to resolve this debate via this paper; instead, we focus on the (seemingly less-contested) term ‘sample diversity’. While we acknowledge that this term is not codified in qualitative textbooks such as the SAGE Encyclopedia of Qualitative Research Methods , sample diversity is considered desirable by the various qualitative scholars we reviewed. Boulton and Fitzpatrick demand, for instance, that qualitative researchers ‘ensure that the full diversity of individuals […] is included [in their sample]’ ([ 50 ], p. 84), a mandate echoed by other scholars [ 16 , 51 – 53 ].

In order to operationalise the concept of sample diversity, we used five key methodological recommendations to inform our research. In this paper, we use quantitative analyses from our experiences with snowball sampling to further reflect on these recommendations, which are briefly described below.

Prior personal contacts of the researcher are required

Patton ([ 5 ], p. 176) notes that snowball sampling ‘begins by asking well-situated people: “Who knows a lot about ____? Who should I talk to?”‘. In the absence of a sampling frame for the population of interest, however, the researcher must retain at least some prior personal or professional contacts in the population of interest which can serve as the seeds of the snowball sample [ 2 , 54 ]. Waters contends that building a diverse snowball sample ‘depend[s] almost exclusively on the researcher’s [prior personal or professional] contacts’ ([ 39 ], p. 372).

Sample seed diversity is important

Morgan [ 21 ] has claimed that the ‘best defence’ against a lack of sample diversity is to begin the sample with seeds that are as diverse as possible. Others echo this advice [ 3 , 39 , 55 ], arguing that it is ‘compulsory for the researcher to ensure that the initial set of respondents is sufficiently varied’ ([ 55 ], p. 55). The term ‘chain referral sampling’ has been used for snowball samples that are strategically built via multiple varying seeds [ 3 ].

Technology means face-to-face interviews are no longer required

Some researchers have argued that face-to-face interviews are obsolete. For instance, over 25 years ago, it was claimed there were ‘no remarkable differences’ ([ 56 ], p. 211) between information collected via telephone and information collected via face-to-face interviews. The increasing use of telecommunications in recent years is likely to have further reduced barriers to remote interviewing, and various scholars [ 57 , 58 ] continue to claim that ‘evidence is lacking that [telephone interviews] produce lower quality data’ ([ 59 ], p. 391). In particular, they have highlighted the benefits of using Skype for semi-structured interviews [ 57 ].

However, for snowball sampling, face-to-face interviews help to generate the trust that scholars claim is required in order to gain referrals [ 1 , 31 , 39 , 60 ]. Noy argues that ‘the quality of the referring process is naturally related to the quality of the interaction: […] if the researcher did not win the informant’s trust […], the chances the latter will supply the former referrals decrease’ ([ 1 ], p. 334).

Persistence is necessary to secure interviews

Although the value of persistence may be considered self-evident by some scholars, it is seen by multiple academics [ 61 – 63 ] as a central virtue of qualitative researchers. Many young career scholars who embrace snowball sampling are likely to hear such advice as, ‘If you cannot interview your envisaged interviewees initially, don’t give up!’. A ‘helpful hint’ for qualitative researchers seeking informants is, ‘Persevere–repeat contact’ [ 64 ].

More waves of sampling are required to access more reluctant interviewees

As a remedy for snowball sampling’s previously discussed bias towards excluding those least keen to be interviewed, multiple scholars suggest pursuing a snowball sample for multiple waves (with a new sampling wave reached once an interviewee introduces the interviewer to one or more potential interviewees) [ 65 – 68 ]. Those suggesting this remedy assume that pursuing more waves increases the likelihood of being referred to an interviewee from a particularly difficult-to-reach population who is at least somewhat keen to be interviewed.

Approval for this study was granted by the Central University Research Ethics Committee (CUREC) of the University of Oxford. Our population of interest for our research project were stakeholders in Southeast Asia’s dam industry. Since ‘the most dramatic conflicts over how to pursue sustainable development’ ([ 69 ], p. 83) have occurred over the construction of large dams, we see this industry as a conflict environment with widely varying viewpoints. A conflict environment is one in which people perceive their goals and interests to be contradicted by the goals or interests of the opposing side [ 70 ]. The major conflicting parties in the dam industry tend to be local and international non-governmental organisations (NGOs) and academics (usually keen not to construct a particular dam) versus international donors, the private sector and governments (usually keen to construct a particular dam) [ 71 , 72 ]. Each sub-population operating in a conflict environment can be considered difficult to reach since fear and mistrust are often pervasive [ 7 ]. Snowball sampling is a suitable research method in conflict environments because the introductions through trusted social networks that are at the core of this method can help interviewees to overcome fear and mistrust, which, in turn, ensures access [ 7 ]. This access is needed to gather the widely varying viewpoints in the hydropower industry, in particular viewpoints with regards to what constitutes just resettlement [ 73 , 74 ]. Based on this rationale, we chose snowball sampling as the main method for our research.

