A random forest-based analysis of household survey data to infer insights on digital inequality

This paper examines digital inequalities in Nepal based on a publicly available dataset. We build different random forest classification models and apply sensitivity analysis, permutation variable test, and partial dependence analysis to characterize digital inequality on access, skill, and use at household and individual levels. Our analysis reveals important Nepal-specific findings about digital inequality. In addition, our random forest-based analysis illustrates how non-parametric methods can explicate complex nonlinear relationships that prevail between demographic variables. This paper also illustrates how sensitivity and partial dependence analysis can aid in interpreting the so-called ‘black box’ models like random forests. One of our notable findings is that caste has very little explanatory power in explaining the adoption of digital technologies. Gender, on the other hand, is still a strong predictor of an individual’s computer skills. Although the analysis in this paper is limited to Nepal, the methodology applies to similar datasets for other countries too.