A Nationwide Study on the Academic Achievement of Homeschool Students and Parents’ Reasons for Homeschooling

Not recoverable from the abstract; described as "a nationwide sample of homeschool students' standardized academic achievement test scores"

US homeschool students taking standardized achievement tests, with parent-reported reasons for homeschooling. The sampling frame is not stated in the abstract; Ray's prior nationwide studies (1990, 1997, 2000, 2010) all drew from fee-for-service testing organisations and homeschool advocacy bodies.

The first nationwide achievement study of homeschoolers in over 15 years and the successor to Ray 2010. Two analyses: a within-sample search for predictors of achievement, and a comparison of homeschoolers to the conventional-school test NORM GROUP applying controls for socio-economic status and race/ethnicity. Adding SES and race controls to a norm-group comparison is a real improvement over the raw percentile comparison of Rudner 1999 and Ray 2010 — and it is still not a counterfactual: the comparison group is a test publisher's norming sample, not matched children, there is no prior achievement measure, and selection into paid standardized testing is unaddressed. Brian Ray is founder and president of NHERI. Recorded from the published abstract; the full text sits behind a Cloudflare-protected Taylor & Francis paywall that could not be retrieved, so the sample size, sampling frame, funding and the specific weighting procedure are unknown and the effect rows below carry only what the abstract states.

Homeschool students scored at mean percentiles of 75 to 79 in reading, language, math, social studies and science. With SES and race/ethnicity controls applied, they scored significantly higher than the conventional-school norm group, weighted mean percentiles 74 to 82. Relatively few predictors explained variation in scores within the sample — the same negative result Ray 2010 reported and read as range restriction. The number that matters for this archive is the drift: Ray 2010, same author and the same testing-service design, put the mean at the 84th-89th percentile; this study puts it at 75-79 raw. A 5-10 percentile-point decline across the same programme's own nationwide samples is what regression toward the mean looks like as the homeschooling population broadens, and it is precisely the decline Ray 2010 said it had looked for and failed to find.

High and, on the evidence of the abstract, only partially addressed. Controlling for SES and race/ethnicity removes two coarse proxies for the family characteristics that select into homeschooling; it does not touch parental education beyond SES, parental commitment, child ability, or the decision to pay for testing and report the result. A norm-group comparison has no counterfactual children in it at all.

Same author, same institute and the same testing-service design as Ray 2010 and Rudner 1999, so this is the third iteration of one design rather than an independent replication — the shared selection mechanism is what is under suspicion, and repeating it cannot test it. What is informative is that the iteration moved the estimate DOWN by 5-10 percentile points, toward the norm.