Many factors determine sample size for qualitative research. They usually include practical considerations such as size of budget, number of unique groups being interviewed and in the case of in-person, the number of cities in which interviews are being conducted. They may also include subjective “comfort” factors such as what has been done in the past and good old “gut feel” for what will be sufficiently reassuring and persuasive to decision makers.
Segment detection is a potentially value outcome that typically is not considered in sampling, but it should be. But what if qualitative is your only option for a given research effort, based on market dynamics or simply available time and budget? A sample sufficiently powered to detect a need in a minority segment could help identify a source of new business. Similarly, a sample large enough to identify the existence of a discordant minority could provide valuable insight in evaluative research. Finally, the ability to detect the direction of early adopters could be the key to being properly prepared for the future. If qualitative is the right fit for your insight needs or budget, you don’t need to give up the ability to detect meaningful segments. With appropriate planning, it is possible to design an appropriate sample size for detecting segments in qualitative research.
We recently ran a series of calculations to address the issue of sample size and segment detection. The goal was to identify efficient samples that would give a high probability of producing at least three members from segments. The segments evaluated ranged from 20% to 40% – sizes large enough to be of interest to marketers. We also examined a segment of 16% – the approximate percentage of people who are innovators/early adopters. We chose a target of three or more respondents to facilitate distinguishing a potential segment from outliers.
Table 1 below shows the probability of getting three or more respondents given segments and samples of different sizes. The table is color coded with green highlighting when there is a 90% of greater chance of getting three or more respondents from the segment, yellow representing when there is a 70 to 90% chance and red representing when the probability is less than 70%. Please note that the calculations assume a representative sample.
The results suggest that you should avoid samples of 10 or fewer if you are interested in detecting the opinions of minority segments. If the market splits 60-40 or 50-50 you have a reasonable chance of detecting both perspectives. If the segment is smaller, there is a sizable risk that you will not have enough representatives in the interviews to detect it.
Samples of 15 to 20 are good for the moderately sized segments typically of interest to marketers. Samples of 15 to 20 respondents have at least an 80% probability of detecting segments in the 25% to 30% range. Larger samples increase the probability of detection but at an additional cost that may outweigh the benefits to probability. For most qualitative research a sample of 20 is likely to provide the best power to detect important segments at the most reasonable cost.
Finally, samples of 25 to 30 are needed if you want to detect smaller segments such as innovators/early adopters. If you are only interested the perspective of these market leaders a sample of 25 is a little more efficient than a sample of 30. However, if you are looking at the entire spectrum of adoption status and would like to know what is happening with the majority respondents and laggards a sample of 30 is a better choice. The probability of getting 3 or more from both early adopters and laggards segments is 77% with a sample of 30 but only 62% with a sample of 25.
This table only scratches the surface of how to design qualitative samples scientifically.
Originally published on the RGA blog on October 30, 2015 by Bruce Duncan.