Nuray Güner presenterar sitt examensarbete

A Comparison of Models for Longitudinal Count Data

Abstrakt: This thesis is concerned with the analysis of count data as
longitudinal  time ordered observations. The difficulties with these
kinds of data are that the responses for each subject are correlated,
and thus complicating the estimation procedure. We compared two
different models: the overdispersion and the gamma frailty model, by
theoretical and numerical ways. This included simulations from both
models and fitting both models to the simulated data and a real data
set in order to evaluate the best model and fit. The conclusion is
that the main difference is that the correlation is in the frailty in
the gamma frailty model while the correlations are in the observations
in the overdispersion model. This leads to different estimates of the
correlation parameters. However the gamma frailty model shows
superiority with convergence and average better estimates, although it
can not converge for small sample sizes. The overdispersion model
gives consistent estimates for the covariates and less satisfying
estimates for the correlation. A suggestion is to choose the
appropriate model depending on the shape of the correlation in counts.