Homeschool socialisation — the most-asked question, and the least answerable one

Verdict

insufficient on the general question — but the specific question has a usable answer, and it is not the one either side argues about.

The popular worry is that homeschooled children are isolated. The one study that actually counted found they are not. Chatham-Carpenter had adolescents log every interaction lasting two minutes or more for a month: total contacts 49.3 for homeschoolers against 56.3 for public schoolers, not significantly different, with homeschoolers recording twice as much contact with younger children. Volume is not the problem.

What that same diary found is the finding worth carrying: peer contacts of 11.0 against 30.0 (p < .01), and rated closeness of relationships 1.90 against 2.12 on a three-point scale (p < .035). Homeschooled adolescents were not short of people. They were short of peers, and rated the relationships they had as less close. The author's own formulation is the honest one — not at risk in the number of people they interact with, at risk in feeling less closeness toward and receiving less support from peer friendships.

Everything above that level of specificity dissolves. The reassuring review corpus is a vote-count of small self-selected samples with parent-reported outcomes; its own author concedes that "what homeschooled children think about their own social skills is less clear." And the two best-controlled studies of adult civic behaviour contradict each other outright.

What the evidence shows

Three measurement methods, three different pictures, and the disagreement is the information. When a parent rates the child (the bulk of the review corpus), homeschoolers look clearly better. When the child rates themselves (Medlin's SSRS), homeschoolers look at or above published norms — which rules out one specific version of the worry, that these children privately know they are struggling. When a blind observer scores videotaped behaviour (Shyers), homeschooled children show significantly fewer problem behaviours, and the self-report measures in the same study return nulls. And when the measure is a behavioural count rather than any rating (Chatham-Carpenter), the favourable picture holds for volume and breaks for peers and closeness. The archive's measure-type rule predicts exactly this ordering, and the most objective measure gives the least flattering answer.

Shyers' result should not be over-read in the favourable direction either. Fewer disruptive behaviours in a novel mixed-group play session is at least as consistent with compliance, adult-oriented conduct, or unfamiliarity with peer-group norms as with superior social competence — and the null on assertiveness is the datum most compatible with the isolation worry. What it does establish is that the strong form of the concern is wrong: children with a serious social deficit do not usually look better than controls to blind raters.

On adult civic behaviour, the two best-controlled studies flatly contradict each other. Chen's GUTS cohort finds homeschooling associated with more volunteering in adulthood (β = 0.33, surviving Bonferroni). Hill & den Dulk's NSYR national probability panel finds formerly homeschooled young adults at a 23% predicted probability of volunteering against 48% for public school, with 77% reporting none in the past year, and the effect strengthens under controls for family religiosity, parental modelling and organised-activity counts. Both are prospective, both control heavily, and they disagree on the same construct. The plausible reconciliations — different cohorts (nurses' children versus a national sample), different outcome definitions (frequency versus any), the fact that Hill's sample is conditioned on having volunteered as an adolescent — are all untested. Any claim that homeschooling produces civically engaged adults has to explain Hill; any claim that it produces disengaged ones has to explain Chen.

And the two Cardus waves make it three answers from four probability samples. Sikkink and Skiles, pooling the 2011 and 2014 Cardus surveys — random panels rather than alumni networks recruited through movement newsletters — find home-school graduates volunteering less, giving less, reporting lower trust in neighbours, strangers and institutions, and more likely to describe a weak sense of direction and to feel helpless. That is the same sign as Hill, from an independent frame. Then Cheng and Watson's 2025 Cardus wave, on a fresh panel using the same survey brand, reports the opposite: long-term home-schoolers volunteering more and feeling helpless less. The one detail in that reversal worth holding is that its medium-term home-schoolers posted the worst mental-health scores in the study, which reads less like a dose-response and more like duration proxying for why the family home-schooled — long-term exposure meaning "chose it and stayed", medium-term meaning "pulled the child out when something went wrong". Four probability samples, three answers, and the variable that best predicts which answer you get is how the study defined the exposure. That is what an unidentified literature looks like.

