|
Abstract:
|
The NSDUH produces and monitors annual prevalence estimates which are comparable over time in the population. Assessing the significance of changes across time is often referred to as trend analysis or trend testing. The trend testing literature and the current practices for NSDUH and other federal data collections were thoroughly reviewed. Emphasis was placed on trend testing practices for cross-sectional designs from a complex sample. Then, a series of simulations with ten years of NSDUH data were conducted to compare the statistical properties (e.g. classification error rates) of the most common trend testing approaches as identified through the review. These include pairwise testing that compares the current year estimates with prior year estimates, linear and quadratic orthogonal contrasts, statistical regression with time-dependent covariates, and joinpoint regression method. Outcome variables with a wide range of prevalence in the population were considered. The goal of this study is to facilitate the choice of trend testing methods for different statistical products in terms of credibility, ease of implementation and interpretation, and time implications.
|
