Social science panel surveys typically collect data at regular intervals. In most panel surveys, the time between two consecutive interviews is one year. As much as for practical reasons, this interval is chosen, because year-on-year changes suffice to properly analyze macro- and micro-changes. For some variables however, it is worthwhile to record changes at a monthly level.
Collecting monthly data in an annual panel survey bring new challenges for survey designers. Annual interviewing intervals imply that monthly data have to be reported retrospectively. The occurrence of the seam effect (the fact that changes are overreported around the dates of the interviews), is one form of retrospective measurement error that is typically found in such data (Callegaro, 2008; Jäckle, 2008). Another type of measurement error is that the total number of reported events or status changes is usually underreported (Pascale, Roemer, & Resnick, 2009) due to recall problems or telescoping effects.
This paper reports on three combined experiments conducted in the Dutch LISS panel between December 2010 and December 2011. In the first experiment, about 2000 randomly selected respondents from the LISS panel were asked to monthly report on aspects of their health for a period of 12 months. After 12 months, these respondents also received a lengthier retrospective questionnaire on their health. In the control condition of this experiment, 2000 respondents only receive the retrospective questionnaire. The difference between the prospective and retrospective reports in the experimental group indicates measurement errors that are present in retrospective data. The retrospective data from the control condition serve as a check for possible panel conditioning effects in the experimental group.
The experiment with retrospective and prospective health-interviewing was combined with two further experiments: 2. the use of an Event History Calender (EHC) related to health and 3. Dependent Interviewing (DI). Both EHC’s and DI have been used to improve the accuracy of retrospective reporting. Rarely have they however been used in combination, and few studies have been able to use prospective data from 12 waves to study the effects of DI and EHC in retrospective interviewing.
The 2 x 2 x 2 design that results from these 3 experiments combines retrospective/prospective interviewing with the absence/presence of EHC and the presence/absence of DI. In handouts distributed before the workshop we will report the most important results in a series of graphs and tables. During the presentation, we want to outline the design of the study, and focus on the most important findings and implications for survey practices.