Using decomposition techniques and machine learning to investigate the determinants of socioeconomic inequalities in early childcare access

Formal early childcare has a strong equalizing potential, yet access remains socioeconomically stratified. This study examines how these socioeconomic inequalities emerge and widen across three stages of the formal early childcare access process: intention to use early childcare during pregnancy, application, and actual access during the child’s first year. Using longitudinal data on approximately 2,000 families in France, collected during pregnancy and followed one year after birth, we document a progressive widening of gaps along the access pathway. Compared with high–socioeconomic status (SES) households, low–SES households are about 18% less likely to intend to use early childcare, 25% less likely to apply, and 46% less likely to obtain access. To identify the determinants of these gaps, we combine machine learning for variable selection with decomposition analyses that quantify the contribution of observable factors at each stage across a wide range of 39 predictors. At the intention stage, most of the SES gap is accounted for by differences in observable characteristics related to resources, constraints, and available alternatives, with norms contributing little. At subsequent stages, inequalities increasingly reflect institutional barriers. The largest disparities emerge at the access stage, where spot allocation-related factors favoring higher-income, working, and earlier-applying households, and knowledge of the childcare system, account for most of the gap. Overall, the results show that socioeconomic stratification in early childcare access is closely linked to the timing and design of access processes, even in systems intended to be universal.