## Snowball sampling

Snowball sampling is a type of non-probability sampling technique. Non-probability sampling focuses on sampling techniques that are based on the judgement of the researcher [see our article Non-probability sampling to learn more about non-probability sampling]. This article explains (a) what snowball sampling is, (b) how to create a snowball sample, and (c) the advantages and disadvantages (limitations) of snowball sampling.

#### Snowball sampling explained

Some populations that we are interested in studying can be hard-to-reach and/or hidden. These include populations such as drug addicts, homeless people, individuals with AIDS/HIV, prostitutes, and so forth. Such populations can be hard-to-reach and/or hidden because they exhibit some kind of social stigma, illicit or illegal behaviours, or other trait that makes them atypical and/or socially marginalized. Snowball sampling is a non-probability based sampling technique that can be used to gain access to such populations.

#### Creating a snowball sample

To create a snowball sample, there are two steps: (a) trying to identify one or more units in the desired population; and (b) using these units to find further units and so on until the sample size is met.

##### STEP ONE Try to identify one or more units in the desired population

Imagine that the population we are interested in are students that download pirate music over the Internet or that take drugs. Let's go with the latter: students that take drugs. Each student is referred to as a unit [see our article, Sampling: The basics, if you are unsure about the terms unit, case, object, sample and population]. Collectively, all student drug users make up our population. However, we are only interested in examining a sample of these student drug users.

First, we need to try and find one or more units from the population we are studying (i.e., student that take drugs). Finding just a small number of individuals willing to identify themselves and take part in the research may be quite difficult, so the aim is to start with just one or two students (i.e., one or two units).

##### STEP TWO Use these units to find further units and so on until the sample size is met

Due to the sensitivity of the study, the researcher should ask the initial students who agreed to take part in the research to help identify other students that may be willing to take part. For ethical reasons, these new research participants should come forward themselves rather than being identified by the initial students. In this respect, the initial students help to identify additional units that will make up our sample. The process continues until sufficient units have been identified to meet the desired sample size.

Snowball sampling is a useful choice of sampling strategy when the population you are interested in studying is hidden or hard-to-reach. This includes populations are such to social stigma and marginalization, such as suffers of AIDS/HIV, as well as individuals engaged in illicit or illegal activities, including prostitution and drug use. Snowball sampling is useful in such scenarios because:

• It can be difficult to identifying units to include in your sample, perhaps because there is no obvious list of the population you are interested in. For example, there are no lists of drug users or prostitutes that a researcher could get access to, especially lists that could be considered representative of the population of drug users or prostitutes.

• The sensitivity of coming forward to take part in research is more acute in such research contexts. Individuals that are drug users or prostitutes, for example, are likely to be less willing to identify themselves and take part in a piece of research than many other social groups. However, since snowball sampling involves individuals recruiting other individuals to take part in a piece of research, there may be common characteristics, traits and other social factors between those individuals that help to break down some of the natural barriers that prevent such individuals from taking part.

• The unknown and/or secretive nature of some social groups may also make it difficult to identify strata that warrant investigation. Strata are simply sub-groups within a population. In the case of drug users, it may be obvious to identify strata such as gender (i.e. male or female), types of drug user (e.g. causal, addict), and so forth, but others may be unknown to the researcher. Whilst is it typical to define the characteristics of the sample you want to examine at the start of the research process, the snowball sample may also be helpful in exploring potentially unknown characteristics that are of interest before settling on your sampling criteria.

• There may be no other way of accessing your sample, making snowball sampling the only viable choice of sampling strategy.

Snowball sampling may also be viewed as an effective sampling strategy from a perspective of research design and the choice of research methods. Whilst the use of quantitative research designs, surveys methods, and statistical analyses are geared towards the use of probability-based sampling techniques that make it possible to make statistical inferences from a sample that can be generalised to a population [see Probability sampling], if we were to use such a research design to compare students who were frequent as opposed to causal drug users, it could actually lead to significant sampling bias. Taking this example, imagine that we were to conduct our survey during morning lecturers at a university. Whilst stereotypical, we may expect that a larger proportion of these frequent drug users did not show up to the lectures compared to the causal users. If this were the case, and we would likely not know if it was or not, the sample that took part in the survey could include an over-representation of causal drug users compared with frequent drug users. This would lead to sampling bias. Whilst it could be said that such as criticism is more about research design than sampling strategy, the point of this example is to highlight that sometimes a statistically inferior sampling design can result in a more representative sample.