Use of panel data analysis for V4 households poverty risk prediction

Abstract

One of the main approaches to tracking causality between income, social inclusion and living conditions is based on regression models estimated using various statistical methods. This approach takes into account quantitative and qualitative information about individuals or households that is collected in different periods of time (years in particular), thus allowing it to be transformed into multidimensional data sets, called panel data. Regression models based on panel data are able to describe the dynamics over time periods, so that the patterns can be related to changes in other characteristics. This paper utilises one of these approaches to panel data analysis RE-EM trees which are used to predict the risk-of-poverty rate of households located in the four “Visegrad” countries. The risk-of-poverty rate of individual households is computed on the basis of cluster analysis results, and it takes into account household living conditions as well as income. Subsequently, the risk-of-poverty rate is used as the outcome for the prediction model above. Certain household characteristics were chosen as predictors including: information about the “head” of the household (age, education level, marital status, etc.) and information about the number of members in the household. The results show slight differences in poverty determinants among Visegrad countries. The determinants with the highest impact on the risk-of-poverty rate are: number of household members (Czech Republic, Hungary and Slovakia) and education level (Poland).