Multivariate analysis of Longitudinal Data in the Presence of Non-Ignorable Missing Values

Multivariate analysis of Longitudinal Data in the Presence of Non-Ignorable Missing Values

Associated Students: 
Project Level: 

HIV and AIDS remain a challenge globally, regionally and locally. As a result, we see a number of interventions like voluntary testing and counselling, safe male circumcision, prevention of mother to child transmission, etc. being put into place to manage the epidemic in Botswana. A need to estimate the efficacy and mechanisms of these interventions over time arises and is often complicated by the presence of non-random missing data, which can induce bias in estimates, standard errors and confidence intervals. In estimating intervention effects, researchers often use univariate methods and ignore missing data. This omits the joint evolution of multiple outcomes over time and underestimates the effect of missing observations in the data. An investigation of mechanisms by which interventions work becomes essential when trying to minimise costs and the presence of missing data complicates things further. Molebatsi’s project explores existing multivariate analysis methods in longitudinal studies, as well as techniques for handling missing values in estimating global intervention effect and its mechanisms. Alternative methods using BCPP (Botswana Combination Prevention Project) and simulated data sets will ultimately be proposed.