The Long-Run Impacts of a Universal Child Care Program

Four separate datasets, all Quebec vs rest of Canada. NLSCY/SYC for child behaviour (ages 2-3 and 5-9); SAIP/PCAP and PISA 2000-2015 for test scores at 13-16; CCHS 2001-2015 for self-reported health at ages 12-20 (~130k obs per biannual wave, ~65k per annual wave); Statistics Canada UCR crime rates 2006-2014 aggregated into age x sex x province x year cells, ages 12-20.

Children exposed to Quebec's universal $5/day childcare expansion introduced in 1997 for ages 0-4.

Difference-in-differences on a large-scale universal childcare rollout — one of the few evaluations of universal (not targeted) provision at national scale. Caveats that matter for how much weight the crime result can bear: there is only ONE treated province, so inference rests on Bester-Conley-Hansen standard errors using four regional groups and a t-distribution with three degrees of freedom; the crime analysis is on aggregated cells (age x sex x province x year), not individuals; and p-values are adjusted for multiple testing following Anderson (2008). Figure 3 shows Quebec accusation rates remain BELOW the rest of Canada throughout — the estimated "increase" is a narrowing of Quebec's crime advantage with exposure, not an absolute excess. First stage: exposure raised childcare use at ages 0-4 by 14.3 pp (early cycles) to 19.7 pp (all cycles) on a base of 46%, about a 40% rise.

The most important cautionary case for UNIVERSAL childcare expansion, but narrower than it is usually quoted. Verified negatives: non-cognitive deficits at ages 2-3 (anxiety +0.12 SD, aggression +0.12-0.17 SD) that do NOT fade and in the case of anxiety more than double by ages 5-9 (+0.28 SD); self-reported health +0.073 SD worse and life satisfaction +0.043 SD worse at ages 12-20; youth crime accusations +19% and convictions +22% at ages 12-20. Three important qualifications the usual citation drops: (1) the paper finds NO consistent long-run cognitive harm — PISA math is significantly POSITIVE (+0.20 to +0.31 SD) and the authors conclude "no consistent impact on their cognitive skills"; (2) self-reported mental health, pro-social behaviour and science scores are all nulls; (3) the outcomes are measured at ages 12-20, i.e. adolescence, NOT adulthood — the oldest exposed cohorts were only in their early 20s at writing. Combined with Tennessee, it shows scaling early-childhood provision can produce harm, not merely disappointing nulls — most plausibly when it displaces better care.

Comparable cohorts in other Canadian provinces (difference-in-differences).

Policy-driven cohort variation rather than family selection.

Split record. The contemporaneous negative non-cognitive result is replicated: this paper reproduces Baker-Gruber-Milligan (2008) on a wider sample, Kottelenberg and Lehrer (2013) confirm it persisted as the programme matured, and Haeck-Lefebvre-Merrigan (2015) agree on the PPVT decline. But Kottelenberg and Lehrer (2017) find the sign REVERSES for disadvantaged single-parent families, with the negative mean driven by two-parent families, and the long-run health/crime results in this paper are novel and not independently replicated. Within the paper itself the cognitive results are self-contradictory (SAIP/PCAP negative-insignificant vs PISA positive-significant on math).