The four methods just considered are now fairly well known and understood. Recently, a fifth approach has been proposed, one that appears to combine the virtues of the conditional logit method with those of GEE or the mixed model. Neuhaus and Kalbfleisch (1998) have the most explicit discussion of the method for binary data, but the general strategy was previously considered by Bryk and Raudenbusch (1992) and applied to a study of criminal recidivism by Horney et al. (1995). The method can be summarized in four steps:
Calculate the means of the time-varying explanatory variables for each individual.
Calculate the deviations of the time-varying explanatory variables from the individual-specific means.
Estimate the ...