Using Isotonic Regression to Estimate Order-Restricted Health Indicators for the 1997-2006 National Health Interview Survey

• Activity Number: 427
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Van Parsons*+
• Companies: Centers for Disease Control and Prevention
• Keywords: monotonic estimation ; complex survey ; replication

Abstract

Estimates of health measures within a population can often be modeled by order-restricted population parameters. Population subdomains based on gender, race or a time period often define order relations for selected disease prevalence among the subdomains, for example, interventions have decreased smoking prevalence over time. With complex survey data such estimation and inferences are typically made using design-based (general) linear models or direct-estimate cell-mean models. These models are easy to implement, but have possible drawbacks including linear model distortion of the true monotonic nature or violation of hypothesized orderings by the direct estimate orderings when using a "small sample" cell-means model. As an alternative to the standard methods, isotonic regression techniques can be used as a basis for order-preserving estimation and inference. These techniques avoid the drawbacks of the standard methods mentioned. Data from the 1997-2006 National Health Interview Survey will be used to estimate parameters for hypothesized total and partial orderings for select health statistics. Comparisons of the standard and isotonic methods will be presented and discussed.