Vaughn versus Schepis is the same lesson in the substance-use literature, and it is not about homeschooling at all. Both are competent analyses of the same nationally representative survey series. One asks "compared to everyone else" and gets alcohol AOR 0.50; the other splits the schooled group by dropout risk and finds homeschoolers worse than well-functioning schooled students on opioid misuse, tobacco, illicit drugs and nicotine dependence. Anyone quoting either number without naming the comparison group is quoting a design decision.

The within-homeschool heterogeneity is more robust than any between-group finding. Hodge's latent profile analysis pulls four subgroups out of 1,321 homeschooled adolescents — highly religious and engaged; limited parental monitoring; high parental warmth and support; secular permissive — and only the first is distinctly protected. Green-Hennessy, on overlapping data with a different method, finds the same split by religious tie: weaker-tie homeschoolers three times more likely to be behind grade level and 2.5 times more likely to report no extracurricular activity at all. The average homeschooler does not exist, and the risk concentrates in households with no outside structure.

Hereditarian-lens assessment

Risk: high, and here the confound is not only genetic — it is definitional.

Social and behavioural outcomes are heritable, and they are also directly produced by the family characteristics that select a family into homeschooling: religiosity, parental supervision, marital stability, community embeddedness, parental personality. Every comparison in this file is between children of different families. When Vaughn finds homeschooled adolescents half as likely to drink, the leading explanation is not the delivery of instruction — the same households transmit both the alleles and the norms. Plomin's replicated finding that measures of "the environment" are themselves heritable at h² ≈ 0.27 applies with full force to every variable here.

The definitional problem compounds it. Guterman & Neuman's matched Israeli sample is among the best designs available — matched socioeconomic background, common research procedure — and it still cannot match on why a family withdrew a child. A family that homeschools because a child was miserable at school and a family that homeschools for religious conviction are the same row in every dataset above. Chen's suggestive PTSD signal, Green-Hennessy's at-risk subgroup, and the maltreatment association Luck reviews are all exactly what reactive withdrawal would produce.

Two consequences the premise makes unavoidable:

• Every pro-homeschooling socialisation finding is confounded. Lower depression, better friendships, less substance use, fewer observed problem behaviours, more volunteering: all are family traits before they are schooling effects.
• Every anti-homeschooling socialisation finding is confounded by the same thing. Fewer peer contacts, lower rated closeness, 23% volunteering, elevated illicit drug use relative to low-risk students, maltreatment associations: all are equally consistent with which families homeschool and with withdrawal following the problem. Refusing the first inference while accepting the second would be using the premise as a weapon rather than a rule.

