There is an enormous literature in the area of food choices. While much of the literature is based on surveys, the use of experiments is also popular. Experiments have a clear superiority over surveys for the ability to draw cause-and-effect relationships due to the control exercised over stimulus presentation. On the other hand, well-designed surveys allow us to make population estimates and regulators are often most interested in population-based figures.

Both experiments and surveys have their drawbacks, and the purpose of this talk is to illustrate some issues unique to experiments, and some issues unique to surveys, and also to discuss some problems associated with both. For example, an issue with food labelling experiments can be the small sample sizes achieved, which typically mean that each experiment concentrates on a relatively small subset of the population, preventing generalisations to the overall population. An issue associated with surveys is that we can only draw conclusions around associations, because there is no manipulated variable, and our measures tend to be proxies. For both types of study, a key concern is the low proportions of variance explained by the models.

This talk will draw upon recent literature on food labelling to illustrate the issues, and present some ideas for improving research design. The audience is encouraged to offer their perspectives.