Selection bias in national surveys due to screening: implications from a county general population survey

National alcohol surveys using face-to-face interviews, such as the US 1988 National Health Interview Survey Alcohol Supplement and the 1990 US National Alcohol Survey, for reasons of efficiency often use screener items to identify individuals who are likely to have experienced alcohol-related problems and only those individuals are chosen to respond to a list of alcohol-related problem questions. The consequence of screening is that only a subset of the current drinkers have complete data on such items. This paper examines the bias introduced by the exclusion of cases with incomplete information due to screening. Data from a regional general population survey were used to investigate possible bias due to screening because it included the screening questions used in NHIS and NAS but did not screen on those items. Risk curves and estimated probabilities from logistic regressions of three alcohol dependence symptom items and two problem indices were compared by gender across three subsamples: (1) all current drinkers (without screening); (2) those who passed the screener item for the NHIS and (3); the NAS, respectively. Results indicate that the effect of the screener items on the estimated prevalence of the measures concerned are minimal, supporting the practice, judiciously applied, when greater survey efficiency is required.