On LPG usage in rural Vietnamese households

Journal article, 2017

Cooking with solid biomass fuels, a common practice in the developing world, is associated with numerous problems. Hence policy makers wish to facilitate households to switch to more modern fuels. To better understand the potential for policy interventions, an enhanced understanding of household fuel choices and the process of fuel switching is paramount. The primary aim of this paper is to perform an exploratory data analysis in order to obtain a set of factors associated with rural households that are using liquefied petroleum gas (LPG). This is achieved through Random Forest analysis, a statistical technique commonly employed for solving classification problems. In this context, fuel choice defines groups to which households belong, while the random forest modelling is used to determine the importance of the variables on the correct classification. The results from this study can be used for constructing further statistical models, as basis for experimental work as well as for questioning previous models. This study ranks the overall predictive importance of a great number of variables previously used in literature on fuel switching. High importance is given to variables coupled to household wealth and history of income together with various commune level variables such as distance to nearest town. The results indicate that whether households have or will undergo fuel switching can be predicted based on area characteristics associated with rurality together with a history of household income and wealth, and furthermore there appears to be an interaction effect between these characteristics. Current income is unable to fully account for a household's wealth and history of income, which in turn appear to be associated with both the current and future use of LPG. The likeliness that a household is or will start using LPG increases with increased wealth if the household resides in a less rural environment. Furthermore, some previously used variables for modelling fuel switching may instead be explained by their association with either wealth, stable income or with level of rurality.