The Effect of Admissions Test Preparation: Evidence from NELS:88

NELS:88 SAT sample n = 3,492; ACT sample n ~ 2,390; nationally representative US class of 1992

US high-school students, PSAT-to-SAT and ACT score changes with self-reported preparation

Observational, using the PSAT as a pretest proxy and regression controls for demographics, high-school performance, motivation proxies and other preparation activities. Coaching is self-reported. The same author's 2002 dissertation is the extended treatment of the same data - the two are ONE piece of evidence, not two, and are recorded separately only because the dissertation carries a literature census and a selection-model sensitivity analysis the article does not.

Raw comparisons make coaching look worth about 17 maths and 13 verbal for a commercial course, and 19 maths and 7 verbal for a private tutor. Full regression controls cut that to roughly 14-15 points SAT-M and 6-8 points SAT-V - about 14% and 5% of a standard deviation. The finding the archive should not lose is on the OTHER test: effects on the ACT were essentially nil (ACT-maths 0.33 to 0.73 scale points, mostly non-significant; ACT-reading negative). A preparation industry that moves one admissions test and not the other is not moving the underlying skill. Coached students were wealthier, had higher PSAT scores and were more motivated, so every naive comparison is biased upward, and the paper documents an additional selection problem at the other end - disadvantaged aspirants select out of taking the test at all.

Medium-to-high, and acknowledged. Coaching is purchased by families that differ on income, parental education and academic push; each additional covariate block shrinks the estimate, which is the signature of residual selection rather than of a stable causal parameter.

Consistent with Powers & Rock's independent national-sample estimates and with Becker's controlled-study meta-analytic figures; the ACT null is independently interesting and has not been directly re-tested.