Consider the longitudinal medical problem of monitoring the evolution of a response variable on of several individuals over time. There is sometimes the need to monitor a specific individual with greater or lesser frequency, this depending on their health condition (depending on previous measurements, individual is measured more or less often). In this talk we consider the study of Lipsitz et al. (2002). In this article they assume that the estimation of the longitudinal model is done through a likelihood function that is decomposed into two components: one for the follow-up time process and the other for the outcome process. We conducted a simulation study of longitudinal data and we estimate the model parameters taking into account the likelihood function proposed by Lipsitz et al. (2002). Our contribution is given in the proposal for a longitudinal model in which the likelihood function can not be decomposed. The aim of our work is to estimate the model parameters according to the longitudinal likelihood function here proposed and compare the results with those obtained in the previous article. We also show that by specifying an incorrect covariance model, inference produce bias estimators.