In order to ensure sample diversity for our research project on anti-dam movements in Southeast Asia, we aimed to gather perspectives mostly from six main sub-populations: (1) local NGOs, (2) international NGOs, (3) international donors, (4) academia, (5) the private sector and (6) the government. We hypothesized that ‘dam developers’, a main sub-category of the interviewee category ‘private sector’, would be the most significant challenge to ensuring the diversity of our sample. Early in our process, many of the scholars with whom we discussed our research project argued that it would be impossible to interview a dam developer from a Chinese institution; meanwhile, researchers from a comparable research project that ended approximately when our project started reported being unable to interview any dam developers from European institutions. We also initially failed to collect data from dam developers: for instance, a survey we initiated that was distributed by Aqua~Media (host of a major global dam developer conference) to more than 1,500 dam developers yielded just five responses, only one of which was complete. We considered this weak response rate to be due, at least in part, to the dam industry’s negative view of academicians since the publication of Ansar et al. [ 75 ], which Nombre ([ 76 ], p. 1), the president of the International Commission on Large Dams (ICOLD), called ‘[highly] misleading’.

None of our researchers had significant direct links to the dam industry upon the start of the project; however, we did retain a variety of indirect links. Our researchers had past links to a management consultancy that serves various dam industry players, (more limited) links to an international donor working in the hydropower sector and links to activists in Myanmar advocating against dam projects.

After a favourable ethics review of our study by the CUREC of the University of Oxford, we commenced semi-structured interviews in April 2015, mostly via cold calls (we include cold e-mails in the term ‘cold calls’ throughout this paper). Initially, we conducted research via telephone only. We then undertook field research in Singapore, Myanmar and Thailand from June to August 2015 and terminated our data collection in late August 2015.

In total, 81 semi-structured interviews were carried out during this period. From a qualitative perspective, this is a relatively large sample size (for instance, the average qualitative PhD dissertation is based on 31 interviews [ 77 ]); from a quantitative perspective, however, the sample size is quite small [ 78 ]. Of our 81 interviews, 48 (59%) were conducted via telephone, 26 (32%) face-to-face and 7 (9%) online, either via e-mail or an online survey. Most of our interviews (57%) were carried out in July in Myanmar. Of our 81 interviewees, only 24 (30%) were women. Researchers who employ snowball sampling frequently employ personal/professional contact seeds and cold call seeds to build their sample (e.g., [ 2 , 79 , 80 ] with a seed defined as the starting point of a sample [ 65 ]). Of the 81 interviews analysed, 53 (65%) were rooted in a personal or professional contact ( Fig 1 ) (i.e. the seed of the interview pathway was a contact we had already retained prior to the research project). The remaining 28 (35%) interviews were rooted in cold calls.

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

https://doi.org/10.1371/journal.pone.0201710.g001

Given the sensitive nature of the interview topic, all interviewees were assured anonymity. Thus, all of the interviews are coded, with the first letter indicating the mode of interview ( T for telephone, F for face-to-face, O for online survey or e-mail), the second letter indicating the category of interviewee ( A for academia, G for government, I for international donor, NI for international NGO, NL for national NGO, P for private sector) and the sequence of numbers indicating the interview number within a particular mode. Researcher A is indicated by RA , Researcher B by RB ; CON represents a conference event. Bold type indicates that an interview was completed, while X that an interview was not completed.

As outlined in the previous section, snowball sampling is sometimes criticised for producing samples that lack sample diversity. To address this criticism, we reviewed the (scarce) literature on enhancing sample diversity via snowball sampling procedures prior to commencing our study. Upon reflection during our research, we chose to pursue our analysis retrospectively in order to challenge some of the recommendations provided in literature. Our analysis is structured alongside the five core pieces of advice found in this literature ( Table 1 ). Our results are based on a quantitative analysis of the 81 interviews we conducted. Although we endeavoured to include all interview attempts, some initial cold calls may have been overlooked in this retrospective approach. Therefore, some of our analysis, particularly in Section 4.4, may be too optimistic. Overall, we were able reconstruct 211 reach-out attempts.

