A0611
Title: Propensity score matching for cross-classified data structures
Authors: Bruno Arpino - University of Padua (Italy) [presenting]
Daniela Bellani - Catholic University (Italy)
Abstract: Cross-classified data structures arise in various settings where individual units can be grouped along two or more dimensions that are not nested within one another. As a motivating example, the effect of parenting style on the educational performance of children of immigrants is considered. When studying immigrants, it is common to account for the effects of both the country of origin and the country of destination, which together define a cross-classified structure. Additionally, it is often crucial to account for the community effect, which refers to the impact of the specific immigrant group residing in a particular destination country. Propensity score methods have been developed to address multilevel structures of different types. An approach is developed to match treated and control units within communities as much as possible. If this is not feasible, the method seeks matches within the country of destination or the country of origin, with a preference order that can be adjusted by the researcher on a case-by-case basis. The proposed approach's ability to balance confounders is evaluated, and the bias of causal estimates is reduced using simulated data, which is applied to data from the program for international student assessment (PISA).
