Statistical Modelling of Breastfeeding Data

Abstract

This thesis addresses some key methodological problems in statistical modelling of breastfeeding data. Meta-analysis techniques were used to analyse aggregated breastfeeding data. Generalised linear mixed model and an extended Cox model were used with time-varying exposures to analyse longitudinal and time-to-event breastfeeding data, respectively. Shared frailty models were applied to correlated breastfeeding duration data controlling for heterogeneity. A novel two-part mixed-effects model was proposed for modelling clustered time-to-event breastfeeding data with clumping at zero.