Sample diversity is measured by representation from five identified sub-groups.

https://doi.org/10.1371/journal.pone.0201710.t001

Results and discussion

On prior personal and professional contacts.

Our analysis provides evidence that sample diversity can be reached even if no prior personal or professional contacts to the population of interest have been retained. The seeds of the interviews are depicted in Fig 2 , with the left side of the figure depicting the 53 interviews based on a personal or professional contact and the right side depicting the 28 interviews that were based on cold calls. This figure shows two main points of interest: first, both types of seeds include interviews in each interview category; second, the interview sub-category ‘dam developer’, which we hypothesised would be the most difficult to include in the sample, is also covered by both types of seeds. We can therefore conclude that a diverse sample could have been built even if we had relied solely on cold calls.

https://doi.org/10.1371/journal.pone.0201710.g002

It is acknowledged, however, that building a snowball sample from cold calls is particularly labour-intensive [ 39 ]: in our research, only 25% of our cold calls led to an interview, compared to 62% of the referrals. Significant differences in the value of referrals persist from one interviewee group to another ( Fig 3 ). We measure the value of referrals via a concept we call ‘network premium’. To gauge the network premium, we subtracted the cold call response rate (i.e., the number of interviews initiated via cold calls divided by the total number of cold calls) from the referral response rate (i.e. the number of interviews initiated via referrals divided by the total number of referrals). Referrals were the most valuable when contacting international donors and private sector players, with network premiums of 74% and 52%, respectively, indicating that these groups are particularly difficult-to-reach populations.

(1) Unable to retrace for 13 identified reach-outs if initiated via referral or cold call; four reach-outs coded as ‘Other’. (2) Unable to retrace for one interview carried out via referral coded as ‘Other’. (3) Including personal contacts and contacts via conferences. (4) Referral response rate–Cold call response rate.

https://doi.org/10.1371/journal.pone.0201710.g003

The overall results from these analyses are encouraging for scholars interested in researching a population to which no personal or professional contacts are retained prior to the research project. While personal or professional contacts maintained to the research population of interest can accelerate the research endeavour, our results also showcase that (at least for our topic of interest) a diverse sample can be built from cold calls if a researcher is willing to invest some time in reach-outs.

On seed variation

Our research confirms the scholars’ advice that seed diversity is important. Fig 4 (a variation of Fig 2 ) depicts the completed interviews from a seed perspective, with RA, RB and cold calls as the three main seeds of the sample. The sample built via RA, who has a background in the private sector, is largely biased towards this sector, with 47% of all interviews seeded via RA private sector interviews. RB conducted 57% of interviews, whose background is closest to local NGOs, were with local NGOs. Meanwhile, the sample built via cold calls indicates no significant biases towards any interviewee category. Interviews based on the network of RB included one (TNL17) with a leading activist from a remote area of Myanmar who provided unique insights into the early days of an anti-dam campaign. This insight helped us to develop a narrative of the campaign that was not skewed to the later days of the campaign and the activists prominent in these later days. The sample diversity ensured via RB was thus central to the quality of our research.

https://doi.org/10.1371/journal.pone.0201710.g004

It is noteworthy that the three different seeds in Fig 4 include interviews in all interviewee categories, including the sub-category ‘dam developer’ (the sole exception is the interviewee category ‘international NGO, which contains zero interviews for RB). This indicates that, at least for our topic of interest, a fairly diverse sample can be generated even if the researcher is unable to vary her or his seed, although the overall data suggest that seed variation can significantly enhance sample diversity. Fig 3 may therefore be viewed as a case for collaboration among researchers; if researchers with different backgrounds and different personal and professional contacts to the population of interest begin to collaborate, such collaborations are bound to contribute to sample diversity.

