In this paper, the main aim is to analyze longitudinal ordered responses that observed over discrete time and are in operation prior to the time they are sampled. In this context, a transition model was used in which current responses are modelled by conditioning on the previous measurements and covariates. The problem is that the marginal distribution of the initial observations cannot be determined from the conditional distribution without additional assumptions. This leads to lose the information in the initial measurements especially in the short sequences. We present the basic concepts of fitting transition models considering initial measurements using a very simple practical approach. We use simulation study based on 10000 replicates of data to investigate the finite sample behavior of our method and compare the results with two other approaches, naive and EM algorithm approach. Finally, we apply our suggested method in a real after birth pain data.

transition models, initial condition problem, multiple imputation, simulation study, longitudinal data.