Quantifying Skip-Out Information Loss When Assessing Major Depression Symptoms

Large-scale mental health surveys screen participants for the presence of the core diagnostic criteria of a mental disorder such as major depressive disorder (MDD). Only participants who screen positive are administered the full diagnostic module; the remainder "skip-out." Although this procedure adheres faithfully to the psychiatric classification of mental disorders, it limits the use of the resulting survey data for conducting high-quality research of importance to scientists, clinicians, and policymakers. Here, we conduct a series of exploratory analyses using the Virginia Adult Twin Study of Psychiatric and Substance Use Disorders (VATSPSUD) data, a unique survey which suspended the skip-out procedure for assessing past-year MDD. Adult twins (N = 8,980) born between 1930 and 1974 were recruited from a multiple-birth record database established in 1980 and interviewed in mid-adulthood between 1987 and 1996. We compared the: (a) prevalence and levels of impairment of the diagnostic criteria (and disaggregated symptom items) of adults screening positive/negative and (b) patterns of associations between MDD diagnostic criteria (and disaggregated symptom items) under three conditions: (a) full data; (b) "skip-out" data substituted with zeros; and (c) "skip-out" data treated via listwise deletion. Important differences in the patterns of associations between diagnostic criteria and disaggregated symptom sets emerged which changed the statistical evidence regarding the dimensionality of the criteria/symptom items (i.e., Condition C). An ill-defined correlation matrix which was unsuitable for statistical analysis was produced (i.e., Condition B). Given the problems with these widely used approaches, we offer researchers and data analysts practical alternatives to using the skip-out procedure in future surveys.