On face-to-face interviews

Our descriptive analysis provides evidence to further support the argument that face-to-face interviews are redundant, with our data indicating that face-to-face interviews can lead to more sought referrals than telephone interviews (perhaps since trust may be more readily established via face-to-face conversations than over the telephone). Fig 5 aims to quantify the value of face-to-face interviews. Overall, 30 (37%) of our interviews were initiated via prior face-to-face conversations, while prior telephone conversations and online contact each led to only eight interviews (10%). An examination shows that of the nine interviews conducted with dam developers, the interviewee sub-category deemed most difficult to access, seven (78%) were initiated via prior face-to-face interviews, while not a single telephone interview led to a referral to a dam developer. These interviews proved to be essential for our research. For instance, one Chinese dam developer challenged a claim from numerous NGOs that his company would not engage with NGOs, which, in turn, allowed us to present a more balanced portrayal of the interplay between Chinese dam developers and NGOs.

(1) Comprises interviews with those already retaining a personal or professional contact prior to the research project.

https://doi.org/10.1371/journal.pone.0201710.g005

While our research did not investigate whether face-to-face interviews lead to lower-quality data than telephone interviews, our data provide tentative evidence that face-to-face interviews are not obsolete; they can still be helpful for those employing or intending to employ snowball sampling, since these interviews can lead to more sought referrals and thus enhanced sample diversity. We acknowledge, however, that this finding may not be true for all populations. For instance, studies on individuals with sexually transmitted diseases have found that these interviewees (particularly men) tend to report more truthfully in an audio-computer-assisted self-interview (ACASI) than in a face-to-face interview, since interviewees tend to be more comfortable reporting on sexually transmitted diseases to a computer than to a live person [ 81 , 82 ].

On persistence

Our data suggest that persistence can indeed enhance sample diversity, but we can also conclude that excessive persistence does not necessarily yield dividends. Instead of distributing a great many interview reminders during our study, we reached out to the majority of our proposed interview subjects only once. Nevertheless, the scarce data we collected regarding persistence indicates its value. We map this data in Fig 6 , with the left side depicting our success rate in relation to the number of reach-outs (either one, two or three) and the right side depicting a deep dive on success rates achieved with two reach-outs (distinguishing between reach-out attempts to unknown potential interviewees and those to whom we were referred by other interviewees). We sent one interview reminder to 28 of our proposed interviewees. This led to 10 additional interviews, a success rate of 36%, equalling 12% of the total interviews analysed for this paper. Reminders appear to be only somewhat more helpful when contacting referrals in comparison to their usefulness with cold calls–a single reminder led to an interview in 39% of our cases for the former group and 38% for the latter. One of the most valuable interviews for our research gained via a reminder was with the CEO of a Burmese dam developer. This interviewee compared Chinese and European dam developers in Myanmar, which helped us to further refine our narrative on social-safeguard policy adherence by Chinese dam developers in Myanmar.

(1) Number of reach-outs unknown for 32 reach-outs. Eight potential interviewees responded, but refused interview.

https://doi.org/10.1371/journal.pone.0201710.g006

Excessive persistence, however, does not appear to be worthwhile. We sent three reminders to seven of our envisaged interviewees, but as Fig 6 shows, this did not lead to a single additional interview. While our data does not suggest that excessive persistence is helpful to researchers, it may also not be recommended for ethical reasons. A potential interviewee who does not respond to an interview request after two reach-outs may be indicating via this non-response that she or he is not interested in participating in the research. If a single request remains unanswered, the researcher may hypothesise that, for instance, the e-mail was overlooked, a hypothesis particularly likely when conducting interviews with time-pressed leaders of organisations. Indeed, all 10 interviews only carried out upon the second reach-out were interviews with interviewees in management positions.

Our data on persistence provide some evidence that those employing or intending to employ snowball sampling can enhance sample diversity if every reach-out is carefully tracked and followed by a reminder. We typically sent a reminder after one week if no response was obtained upon the first reach-out. This persistence may help to include those least keen to be interviewed for a research endeavour.

Our data show some evidence that, for our topic of study, pursuing interviews for even a few waves provided the perspectives of particularly difficult-to-reach populations and thus achieved sample diversity. More than 60% of our interviews were conducted in the zeroth or first wave ( Fig 7 ). These include seven of the nine interviews conducted with dam developers, the sub-category we deemed most challenging to interview. The remaining two interviews with dam developers were conducted in the second wave. However, not a single interview with a dam developer was carried out in the third wave and beyond, although a fifth of our total interviews were carried out in the third or later waves. Pursuing interviews for multiple waves nevertheless yielded novel insights. For instance, interview FNL12, which was conducted in the sixth wave, yielded insights on small dam construction in Myanmar–a topic of (some) interest to our research endeavour, but not covered in detail by previous interviews. Furthermore, we note that our finding regarding the limited value of multiple waves may also be specific to our population, with this finding perhaps indicating a low degree of network segmentation in the population in question [ 83 ]. Meanwhile, a high degree of network segmentation may impede the pursuance of multiple waves, since interviewees may lack the suitable contacts for a referral [ 84 ].

https://doi.org/10.1371/journal.pone.0201710.g007

While additional waves can lead to novel insights, our overall data on waves provide some evidence that the number of waves pursued is not a definitive indicator for sample diversity. Even very few waves can yield access to particularly difficult-to-access populations.

