Household Poverty-Risk Analysis and Prediction Using Bayesian Ordinal Probit Models

Though the rate of poverty in Ghana has consistently declined over the years, some parts of the country still record substantially high figures [1], and this is a major concern for stake holders. Previous research to identify causal factors has commonly used the binary logit or probit models. These models, however, mask the effect of important intermediate information during the binary transformation of the response variable. This has the potential to misestimate the probability of poverty. In this study, the ordered probit model was used, thus creating a framework that includes the ordinal nature of poverty severity. The model was based on the round 6 dataset of the Ghana Living Standards Survey. Our findings show that poor and extremely poor were negatively affected by rural location, illiteracy, and Savannah ecological zone. Policies to eradicate poverty must therefore aim at optimizing these significant variables contributions to welfare conditions in the country.