The article by Regier et al1 should become a classic for mental health epidemiology. It marks the end of the third generation in mental health epidemiology, and one can draw important conclusions for the future.
Clearly, important discrepancies exist in the prevalence rates between 2 similar US population surveys for mental and addictive disorders that had been "planned to give estimates of the need for treatment." Regier et al have made a scholarly effort to compare the findings of 2 surveys that they expected to have given similar results. The Epidemiologic Catchment Area (ECA) study was done in 1980 to 1985 using DSM-III and the Diagnostic Interview Schedule (DIS). The National Comorbidity Survey (NCS) was done in 1990 using DSM-III-R and the University of Michigan Composite International Diagnostic Interview (UM-CIDI) as a partial "replication" of the ECA. After all, the surveys and instruments were of the same ancestry, so they should not differ wildly. Whatever one could possibly do to control for these differences between these 2 surveys, the authors have already compared in the article. You almost have to become an "ECAologist" rather than an epidemiologist to compare these 2 surveys: the instrument (DIS vs CIDI?), the diagnostic criteria (DSM-III vs DSM-III-R?), or special survey style (screening questions were put up front in the UM-CIDI; there was a commitment probe), the samples (ECA urban centers vs younger NCS sample?), and the algorithms (scoring in different styles).