Our quantitative analysis of pathways to delivering sample diversity in snowball samples yielded the following revisions to the literature’s recommendations:

• Prior personal contacts are not essential for achieving sample diversity but tend to be helpful, as generating new contacts during research can be labour-intensive.
• Sample seed diversity is important to achieving sample diversity.
• Face-to-face interviews build trust and can help to generate further referrals.
• Persistence (within reason) is helpful in securing interviews.
• Sample diversity is not necessarily enhanced if a seed is advanced over numerous waves.

We do not claim that these insights are comprehensive, but we believe that these interpretations of our data may serve as a starting point for future scholars using snowball sampling procedures. All of the analyses presented in this section are based only on descriptive statistics. This means, for instance, that we cannot control for confounds such as effort [ 85 ]. An experimental research design would yield the most robust insights on sampling procedures to enhance the sampling diversity of a snowball sample (with, for instance, one research project staffed with scholars with relevant personal or professional contacts and another staffed with scholars without relevant contacts).

Overall, this work aims to advance the literature on snowball sampling as a qualitative sampling approach. While snowball sampling procedures may qualify ‘as the least “sexy” facet of qualitative research’ ([ 1 ], p. 328), these procedures are ‘not self-evident or obvious’ ([ 20 ], p. 141), since the snowball sample does not ‘somehow magically’ ([ 20 ], p. 143) start, proceed and terminate when a scholar attempts to develop a diverse sample. Rather, continuous, deliberate effort by the researcher(s) is required. Our paper has attempted to provide some insights on this effort.

Unfortunately, we developed the idea to write this paper only during the course of our research project, and thus some of our data may be skewed. For instance, we may not have been able to trace all original reach-out attempts and our data on persistence may therefore be biased. Some of those scholars grounded in quantitative thinking may also claim that the insights outlined in Section 4 lack external validity since our sample size is relatively small from a quantitative methodological perspective. In addition, our population was very specific and thus may not be comparable to other difficult-to-reach populations, and we also did not adopt an experimental research design as described above. Hence, we encourage scholars to replicate our findings via their respective research projects that employ snowball sampling. With many scholars claiming to feel more pressed than ever to deliver research results with maximum efficiency, we hope that these initial descriptive analyses of snowball sampling procedures provide some valuable insights to those employing or intending to employ this method and aiming to improve their management of it.

Supporting information

https://doi.org/10.1371/journal.pone.0201710.s001

Acknowledgments

We wish to thank our reviewers at PLOS ONE who provided constructive thoughts on this piece of work. We also thank Ralf van Santen for his outstanding contributions to this work as a research assistant.

