Understanding Mobile Internet Access and Data Plan Choice in Brazil: A Machine Learning Approach
This paper applies the Elastic Net Machine Learning technique to choose the variables that best represent the characteristics of mobile internet use in Brazil. We use regularized models to estimate the importance of a large number of variables, including socioeconomic attributes, internet and device utilization patterns, and digital skills to explain (a) access to the internet through mobile devices and (b) choice of mobile data plan. After identifying the most important variables, we estimate their marginal effects on the two dependent variables with nonlinear econometric models. The results suggest that socioeconomic characteristics and user skills have significant explanatory power in both estimations. Specifically, barriers such as age, income, and skill gaps persist, hindering inclusive mobile internet adoption. Conditional on mobile internet use, these characteristics are more common among postpaid internet data plan subscribers. Moreover, communication skills like messaging and social media use stand out regarding internet access, whereas internet utilization patterns (on the move and at work) have high explanatory power in the data plan choice.
