Effect of Baseline Hazards and Functional Form of Continuous Covariates on Model Misspecification

Simulation studies involving linear and quadratic functional forms of continuous covariates were carried out under various combinations of two baseline hazards. These include combination of two Weibull baseline hazards, combination of two Lognormal baseline hazards and combination of Weibull and Lognormal baseline hazards. The study examined the performances of various models progressively within the context of model misspecification. Three criteria; Likelihood Ratio (LR), Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used for model comparison. Under LR and AIC, the correctly specified models generally performed better than the misspecified models across all the baseline hazards combinations and under both linear and quadratic functional forms of continuous covariates but not for all cases under BIC. Models with combination of two Weibull baseline hazards performed best under the two functional forms of continuous covariates while those having combination of two Lognormal baseline hazards performed least.