Adjusting for Multimethod Bias Through Selection Modeling

Abstract

This article presents results of a beverage server intervention project that attempted to reduce levels of alcoholic intoxication through a set of policy changes and server training. Outcome data were collected through the use of personal interviews and structured observations in two enlisted clubs, one receiving the intervention and one acting as the comparison group. Because of the logistics of the data collection, the pooled sample of observations and interviews reflects systematic method biases as well as self-selection effects. Probit regression to model method and self-selection biases is used to adjust for these two biases in the analysis of consumption measures for total alcoholic drinks, beers, and blood alcohol concentration.