A colleague recently asked, “How many people need to be sampled (interviewed) to make the study valid?”

Interesting question.  Takes me to the sampling books.  And the interview books.

To answer that question, I need to know if you are gathering qualitative or quantitative data; if you are conducting two measurements on one group (like a pretest/posttest or a post-then-pre); if you are conducting one measurement on two groups; or any number of other conditions that affect sampling (like site, population size, phenomenon studied, etc).

So here is the easy answer.

If you are conducting two observations on one group, you will need a minimum of 30 participants with complete responses.

If you are conducting one observation on two groups, you will need at least 30 participants in each group with complete responses.

If you are conducting focus groups, you will need 10 -12 participants in each group and you will need to conduct groups until you reach saturation, that is, the responses are being repeated and you are not getting any new information (some folks say reaching saturation takes 3 – 4 groups).

If you are conducting exploratory qualitative research, you will need…it all depends on your research question.

If you are conducting confirmatory qualitative research, you will need…it all depends on your research question.

If you are conducting individual interviews, you will need…and here my easy answer fails me…so let me tell you some other information that may be helpful.

Dillman has a little chart on page 57 (Figure 3.1) that lists the sample size you will need for various populations sizes and three sizes of margin of error.  For example, if your population is 100 (a good size) and you want a margin of error of 5% (that means that the results will be accurate within + or – 5%, 95% of the time), you will need 49 participants with complete data sets if you think that for a yes/no question, participants will be split 50 yes/50 no (the most conservative assumption that can be made) You will need only 38 participants with complete data sets if you think that the responses will be unevenly split (the usual case).

This chart assumes random selection.  It takes into consideration variation in sample (the greater the variation, the larger the sample size needed).  It assumes maximum heterogeneity on a proportion of the population from which the sample is drawn.

Marshall and Rossman say very clearly, “One cannot study the universe…”  So you need to make selections of sites, samples of times, places, people and things to study.  The how many depends…sometimes it is one person; sometimes it is one organization;  sometimes it is more.  They say one to four respondents for case studies and mixed-methods studies; 10 groups was the average number of focus groups; 16 – 24 months was the average for observational fieldwork;  and one set of interviews they cite involved 92 participants.  It all depends on the research purpose–an unknown culture could be studied with a single in depth case study;  a study of mother’s receptivity of breast-feeding could have a huge sample or a small sample could provide thick description while a large sample would enhance transferability.  Credibility and trustworthiness of the findings must be considered.

The best answer, then, is…it all depends.

Print Friendly, PDF & Email
Be Sociable, Share!

One thought on “Sampling

  1. Molly and others. I’ve created a handy sample size calculator in an Excel spreadsheet that takes into account the desired margin of error for the sample estimates, the distribution of responses, and the desired probability level. It relies on formulas published by Dillman (2007) and Czaja and Blair (1996). All a person has to do is enter the size of the population from which he or she is sampling, specify the desired margin of error, the proportion of responses in each of two categories, and the desired probability level and the calculator spits out the necessary sample size. It has some other nice features too related to typical response rates, allowing a person to take them into account when planning a survey delivery method… I can share it with you and others. Not sure how to do that on this blog though… You can email me if you want it.

Leave a reply

<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>