Homeschooling and child maltreatment: A review of the regulatory context and research evidence in the United States

narrative review; no pooled sample. Catalogues roughly a dozen empirical efforts, the largest of which is a 1,253-respondent retrospective adult survey

Home-educated children in the United States; policy review covers all 50 states

Discussion/review article in a child-maltreatment journal, written by child-welfare researchers with no stake in the homeschooling debate — which is why it is the most useful single entry point on this question despite being grade D. It does not pool anything and contains no original comparison, so it inherits the design of what it reviews, and what it reviews is uniformly weak. The paper's value is that it names every study on both sides and says why each fails: advocacy funding, nonrandom samples, self-reported outcomes, absent confound controls, and — decisively — no reliable denominator, because 11 states require no notification at all and most states have no legal category that distinguishes homeschooled children from other non-enrolled children. Also contains a state-by-state table of filing, recordkeeping, parent qualification, instruction, testing and extracurricular-access rules.

The evidence base for the maltreatment/isolation concern does not currently support a directional claim in either direction. The authors' own summary: the lack of reliable data and the topic's highly politicised nature "have produced a body of mixed and inconclusive research." What exists: (a) Knox et al. 2014, 28 pediatrician-selected severe-torture cases, 47% withdrawn from formal school to be homeschooled and 29% never enrolled — no comparison group, no denominator; (b) Connecticut Office of the Child Advocate 2018, 380 students withdrawn to homeschool across six districts over three years, more than a third from families with at least one prior accepted report of abuse or neglect, and no district had a follow-up protocol; (c) Badman's 2009 England review, 0.4% of home-educated children known to social services vs 0.2% of all children, methodology widely criticised; (d) CRHE's Invisible Children database, which reports an observed fatality rate above expectation but is built from media and court records with no denominator; (e) Homeschool Alumni Reaching Out 2014, 42% of 3,000+ alumni reporting abuse or neglect, sample distributed through an abuse-survivor community; (f) Ray 2015 and Ray & Shakeel 2022/2023 reporting no elevated risk, from advocacy-organisation surveys that the authors say did not address their own methodological limits. The structural argument is the part that does not depend on any of this data: CPS detection runs on mandated reporters, and educational personnel are the single largest referral source (20.7% of reports); COVID-19 school closures produced 20-70% drops in state hotline reports, demonstrating how much detection depends on institutional contact. Homeschooling removes that contact by construction. That is an argument about DETECTION, not about incidence, and the authors are careful not to conflate them.

Not the operative confound here; the operative one is reverse causation and selection. Withdrawal to homeschool frequently FOLLOWS a CPS investigation or truancy concern, so any raw association between home education and maltreatment is partly the abuse causing the homeschooling rather than the reverse. No reviewed study separates the two.

The only finding that replicates across independent sources is the negative one — that the data needed to answer the question do not exist. Two government-commissioned reviews (Badman 2009 England, Forrester et al. 2017 Wales) reached the same conclusion that home education alone is not a risk factor but does remove professional monitoring. The positive claims on both sides (CRHE's elevated fatality rate; Ray's protective effect) each rest on a single non-replicated advocacy dataset.