A Zero-Inflated Ordered Probit Approach to Modeling Household Poverty Levels

This research addressed the limitations of the ordered probit (OP) regression model in handling data that contains an excessive number of zero responses. The zero-inflated ordered probit (ZIOP) model was employed to overcome this issue. This model separates the estimation of structural zeros and ordinal outcomes through two distinct components: a binary probit for zero inflation and an OP for ordered categories. Due to the absence of closed-form solutions, parameter estimation was conducted using the maximum likelihood estimation (MLE) method with the Berndt-Hall-Hall-Hausman (BHHH) iterative algorithm. The analysis was based on 4,067 household-level observations from Indonesia’s National Socio-Economic Survey, incorporating indicators of health, education, and standard of living derived from the multidimensional poverty index (MPI) framework. The result of the Vuong test (4.56) confirmed that the ZIOP model significantly outperformed the conventional OP model for zero-inflated ordinal data. Therefore, the ZIOP model is considered more appropriate for analyzing household poverty classifications with a high prevalence of zero observations.