One of the key questions arising when planning a survey is how many people we need to sample in order to be comfortable with the findings.

In quantitative research the two main issues we look for are validity and reliability. Validity is addressed by making sure we are asking the right questions of the right people. So the questionnaire must be designed to ask relevant questions and the sample universe should comprise people who are clearly in the target market. Reliability is a function of the sample size. The more people we survey, the more reliable the findings. A census is the most reliable sample because it surveys everyone in the sample universe e.g. residents of a country. In commercial research we have to match reliability to budget so you want a sample size that will be big enough to be reliable yet small enough to be cost effective.

A useful way to determine your optimum sample size at an acceptable confidence level is to download this easy to use Sample Calculator App It takes the guesswork out of sampling and often shows a sample size smaller and less costly than you might have imagined.

In qualitative research the answer usually lies in the number of breaks required in the overall audience profile. Like a dinner party, focus groups work best when people are comfortable together. Age, social class, lifestage, lifestyle and special interests all play a part. Geographical spread also needs to be taken into consideration. The important thing to remember is that more groups doesn’t necessarily mean you have a better study. Each group should bring a qualitative difference to the project. If two groups of car mechanics told you a particular make of car was expensive to work on you would probably feel comfortable excluding that car from your wish list. Additional groups might harden your conviction but your ROI would drop significantly with each added group.

If you have a query about sampling for either a quantitative or a qualitative study contact us and we’ll do our best to provide you with the right answer.