Joinpoint Regression Methods for Complex Survey Data
To Know
About this Class
Joinpoint regression is commonly used to model trends in time-specific estimates derived from aggregate data. These methods were developed primarily for non-survey data, such as cancer registry data, under the assumption that estimates at different time points are uncorrelated or follow an autoregressive AR(1) error structure. However, directly applying existing joinpoint methods to complex survey data-for example, multistage cluster samples from the annual National Health Interview Survey-fails to account for covariance among survey estimates induced by the sample design.
To address this limitation, we extended joinpoint methods for aggregate outcomes to accommodate potentially correlated errors arising from complex survey designs. We also developed unit-level models that account for both this correlation structure and the design-based degrees of freedom required for valid inference. This presentation introduces these methods and presents results from empirical applications and simulation studies.