Find information about

Section 4 Techniques and Tools

4.10 Non-random or non-probability samples

If generalising from the sample to the population is less important it is not always necessary to select a strictly statistically random sample. At other times, it may not be feasible if, for example, there is no sampling frame from which to select a sample. In these circumstances, surveys can be based on types of non-random samples.

Qualitative research samples

Qualitative research samples are usually non-random (non-probability) samples because most qualitative research is more concerned to generalise findings to theory development than to generalise findings to populations.

In qualitative research it is not just people who are sampled; time and contexts are sampled too. For example, it is important that people or events are researched at different times of the day and week so that inferences about what people do are not limited to what they do only on dayshifts or at weekends. Also, people behave differently in different settings so the research may require that a variety of settings be selected. However, depending on the research topic, it may not be possible to know the population in advance; sampling frames may not exist and it may not be possible to create them.

Types of non-random samples: quota sampling

Quota sampling aims to produce samples which reflect the broader population on some agreed criteria, without a random selection of cases. It is used widely in market research as it is quicker and cheaper than probability sampling but does not ensure absence of bias or allow any assessment of bias.

Quotas may reflect the geographic or property profile of the population from which the sample is drawn. Whilst not technically statistically representative, this is often adequate for most general survey purposes. Setting quotas requires reliable information about the distribution of properties or other variable used.

The use of some randomising procedure to select cases or addresses to include in the survey will reduce bias in the selection of exactly who to interview, rather than relying on who is available at the time of the survey. Instructing researchers to follow a random route within an area in order to select addresses for inclusion in a survey is also another way of reducing selection bias, although neither of these approaches will produce a strictly random sample.

Types of non-random sample: purposive sampling

Purposive sampling includes cases that are judged by the researcher to be typical of some category of interest to the research, such as recent service users. For surveys, purposive sampling is most likely to be of value for service-specific surveys rather than general surveys.

Flow sampling is a form of purposive sampling and may be appropriate for surveys or qualitative research with callers at a One-Stop Shop or customer service centre. There will be no sample frame or list available in advance for sampling from a flow of people and sampling and data collection happen simultaneously.

In any kind of purposive sampling, defining the population carefully will be important; for example in sampling a flow, you may wish to survey visits or visitors. The time at which the sample is chosen will also be important as the volume of callers or service requests may change during the course of a day and week.

Purposive samples can be taken at fixed times, or by counting and interviewing every nth caller or by using quotas. Each approach brings different degrees of bias and costs associated with having staff available to conduct interviews or distribute surveys or response cards.

Even if statistical representativeness is not the point of a survey, it will still be necessary to consider the range and source of responses to consider whether and how those who have responded are likely to differ from those that have not. If quotas have been used to select a sample, then responses should be compared against the original quotas, to see whether the rate of response differs. The overall response rate should also be assessed against typical responses for the type of survey such as postal or telephone. The absolute achieved numbers in the sub-groups used in the analysis are a useful pointer for judging the adequacy of the sample.

Types of non-random samples: snowball sampling

Snowball sampling is a type of purposive sampling. It involves identifying one or more people from the population being researched who can then identify other members of the population who, in turn can identify further members and so on. In this way a substantial number of people can be identified and approached to take part in the research. Snowball sampling is also known as ‘network sampling’ and is particularly useful when it is difficult to identify members of a population as may be the case when researching hard-to-reach groups.

Types of non-random samples: opportunistic sampling

Opportunistic sampling is also known as ‘convenience sampling’ and involves choosing the nearest and most convenient people to act as questionnaire respondents or as interviewees. However, making selections purely on the basis of ease and convenience is poor research practice. Good research practice involves making selections on the basis of robust criteria related to the research topic and research implementation.

Opportunistic sampling is particularly useful for qualitative researchers working in a context where most people have knowledge about the research topic and are equally suitable as interviewees. However, it is highly undesirable for survey research because it is unknown whether selected respondents are representative of the population and, therefore, the research findings cannot be used to make generalisations about the wider population.

Types of non-random samples: theoretical sampling

Qualitative researchers are typically concerned with how people understand and make sense of things and with finding out about the ‘meanings’ things have for people. Because the concern is with meanings qualitative researchers need to think about the context of the research and the range of activities, processes, events, times and locations that they may need to study. Studying meanings is not usually linked to concerns about statistical representativeness of a population, it is more concerned with the processes involved in how people derive meanings and what these meanings are about. See the section on meaningful representation for a further explanation.

Qualitative researchers often undertake ‘theoretical sampling’ to select who and what they should study. The term ‘theoretical’ relates to the selection of what or whom to study based on developing analysis, ideas or theory. Issues of importance and people are sampled on the basis that they are ‘cultural experts’. In theoretical sampling, data collection and analysis take place simultaneously; the developing analysis informs the researcher of what or whom they should study next.

Non-random or non-probability sampling: checklist

√ If statistical representativeness is the goal of the survey use a random sample.

√ If a sampling frame is unavailable or does not exist use a non-random sample.

√ Ensure researchers provide adequate technical information about sampling.

√ Consider whether events, time, locations and so on need to be sampled.

√ If opportunistic sampling is used, ensure there are justifiable reasons for its use that are related to the research topic and research implementation.

√ If members of a population are hard to identify snowball sampling will be useful.

When a sample may not be necessary

In small organisations or for locally focused surveys, it may not be necessary to sample and surveys can be based on a census of all members of the population.

It will often be desirable to promote the survey widely and encourage all people to give their views, perhaps publicising it through a newsletter. Depending on the context and purpose of the survey, it may be appropriate seek the views of all service users. In these cases, the additional expense of including everyone in the survey is likely to be worthwhile. The scope of the survey can be limited by defining a clear time period in which it will be conducted, so, for example, all those who have requested a repair in the last four weeks could be surveyed.

Previous page Next page