The Com-Poisson (CMP) model is one of the most

\r\npopular discrete generalized linear models (GLMS) that handles

\r\nboth equi-, over- and under-dispersed data. In longitudinal context,

\r\nan integer-valued autoregressive (INAR(1)) process that incorporates

\r\ncovariate specification has been developed to model longitudinal

\r\nCMP counts. However, the joint likelihood CMP function is

\r\ndifficult to specify and thus restricts the likelihood-based estimating

\r\nmethodology. The joint generalized quasi-likelihood approach

\r\n(GQL-I) was instead considered but is rather computationally

\r\nintensive and may not even estimate the regression effects due

\r\nto a complex and frequently ill-conditioned covariance structure.

\r\nThis paper proposes a new GQL approach for estimating the

\r\nregression parameters (GQL-III) that is based on a single score vector

\r\nrepresentation. The performance of GQL-III is compared with GQL-I

\r\nand separate marginal GQLs (GQL-II) through some simulation

\r\nexperiments and is proved to yield equally efficient estimates as

\r\nGQL-I and is far more computationally stable.<\/p>\r\n","references":null,"publisher":"World Academy of Science, Engineering and Technology","index":"International Science Index 108, 2015"}