Bayesian analysis of Amazon’s best-selling books via finite nested mixture models

Online shopping has become increasingly common in recent years and has influenced how we form our preferences and choose the items to buy. This influence also applies to the books we read: other readers’ online reviews are one of the most used tools to determine the next book we will buy. The increasing use of e-commerce websites has also led to a large availability of data to study how the users’ ratings interact with other variables. Here, we consider a dataset of Amazon’s best-selling books in the period 2009-2019. In particular, we study the similarities of the distributions of ratings and prices across different years. To fully capture the complexity of the observed data, we make use of flexible Bayesian nested mixture models to simultaneously avoid strict parametric assumptions and study the clustering structure of observations and years.