|
Abstract:
|
In studies with longitudinal data, missing responses often depend on values of responses. When missingness is monotone in the sense that a subject having a missing response at time t do not have any observation after time t, Tang, Little, and Raghunathan (2003) developed a semiparametric likelihood method for the estimation of parameters of interest. In practice, however, missingness is often not monotone and a direct application of the method in Tang et al.(2003) discards observed data from subjects having nonmonotone missing values. We extend the idea in Tang et al. (2003) to nonmonotone missing data and construct a semiparametric likelihood that utilizes all observed data. Our proposed estimators may be much more efficient than those obtained by discarding data. An application is illustrated using the household income data from the Health and Retirement Study.
|
