How many people should I interview? Deciding on sample size.

You’re putting together a survey and know who you want to interview and what you want to ask but how do you decide on the number of people you want to interview? There is no easy answer, the most appropriate sample size for any survey depends on the resources available and the project objectives.

Firstly, generally the larger the sample size, the better. The more people you interview the bigger the proportion of the target population you speak to and the more accurate your final data will tend to be. However, a large sample size does not guarantee an accurate or unbiased sample (see our blog on achieving representative samples). Other factors, such as the structure of your sample also impact on the precision of your data.

But budgets and timescales usually restrict the size of our sample. For most, but not all, methods of data collection the more interviews required the higher the cost and the more time we need to get the interviews. Our proposed sample size must be within budget and achievable within the project time-frame.

In some cases online interviewing has reduced the marginal cost of additional interviews to zero, making budget less of a consideration. For example, if we have an email database for all customers of an organisation we can theoretically invite all of them to complete our survey and obtain as many interviews as possible. We may, though, feel it is better not to blanket email the whole customer database as we may want to send out further surveys in future and don’t want to over-research individual customers.

Time and money tend to pressure researchers into reducing sample sizes but there are other considerations we need to make to ensure that our sample size is big enough to be fit for purpose:

• How accurate do the survey results need to be? The larger the sample size the smaller the degree of confidence around the results. Tools like this sample size calculator allow you to work out the sample size you need from your desired confidence level, confidence interval and size of your target population (Explanations of confidence level and interval are given on the page alongside the calculator).

• Which sub-groups within the sample will you need to look at in isolation? For example, if you are collecting a sample of all adults but want to be able to look at results for individual age groups alone (e.g. 16-24, 25-34 etc.) then you will need enough interviews within each age group to be confident that the results are reasonably accurate. The more varied your target population the more sub-groups you will probably need to be able to analyse and the bigger the total sample size you will need. As a guide I aim to get 100 interviews with each sub-group of interest. Budgets sometimes mean that this isn’t possible and the sample size is compromised but you should have at least 50 interviews within each key sub-group to be able to report on them with any confidence at all.

• Sometimes we may want to perform analyses that require a minimum sample size. For example multivariate analyses used for segmentation need a total sample size that is large enough to be split into groups that can be analysed separately.

• The actual size of your target audience can also limit the number of interviews you can achieve. For instance if you want to interview people who have spent more than £500,000 on a car then the number of people who actually meet this criteria is relatively small. This will limit the number of interviews you can reasonably expect to achieve. However, 100 interviews with a target population that numbers 1,000 will yield more accurate results than a sample size of 100 with a target population of 1,000,000 as you will be interviewing a much higher proportion of people that meet your criteria for selection.

Related posts:

The importance of representative samples and how to get them

Why are we weighting? A basic introduction to the concept of weighting


Submit comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Recent Posts