• View Article
• Google Scholar
• 4. Handcock MS, Gile KJ. On the Concept of Snowball Sampling. 2011 Aug;
• 5. Patton M. Qualitative evaluation and research methods. Beverly Hills: SAGE Publications; 1990.
• 6. Atkinson R, Flint J. Accessing Hidden and Hard-to-Reach Populations: Snowball Research Strategies. 2001.
• 8. Bhattacherjee A. Social Science Research: Principles, Methods, and Practices. 2012.
• 14. Bailey K. Methods of Social Research. New York, United States: The Free Press; 1994.
• 15. Miles MB, Huberman M a. An Expanded Sourcebook: Qualitative Data Analysis. 2nd ed. SAGE Publications; 1994.
• 16. Boeije HR. Analysis in Qualitative Research. Thousand Oaks, United States: SAGE Publications; 2009.
• 17. Atkinson P, Delamont S. SAGE qualitative research methods. Sage; 2010.
• 18. Daniel J. Sampling essentials: Practical guidelines for making sampling choices. SAGE Publications; 2011.
• 19. Emmel N. Sampling and choosing cases in qualitative research: A realist approach. Sage; 2013.
• 21. Morgan DL. Snowball sampling. In: The SAGE Encyclopedia of Qualitative Research Methods. Thousand Oaks, United States: SAGE Publications; 2008.
• 24. Ragin CC. Fuzzy-Set Social Science. Chicago, United States: The University of Chicago Press; 2000.
• 25. Morgan DL. Sampling. In: The SAGE Encyclopedia of Qualitative Research Methods. Thousand Oaks, United States: SAGE Publications; 2008.
• 26. Morgan DL. Sampling frame. In: The SAGE Encyclopedia of Qualitative Research Methods. Thousand Oaks, United States: SAGE Publications; 2008.
• PubMed/NCBI
• 33. Whyte WF. Street Corner Society—The Social Structure of an Italian Slum. Chicago, United States: Chicago University Press; 1955.
• 38. King G, Keohane RO, Verba S. Designing social inquiry: scientific inference in qualitative research. Princeton, United States: Princeton University Press; 1994.
• 40. Morgan DL. Random Sampling. In: The SAGE Encyclopedia of Qualitative Research Methods. Thousand Oaks, United States: SAGE Publications; 2008.
• 41. Donmoyer R. Generalizability. In: The SAGE Encyclopedia of Qualitative Research Methods. Thousand Oaks: SAGE Publications; 2008.
• 42. Creswell JW. Educational Research. Planning, Conducting, and Evaluating Quantitative and Qualitative Research. Upper Sadle River, United States: Pearson Education; 2005.
• 44. Daniel J. Choosing the Type of Nonprobability Sampling. In: Sampling Essentials: Practical Guidelines for Making Sampling Choices. Thousand Oaks, United States: SAGE Publications; 2012.
• 45. Maxwell JA. Qualitative Research Design: An Interactive Approach. Thousand Oaks, United States: SAGE Publications; 1996.
• 47. Shadish WR, Cook TD, Campbell DT. Experimental and quasi-experimental designs for generalized causal inference. Boston, United States: Wadsworth Cengage Learning; 2002.
• 51. Olsen W. Triangulation in Social Research: Qualitative and Quantitative Methods Can Really Be Mixed. In: Developments in Sociology. Causeway Press; 2004.
• 54. Berg S. Snowball sampling. In: Encyclopedia of Statistical World. Thousand Oaks, United States: SAGE Publications; 1988. p. 529–32.
• 62. Harrison RT, Leitch CM. Fighting a rearguard action? Reflections on the philosophy and practice of qualitative research in entrepreneurship. In: Handbook of Research Methods and Applications in Entrepreneurship and Small Business. Cheltenham: Edward Elgar Publishing Ltd.; 2014.
• 64. Feldman MS, Bell J, Berger MT. Gaining access: A practical and theoretical guide for qualitative researchers. Rowman Altamira; 2004.
• 66. Heckathorn DD, Magnani R. Snowball and Respondent-Driven Sampling. In: Behavioral Surveillance Surveys: Guidelines for Repeated Behavioral Surveys in Populations at Risk of HIV. 2004.
• 68. Robins G. Doing Social Network Research: Network-Based Research Design for Social Scientists. London, United Kingdom: SAGE Publications; 2015.
• 69. Khagram S. Toward Democratic Governance for Sustainable Development: Transnational Civil Society Organizing Around Big Dams. In: Third Force : The Rise of Transnational Civil Society. Carnegie Endowment for International Peace; 2000.
• 70. Kriesberg L. Constructive Conflict: From Escalation to Resolution. Lanham, United States: Rowman & Littlefield Publishers; 1998.
• 71. Nüsser M. Political Ecology of Large Dams: a Critical Review. 2003.
• 72. Nüsser M, Baghel R. The emergence of technological hydroscapes in the Anthropocene: socio- hydrology and development paradigms of large dams. In: Handbook on Geographies of Technology,. Cheltenham, United Kingdom: Edward Elgar Publishing; 2017. p. 287–301.
• 73. Scudder T. The Future of Large Dams: Dealing with Social, Environmental, Institutional and Political Costs. London, United Kingdom: Earthscan; 2005.
• 76. Nombre A. Yes, we need to build more large dams for water storage and energy for sustainable development! 2014.
• 77. Mason M. Sample Size and Saturation in PhD Studies Using Qualitative Interviews. Vol. 11, Forum Qualitative Sozialforschung / Forum: Qualitative Social Research. 2010.

