Abstract:
|
Longitudinal or repeated measures zero-inflated continuous (or semi-continuous) data occur frequently in biometric studies. Such data present challenges due to the presence of a large portion of zero values in addition to an often right-skewed continuous distribution of the non-zero (i.e., positive) values. One popular approach is to use a two-part model with one component that models the probability of the presence of zero values, and a second component that models the distribution of the non-zero values. Typically the non-zero values are assumed to follow a normal distribution after some known transformations. This assumption, however, may not be true and different transformations may lead to different results. We propose a marginal two-part model with a semiparametric transformation model for the second component. The transformation is unknown and can be estimated from the data. We propose a pseudo-likelihood approach to estimate the unknown parameters. The proposed maximum pseudo-likelihood estimators are shown to be consistent and asymptotically normal. Simulation studies demonstrate that the proposed approach performs well. An application to a nutrition study is provided.
|