|
Abstract:
|
Respondent fatigue is often a problem in surveys. Participants may stop responding to a survey or may start choosing default answers. We developed a model to guide the order in which to ask survey questions. Our conclusions were that if we could ask all questions of all participants, we should ask them in a fixed order. If this was not possible, a subset should be randomly assigned at baseline. Monotonicity assumptions about item response could limit the number of follow-up questions. For example, if a person reported chronic health problems at baseline, it might be reasonable to assume the same at follow-up. Follow-up questions that needed to be asked should be asked in the same order as at baseline. Although not immediately intuitive, respondent fatigue can actually increase the power to detect effects in certain situations by decreasing estimator variance.
|
