Understanding the Causes, Relationships, and Consequences of the Undercount of Divorce in Surveys

Colter Mitchell, University of Michigan

By comparing matched divorced records and survey completed marital histories, this paper decomposes the underreport of divorce into the “intermediate factors” of survey statistic quality. Particularly important for divorce are errors due to coverage, unit and item nonresponse, and measurement error (due to both motivated misreporting and recall errors). After decomposing the error, individual and couple characteristics are used to predict the errors. Gender differences in error are also discussed and analyzed. Using the decomposition, I simulate what effects these errors have on divorce rates, hazard models, and indicators of divorce calculated from survey data. Findings suggest that the error is distributed among all types of error, but is particularly large for unit nonresponse and coverage error. As for gender differences, much of the underreport is due to response differences. Also female’s better recall of divorce dates appears to be due to having custody of the children from the divorce.