Identification of Effective HIV Interventions When P >> N: A Cross-Country Application of Variable Selection Models and Elastic Net Regression Modelling
Feb 3, 2016 6:05 AM
This paper has been withdrawn from the conference.
Pascal Schäfer, Graduate Student, University of St. Gallen
Christian Hildebrand, Project Leader, University of St. Gallen
The current work identifies the most effective HIV interventions based on the analysis of a large cross-country data set involving 176 countries and more than 400 variables (i.e., a study where the number of cases is substantially smaller than the number of predictors). Specifically, this research is based on the analyses of country-level data on HIV prevention activities and a combination of other country-specific indicators such as GDP per capita, investments in the education system, male to female sex ratios, degree of urbanisation, among others. The data is based on the Gapminder database for a total of 176 countries. Based on a combination of exploratory factor analyses and elastic net regressions with bootstrapped estimates (see Tibshirani 1996; Zou and Hastie 2005), the current work reveals that instead of GDP per capita or other economic welfare related indicators, a country’s educational system and prior investments in general health care are the main drivers to reduce HIV rates. Counter to intuition, the use of contraceptives is not a major driver to reduce HIV prevalence across countries despite the vast dominance in current interventions policies worldwide. The current work has important implications for the effective design of public policy interventions and the use of variable selection models to identify the most effective indicator.