2005 Volume 26 Issue 1 Pages 15-29
A varying-coefficient regression model for repeated measurements is considered. It can be applied to variety of data structures including binary and count data. A local linear estimating function is proposed. The resulting estimators of time-varying regression coefficients are consistent and asymptotically normal. Based on these asymptotic results, one can construct pointwise confidence bands of time-varying regression coefficients. For binary data, time-varying covariate effects over time on the risk ratio can be investigated according to our proposal. An illustration is provided with a dataset from a clinical trial and results of a simulation study are presented.
