Comparison of survival models in studying breastfeeding duration

Survival analysis is the statistical method for the study of time to an event, but there are various models to fit the data so the important question is which of these models are the most appropriate. In survival analysis, like all regression models, we can use residuals for assessment of the model fitness. There are varieties of residuals and each of them can be used for an especial purpose. In this study to compare different models in the study of factors associated with duration of breastfeeding, we used the data collected on women in Mazandaran Province, Iran during 2003-2004. For determination of factors associated with breastfeeding duration, we fitted Cox proportional hazard, exponential, Weibull, Gompertz, log-normal, log-logistic and gamma generalized models to the data. The assumptions of the above models were also checked. Cox-Snell residuals were used for assessment of the fitness of each model. Then a method was suggested to compare different models on the basis of these residuals. It was concluded that Cox proportional hazard model was the most appropriate.