• Skip to main content [s]
• Infonet (For Students and Staff)
•    
• Research Priority Areas
• Research Degrees
• Library Research Resources

Snowball Sampling

Snowball sampling is one of the most popular methods of sampling in qualitative research, central to which are the characteristics of networking and referral. The researchers usually start with a small number of initial contacts (seeds), who fit the research criteria and are invited to become participants within the research. The agreeable participants are then asked to recommend other contacts who fit the research criteria and who potentially might also be willing participants, who then in turn recommend other potential participants, and so on. Researchers, therefore, use their social networks to establish initial links, with sampling momentum developing from these, capturing an increasing chain of participants. Sampling usually finishes once either a target sample size or saturation point has been reached. This entry begins with a description of the conveniences of snowball sampling, followed by some criticisms and limitations of the technique. The next section provides examples of how snowball sampling is used in qualitative research projects. Subsequent sections examine instances in which snowball sampling stalls or fails to produce participants, and offers two examples of cases in which researchers successfully overcame those obstacles. The entry concludes with a look at some variants of snowball sampling that have emerged given technological advances.

University Staff: Request a correction | Repository Editors: Update this record

• About the Repository
• Repository Policies
• Open Access Policies
• Statistics Overview
• Accessibility

University Of Gloucestershire

Find Us On Social Media:

Other University Web Sites

• Staffnet (Staff Only)
• © UoG 2008-18
• Privacy and Cookies
• Comments concerning this page to Webmaster

Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to  upgrade your browser .

Enter the email address you signed up with and we'll email you a reset link.

• We're Hiring!
• Help Center

Snowball Sampling: A Purposeful Method of Sampling in Qualitative Research

2017, Strides in Development of Medical Education

Related Papers

Journal of Occupational and Organizational Psychology

Adrian Furnham

Middle Eastern Journal of Disability Studies

esmaeil sharifian

mohammad rostaminezhad

Iranian educational systems heavily focus on Smart schools and equipping. Based on social-cognitive theory student attitude toward school affect student achievement and motivation. With attention to the importance of student attitude toward technology and schools, this study aims to develop and validate a questionnaire in order to estimate student attitude toward Smart schools. A sample 305 middle school student from grade 7 and 8 in South Khorasan was selected randomly. Exploratory factor analysis identified 9 factors that explain 62 present of the variance of student attitude toward smart schools. These factors include: attitude towards interactive whiteboard; attitude towards teacher; reluctance to classroom; general attitude towards smart schools; test and feedback; feeling alone in school; student satisfaction from teaching; the impression of school and finally family attitude. The alpha value revealed instrument is reliable (α=0. 89). All Instrument factors can predict 22.5 pe...

Iranian Journal of Medical Microbiology

farshid Danesh

Military Caring Sciences

naser lotfi fatemi

Mahdi Mohammadi

Abstract Objectives: Considering the importance and functional role of social networks in libraries, the present research intends to examine the use of types of social networks by the libraries of Makta member of Qom province. Method: The present research is an applied research in terms of the purpose of the research. A descriptive approach has been used to survey this research. The statistical communty of this study consists of 30 libraries of 34 members of the Makta member libraries. A researcher-made questionnaire was distributed among the directors of the Makta member libraries in order to collect data. Because of the limited research community, all of them were studied and sampling was not done. Results: Findings showed that libraries are The most use social networks in social media. Working with other libraries and training the second and third ranks. In library lending, social networks are also used to reserve the resources requested in use a littlebit. The findings also showed that web-mobile computing tools were the most with 25.7% and microblocking tools with the lowest use rate of 4.2% in libraries. Conclusion: The results of the research show that only 40% of the member libraries of Makta use social networks to manage their library affairs. The most important reason nonuse libraries from social networks was "lack of proper infrastructure". Keywords (s): Special Libraries of Qom Province, Social Networks, Makta, Qom Province

