The Evidence on Teaching

Predicting College Performance of Homeschooled Versus Traditional Students

Yu MC, Sackett PR, Kuncel NR · 2016

grade Cquasi-experimentindependentunreplicated
Sample
732 home-schooled first-year students at 140 institutions, matched 1:1 out of 824,940 traditional students at 195 institutions (College Board data, 2009-2011)
Population
US college entrants identified as home-schooled when they took the SAT. Conditions on having applied to, been admitted to and enrolled in college.
Design
Sequential nearest-neighbour matching without replacement on six variables — institution, gender, ethnicity, HSGPA, SAT and an SES composite (parental education plus log parental income) — producing near-exact matches (matching index median .03). The design answers a narrow question well: conditional on arriving at the same college with the same test scores and the same grades, do home-schooled students perform differently? It cannot speak to home-schooled young people who never applied.
Key findings
Before matching, home-schooled students looked substantially stronger: SAT 603.5 vs 557.8 per section (+45.7 points, roughly 0.52 SD) and HSGPA 3.80 vs 3.46 (roughly 0.71 SD), with higher SES. After matching, no difference in first-year GPA and no difference in retention (88% vs 89%). The finding with teeth is the validity result: high-school GPA predicts first-year college GPA far worse for home-schooled students (r=.21) than for matched traditional students (r=.51), while the SAT predicts about equally well (.39 vs .49). The home-schooling x HSGPA interaction was significant for both GPA and retention after controlling SAT and SES.
Genetic confound
Low for the validity finding, which is a within-group psychometric comparison. High for any read of the raw SAT gap, which is a selection artifact — and the matched null is itself evidence that the raw gap is selection: matching on prior attainment removes it entirely.
Replication notes
Unreplicated, though the direction of the validity finding — parent-assigned grades carry less information than externally scored tests — is what the archive's measure-type rule predicts and is corroborated in kind by every finding that home-school advantages shrink under external measurement.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
First-year college GPA, home-schooled vs matched traditional studentsadjusted-GPA differenceno difference (1.07 vs 0.99 on the institution-adjusted metric)standardizedend of first college yearactive-alternativeunder-1yrattainment
First-year retentionproportion retained88% vs 89% — no differencestandardizedend of first college yearactive-alternativeunder-1yrattainment
Raw SAT gap before matchingmean per-section SAT603.54 vs 557.81 (+45.73 points, ~0.52 SD); HSGPA 3.80 vs 3.46standardizedat applicationactive-alternativenot-applicabledomain-skill
Validity of high-school GPA for predicting first-year college GPAcorrelation rhome-schooled .21 vs matched traditional .51 (SAT .39 vs .49)standardizedfirst college yearnoneunder-1yrattainment

Cited by