using the same covariates as for the whole sample. In the major-ity of pre-treatment variables the difference between covariateswas non-significant (p > 0.01). These analyses provide information on which children benefit the most from ECCE programs and how effective these programs are in reducing cognitive gaps, especially for children from low and middle-low income groups. Children from the middle-low SES group, followed by those from the middle SES group, seem to benefit the most from ECCE programs, according to difference in means analyses done for the matched sampleas well as the adjusted-regression analyses.Table 5 shows how there is no ECCE effect for children from the up permiddle SES group (p > 0.05), and the same being true in reading and social sciences for children from the low SES group.However, low SES children who attended ECCE did perform betteron the math test (by 4.8 points; 0.08 SD) than those who did not attend