State Regulation of Homeschooling and Homeschoolers’ SAT Scores

6,170 SAT test-takers who self-designated as homeschooled in 2001; North and South Dakota dropped for having fewer than seven test-takers. The College Board released only STATE-LEVEL aggregate means and gender breakdowns, not individual records.

Self-designated homeschooled US SAT takers in 2001 — a college-aspirant sliver of the home-education population (Belfield, working the same 2001 cohort with individual data, puts homeschoolers at 0.5% of test-takers against ~1.5% of students).

The design is a three-level comparison of state-average SAT scores across states classified as low, moderate or high regulation using the HSLDA colour-chart categories, i.e. the regulatory taxonomy of the litigation organisation that campaigns against regulation. The unit of analysis is the state, not the child: with individual scores unavailable, all within-state variation and every family-level covariate is unobservable. Nothing establishes how long a test-taker had been homeschooled, or how long they had lived in the state whose regulation is being credited or blamed. Self-designation as "homeschooled" on a College Board form is unverified. Brian Ray is founder and president of NHERI; the paper's conclusion — that regulation does not affect achievement — is the load-bearing empirical claim in the deregulation case, and this is the study that supplies it. Recorded from two independent published reviews (Gaither's Homeschooling Research Notes and the International Center for Home Education Research) plus Ray's own restatement of the finding in Ray 2010; the article itself sits behind a Cloudflare challenge at the publisher's repository and could not be retrieved.

No statistically significant difference in mean SAT score between homeschoolers in low-, moderate- and high-regulation states. Both reviewers note the same wrinkle: the highest-regulation states had the LOWEST mean scores, but not significantly so — which is the direction advocates quote and the significance they omit. The finding is a null on a state-aggregate comparison with no covariates and no individual data; it is compatible with a real regulation effect of any modest size, and equally compatible with none. Ray 2010 reproduces the same null on individual test scores from a different (testing-service) sample, where the ordering is non-monotonic and medium-regulation states score highest.

High and untouched. Families who homeschool and then sit the SAT differ from other families on heritable traits; states differ in who homeschools; nothing here adjusts for any of it. The deeper problem is that regulation is not randomly assigned to states — states regulate in response to their own political and demographic composition, so the regulation contrast is confounded with everything that makes a state a state.

Ray 2010 repeats the analysis on individual test scores from a paid-testing- service sample and again reports no significant relationship (p=.045, .041, .084 against a pre-set alpha of .01), with a non-monotonic ordering. That is repetition by the same author using the same regulatory taxonomy from the same source, not independent replication. Kunzman & Gaither 2020 report that the wider literature (Bhatt 2014; Hail 2003; Levy 2009; Stewart & Neeley 2005) likewise finds regulatory climate correlating weakly or not at all with homeschooler test scores or homeschooling rates — a consistent null across weak designs, which is not the same as an established absence of effect.