Modeling Context Effects in the National Survey on Drug Use and Health
Wang, K. H. (2005, August). Modeling Context Effects in the National Survey on Drug Use and Health. Presented at Joint Statistical Meetings, Minneapolis, MN.
Context effects occur when the content of prior questions in a survey affect subsequent responses. Research on context effects is widespread, but few methods exist for effectively predicting context effects before they occur. Experimental surveys or pretests can be used to test for context effects but such efforts may be costly to conduct or difficult to carry out in a timely manner. In this paper, we attempt to predict the nature of context effects by modeling how respondents use contextual items to shape their responses to a subsequent “target” item in order to predict the impact of removing contextual items on responses to the target item. Data from the National Survey of Drug Use and Health (NSDUH), an in-person cross-sectional study of nearly 67,500 persons conducted each year designed to measure the prevalence and correlates of drug use in the United States, are used to test this approach for predicting context effects. More specifically, we use data from these surveys to model the response process for a question on attitudes towards smoking marijuana, which exhibited a large change in the distribution of responses between 2002 and 2003.