The intergenerational correlation of social assistance and selection bias in the Finnish population data

Abstract

Using the social assistance register we were able to study intergenerational correlations of social assistance recipiency in Finland and how the length of the observation window for identifying recipiency affects on the correlation coefficients. Parents’ social assistance was observed in 1990, and that of their children aged 18–32 was observed in 2005. The intergenerational correlation was .15 on average when the observation window was a calendar year for both parents and children. The correlation varied substantially according to the length of the observation window, the gender and the age of child. The intergenerational correlation was stronger in the early twenties (.20), and substantially lower (.10) in the early thirties. The correlation was stronger for boys (.19) than for girls (.12). As expected from the theory and previous studies, a shorter observation window for parents yielded higher estimates for the intergenerational correlation, and a shorter observation window for children yielded lower estimates. There are two sources of bias when using a shorter observation window. The downward identification bias results when households receiving social assistance for a short spell outside the observation window are classified as non-recipients. The upward selection bias results when households receiving long-term social assistance are over-represented as compared to parents who receive social assistance only for a short spell. These two sources of bias operate in a complex way and the direction of bias they cause for the intergenerational correlation is essentially an empirical question. Hence, when drawing (policy) conclusions from studies on intergenerational correlations, one should keep in mind that the correlation estimates are very sensitive to the length for the observation windows and to the life stage when the children’s social assistance recipiency is observed.