Types of covariate and distribution effects on parameter estimates and goodness-of-fit test using clustering partitioning strategy for multinomial logistic regression / Hamzah Abdul Hamid

摘要

This thesis presents a simulation study on parameter estimation for binaryudand multinomial logistic regression, and the extension of the clusteringudpartitioning strategy for goodness-of-fit test to multinomial logisticudregression model. The motivation behind this study is influenced by twoudmain factors. Firstly, parameter estimation is often sensitive to sampleudsize and types of data. Simulation studies are useful to assess and confirmudthe effects of parameter estimation for binary and multinomial logisticudregression under various conditions. The first phase of this study coversudthe effect of different types of covariate, distributions and sample sizeudon parameter estimation for binary and multinomial logistic regressionudmodel. Data were simulated for different sample sizes, types of covariateud(continuous, count, categorical) and distributions (normal or skewed forudcontinuous variable). The simulation results show that the effect of skewedudand categorical covariate reduces as sample size increases. The parameterudestimates for normal distribution covariate apparently are less affectedudby sample size. For multinomial logistic regression model with a singleudcovariate, a sample size of at least 300 is required to obtain unbiasedudestimates when the covariate is positively skewed or is a categoricaludcovariate.