Career Academies: Long-Term Impacts on Labor Market Outcomes, Educational Attainment, and Transitions to Adulthood

More than 1,400 students followed to the eight-year point (young men n = 604/587 across survey waves; young women n = 854/841); 9 high schools; random assignment from 1993

Medium and large urban US districts; roughly 56% Hispanic and 30% African-American; followed eight years past scheduled high-school graduation, about ages 26-27.

True random assignment among academy applicants — the strongest design in the entire CTE literature and the only randomised evidence on earnings. Weaknesses that bound the claim. Nine sites, all well implemented and self-selected into the study, so this is efficacy rather than effectiveness; MDRC says so explicitly, noting Career Academies have proven challenging to implement at scale with high fidelity. Earnings are survey self-report, not administrative wage records. And both arms outperform national comparisons (50% earned a postsecondary credential against about 28% nationally), so the applicant pool was positively selected and the estimate is a local effect on students who wanted in.

Earnings rose 11% ($2,088 a year; $16,704 over eight years, 2006 dollars) and the gain was entirely male: +$260 a month in years one to four and +$361 in years five to eight, averaging +$311 a month against a control mean of $1,792 — a 17% increase, roughly $30,000 cumulative. For young women the point estimates were +$53 and +$118 a month, neither significant. Educational attainment moved by essentially zero: over 90% of BOTH arms obtained a diploma or GED and about 50% of both earned a postsecondary credential (16% bachelor's, 12% associate's, 22% certificate in the academy arm), with no significant difference in any category. Family formation moved: 33% versus 27% living independently with a spouse or partner and children.

None material for the internal estimate — individual random assignment among applicants. The applicant pool itself is positively selected, which limits generalisation but cannot bias the treatment-control contrast.

No independent replication of the eight-year earnings result exists. Hemelt, Lenard & Paeplow (2019) is a separate lottery on a different (technology) academy and does not measure earnings. The earnings finding rests on a single trial, which under METHODOLOGY caps confidence at medium.