Long-Term Recall Error in Retrospective Family Surveys: Cohabitation Histories in Denmark

Recall questions provide the possibility of obtaining longitudinal information in cross-sectional surveys and are often used in family research. Yet, recall instruments may suffer from measurement issues. This study links a recent cross-sectional survey that includes recall histories of all unions involving cohabitation to administrative cohabitation records for Denmark and assesses the recall error in cohabitation status histories. Under mild assumptions, it distinguishes recall error from other survey error types, allowing for assessing the increase in recall error as time to the recalled period increases. The quality of recall for living with a partner decreases in a linear fashion by 0.5-0.6% points per year on average over a 20-year period. Results are robust to a series of sensitivity tests. The presented estimate of recall error for cohabitation status likely provides a lower bound for general recall error and shows by example how regression calibration methods can be used to correct for recall error even if only a small random subset of data includes true values. Recall instruments used in family research are prone to recall error, and failing to account for this may lead to biased estimates especially for longer recall periods.