Abstract
Of the 48 countries in the World Health Organization Africa region, South Africa has the highest per capita alcohol consumption. South Africa also has the third highest smoking prevalence in Africa. Drinking and smoking excessively are known to diminish health capital, which subsequently hinders productivity. In light of this, it is important to understand the socioeconomic determinants of these behaviours more intricately in South Africa so that current policies may be adapted to accommodate nuance in the determinants of individual and joint consumption of both commodities.
This thesis investigates the socioeconomic determinants of alcohol consumption, smoking and co-consumption for South Africa. Firstly, the thesis contributes to cumulative knowledge-building by investigating the effect of individual socioeconomic status on drinking and binge drinking through replication of an existing study and validation of its original findings. Replication is a key indicator of confidence with respect to the veracity of a predicted effect. Secondly, the thesis contributes to an existing body of knowledge by analysing the determinants of the participation and consumption decisions for drinking using a two-part model. Existing literature focuses primarily on the participation decision, so this thesis adds to existing literature by estimating the determinants of the consumption decision. Thirdly, the thesis examines the determinants of smoking intensity, which is sparsely considered in existing literature, by applying negative binomial and truncated negative binomial regressions to two datasets. Smoking intensity is just as important as smoking status – which is the focus of existing literature – because it describes the extent to which a person smokes. Understanding the determinants of smoking intensity could assist policymakers in identifying at-risk smokers. Fourthly, the thesis explores how depression and drinking explain smoking intensity, since both variables are known to affect smoking status. If they affect smoking status, they may also affect smoking intensity. Fifthly, the thesis explores the determinants of alcohol and tobacco complementarity, using a recursive bivariate probit, which matters because the decisions to drink and smoke are known to be related. Modelling them jointly uncovers any possible correlation between both substances. Lastly, the study increases coverage of alcohol and tobacco complementarity in Africa, by considering the role of emotional well-being, depression and life satisfaction in particular, in mediating smoking and drinking through a decomposition of the recursive bivariate probit estimations into the direct and indirect effect.
The thesis contains six distinct findings and expands coverage of Africa in the alcohol and smoking determinants literature. In Chapter 2, replication confirms most of the existing determinants presented
iv
in literature for drinking, but finds different prevalence rates for binge drinking. In Chapter 3, there are differences in the mechanisms that affect participation and consumption decisions. The baseline dataset in Chapter 4 uncovers specific determinants for smoking intensity, and finds depression insignificant and drinking significant with respect to determining smoking intensity. Chapter 5 finds that race, gender and age significantly affect co-consumption, as does life satisfaction. Depression has no direct or indirect effect on co-consumption, while life satisfaction mediates smoking and drinking.
The thesis offers three salient recommendations. Firstly, findings from this thesis encourage replication as a means of scientific validation, especially where alcohol data are limited. In addition, under-reported alcohol expenditure may be the reason why determinants that were previously significant in the single equation models may become insignificant in the two-part model. Again, findings from the two-part model require confirmation using external validation. To address these two recommendations, the thesis supports the call for a comprehensive alcohol and tobacco data survey over time in South Africa, possibly similar to the International Alcohol Control Study (IACS), which could be used to validate the findings from this chapter, conduct analyses over time and extend this research. Secondly, zero-inflated or hurdle count models are a suitable next step in modelling the participation and quantity decisions of smoking. The negative binomial regression may not provide reliable estimations if there are masses of zeros in the smoking data. Finally, the thesis recommends that evidence-based interventions and policies be used to reduce harmful drinking, smoking and co-consumption in order to improve individual-level health and, subsequently, productivity.