|
Abstract:
|
In surveys that are conducted annually, new questions may be added and old questions may be eliminated from year to year. For the sake of continuity and/or trend analysis, it may be helpful and informative to predict responses to these questions in the survey years when the question did not appear. In the National Survey on Drug Use and Health (NSDUH), an annual national survey of substance use and mental health measures in the U.S. civilian population aged 12 or older, a method has been used in which regression models are fit using data from the years when the question was included in the questionnaire, and predicted values are calculated for respondents during the years when the question was not included. The assumptions required for this method will be delineated and tested using real data from the NSDUH; and the analytic risks involved in such an approach will be noted.
|
