5.5. Cox Regression Using the Hybrid Method

In previous chapters, we saw that we could duplicate or closely approximate the results from a fixed effects analysis by decomposing the time-varying covariates into individual-specific means and deviations from those means, and then putting all the variables into a conventional regression analysis, possibly correcting for dependence among the multiple observations for each individual. Unfortunately, for reasons that are not clear, this approach does not seem to work well for Cox regression. For example, if the hybrid method is applied to the birth interval data, several of the variables have coefficients and p-values that are dramatically different from those shown in Output 5.3. My simulation studies of the hybrid method for Cox regression have also been discouraging. Accordingly, I cannot recommend the hybrid method for event history analysis.