Fixing errors in your survey data

A client with research interests in a particular disease came to us with a problem: they had spent considerable time and expense on a survey of the burdens of the disease only to discover afterwards that some patients and caretakers had been asked the wrong questions. This happened because the skip logic of their survey, the computerized rules for determining what question to ask next based on the most recent answer, had been programmed incorrectly. Therefore, for example, some patients had been asked caretaker followup questions, and vice versa. The survey was long, and the skip logic errors were sometimes nested in several layers of questions. The client had quickly realized that the logical errors were significant, that much in the dataset was still valid, and that they didn’t have the in-house programming ability or bandwidth to fix the errors. We took over, built a map of the survey’s intended logic, and used our data sleuthing abilities to remove all invalid responses and supply the client with several valid data subsets to be used for different types of research questions. This is a great reminder to all researchers who use surveys to be sure your survey is as simple as possible - when we design surveys, we try never to use skip logic if it’s avoidable!