Ahmad Foroughi

Author Address 1. Department of Psychology branch, Islamic Azad University, Isfahan (Khorasgan) branch, Isfahan, Iran; 2. Associate professor Department of statistics, University of Isfahan, Isfahan, Iran; 3. Associate professor of children with special needs, Department of children with special needs, University of Isfahan, Isfahan, Iran; 4. Associate professor, Islamic Azad University, Isfahan (Khorasgan) branch, Isfahan, Iran; 5. Professor, Islamic Azad University, Isfahan (Khorasgan) branch, Isfahan, Iran. *Corresponding Author Address Islamic Azad University, Isfahan (Khorasgan) branch, Isfahan, Iran. *Tel: 09137723258; *E-mail: [email protected]

hamid karyab

Mohammad Ali Bagherpour

تیمها هب هجوت اب نوزفازور نامزاس رد یلخاد یسرباسح یرلخاد یرسرباسح حرشن ررییغت و اه تراهم اه و یگتسیاش یاه دیدج طبترم اب نآ زارین دروم ترسا رر،اح حهوهرپ دره نیودرت ناریدم یگتسیاش عماج یوگلا یلخاد یسرباسح یگهیو هب هجوت اب مكاح طیارش و یطیحم یاه یسرباسح هفرح رب ناریا رد یلخاد یم یفارفتكا دررریور هرب هجوت اب دشاب حهوهرپ رر،اح زا شور یفیك هیرظن یزادرپ داینب هداد هدافتسا دیدرگ روظنم نیا هب قیرط زا هبحاصم راتخاس یاه اب هتفاین دیتارسا ارعف نارصصصتم و هاگرفناد هزورح رد یرلخاد یرسرباسح یرسررب هرب دارعبا یلخاد یسرباسح ناریدم یگتسیاش دش هتخادرپ و هداد یراررگدك شور رمك هب حهوهپ یاه هیزجت دروم یباصتنا و یروحم زاب تفرگ رارق لیلحتو هرب هرجوت ارب سرس هرلوشم ن،رك یاره وشم هل هلوشم درخ و یلصا یاه جارصتسا )میهافم( اه هدش یوگلا یگتسیاش یرلع طیاررش لماش هلخادم لماوع اهدربهار نینچمه و رتسب و رگ نآ یاهدمایپ و ارا اه ئ دیدرگ ه نیا ساسا رب ورگلا ناریدم یلخاد یسرباسح یرفصبرخا ثورلطم مطرس هرب یبایتسد روظنم هب هرب هرعومجم زا یا یگتسیاش یاه یدرف یتیریدم ینف و دنراد زاین یطیحم د هتفای نیا عقاو ر یامیرس رگناریب ار...

Nabi Nazari

Abstract: This cross-sectional study conducted to develop a model for predicting academic achievement of university students by investigating the relationship between teaching skills, academic emotions (positive and negative), and academic stress associated with Mindset (growth and fixed) using structural equation modeling. The statistical population consisted of 360 students of the Islamic Azad University of Hamedan who were selected randomly using a relative stratified method. The study was descriptive and correlational. The data were analyzed by SPSS version 25 and SmartPLS version 3.2.8. First, the validity of the model was estimated using Cronbach's alpha, composite reliability, convergent validity, and divergent validity; then, the coefficient of determination, effect size, and Stone-Geisser criterion were calculated for evaluating the structural model. The results showed that the validity and adequacy of the suggested model were suitable. Thus, it could be used in differe...

RELATED PAPERS

Stochastic Models

Pieter Allaart

Helio Vannucchi

Obstetrics & Gynecology

Barry Schifrin

Revista de Cultura Teológica. ISSN (impresso) 0104-0529 (eletrônico) 2317-4307

Rafael Lopez Villasenor

Mapula Makwela

Indian Journal of Thoracic and Cardiovascular Surgery

Rajnish Garg

Marija Bjekovic

Nature Biotechnology

Brittney Robinett

Agung Nugroho Catur Saputro

Fatih Uzunkaya

Jurnal teknologi pendidikan : jurnal penelitian dan pengembangan pembelajaran

Endah Resnandari Puji Astuti

Revista Tempo e Argumento

Miguel Angel Jara

Solid State Communications

Personnel Assessment and Decisions

Chase Winterberg

Computer Methods in Biomechanics and Biomedical Engineering

Charles A . Taylor

Applied Physics A

Armando Tejeda-Ochoa

Avances del Cesor

Agustin Nieto

Ruth Guzik Glantz

Dhairya Mohan

Journal of Non-Crystalline Solids

cristina Lozano Sanchez

International Series in Quantitative Marketing

Gilles LAURENT

Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment

Gevaldo Almeida

Peter Quayle

Interdisciplinary Issues in Education

Lekhnath Sharma

Annals of Surgery

Frederic Ris

RELATED TOPICS

•   We're Hiring!
•   Help Center
• Find new research papers in:
• Health Sciences
• Earth Sciences
• Cognitive Science
• Mathematics
• Computer Science
• Academia ©2024