Boundaries & what critics say

• "Homeschooled" is not a synonym for "isolated," and the literature is still written as if it were. Most US homeschooling families now buy co-op instruction, private tutoring, online provision or part-time enrolment; families using none of these are a shrinking minority. The socialisation question as popularly posed describes an arrangement that is now the exception, and nobody has measured social outcomes separately by arrangement.
• The behavioural evidence is thirty years old and tiny. Chatham-Carpenter's diary study is n = 41, one region, one season, exclusively religious/ideological homeschoolers, and has never been replicated — and secondary sources disagree about whether the public-school peer-contact figure was 30 or 20. It is simultaneously the most informative and among the least reliable studies in the topic. Shyers is an unpublished 1992 dissertation with n = 140. Both would be footnotes in a healthier literature; here they are the best objective measurement available.
• Medlin 2006 compares to a 16-year-old norm table, not to a control group. A 2006 convenience sample of all-White Christian homeschoolers from two support groups in one Florida town is scored against the SSRS 1990 standardization sample, with no matching on race, income, parental education or religiosity — and Medlin himself notes the instrument's weak subscale psychometrics (α as low as .51). It answers "do these children report themselves as socially deficient?" and nothing broader.
• Lower psychopathology treatment prevalence is not the same as better mental health. Schools are the largest single referral pathway into adolescent services; removing the child removes the referral. Schepis's finding of the lowest treatment prevalence among homeschoolers is at least as consistent with under-detection as with wellbeing, and no study separates them.
• The maltreatment question cannot currently be answered, and both camps overstate it. Luck and colleagues — child-welfare researchers with no stake in the homeschooling debate — catalogue the studies on both sides and find them uniformly unusable: no denominator (11 states require no notification at all), advocacy funding on both flanks, and severe reverse causation because withdrawal to homeschool frequently follows a CPS investigation or truancy concern. Their summary is that the politicisation and data gaps "have produced a body of mixed and inconclusive research." What does not depend on that data is structural: educational personnel file 20.7% of US maltreatment reports and pandemic closures cut state hotline reports by 20-70%. That is an argument about detection, not incidence, and it should not be reported as one about incidence. Two independent government reviews reached the same structural conclusion by different routes: England's Badman review and Wales's Forrester review both found home education is not itself a risk factor while removing professional monitoring. Note also that the "twice the rate" figures routinely attributed to Badman are not in the published report — its text offers a qualitative statement that numbers are "disproportionately high in some authorities" plus an explicit disclaimer of causation.
• The only comparison-group study of maltreatment is developer-led, and its published rebuttal is advocacy-led. Ray & Shakeel report no elevated maltreatment rate among home-educated children from retrospective adult self-report; Stewart & McCracken's response attacks the sampling frame, correctly noting that a survey of home-school alumni who consent to be surveyed structurally cannot reach the children the concern is about. Both are recorded so both poles are visible. Neither is evidence, and the archive's position is that the incidence question is currently unanswerable in either direction — which is not the same as the answer being reassuring.
• The separatism concern has one direct test and it points the other way. Cheng measured political tolerance among undergraduates at an evangelical Christian university using years of exposure rather than a binary: more homeschooling was associated with more tolerance, private versus public schooling was null. One institution, self-selected population, settles nothing — but it is the only study that asked, and it did not find what the concern predicts.
• No study in this topic has any counterfactual worth the name. No randomisation, no lottery, no sibling comparison, no discordant-twin design, no register linkage, no policy shock. Chen's and Hill's covariate-adjusted panels are the ceiling of what exists, and they disagree.

Practical guidance

The academic side of the same decision is homeschooling outcomes, and the question of whether a parent can actually deliver instruction is parent-delivered instruction. All three carry insufficient or mixed verdicts for the same structural reason: the arrangement is chosen by the family, and nothing in this literature ever breaks that link.

• Change the question. Not "will my child be socialised?" — the volume evidence says yes, and the cross-age evidence says homeschooled children do better than average at talking to people who are not their own age. The question is where the close peer friendships come from, because that is the only place the objective measurement finds a gap.
• Activity count is not peer intimacy. Shyers matched his groups on the number and frequency of out-of-school activities and still found differences; Chatham-Carpenter found adequate total contact alongside a third of the peer contact and lower closeness. Recurring time with the same group of age-peers over years is the thing to buy, not a longer activity list.
• Act on the one subgroup finding that keeps replicating. Homeschooling households with weak community ties and no organised outside activity are flagged by two independent methods on nationally representative data. If that describes your arrangement, the concern is pointed at you specifically, and it is addressable.
• Ignore any comparison that does not name its comparison group. The same survey series yields "homeschoolers use half as much alcohol" and "homeschoolers use more opioids and tobacco than their well-functioning schooled peers."
• Do not stake the case for homeschooling on prosocial outcomes. The best-adjusted cohort study delivers volunteering and forgiveness in the same model that delivers a 23% lower probability of a college degree — and the best national panel says formerly homeschooled young adults volunteer at half the public-school rate. Citing one and not the others is not available.
• Supply the missing adult contact deliberately. School is how most adolescent mental-health referrals, learning-difficulty identifications and child-protection reports happen. A homeschooling family has to replace that function on purpose: regular paediatric contact, external assessment, and at least one non-family adult who sees the child regularly.
• Discount your own assessment of your child's social skills. Parent report is the measure most consistently favourable to homeschooling and the one with the least independent corroboration; the objective measures are the ones that found the gap.

Open questions

• Why do Chen and Hill disagree? Two prospective, heavily controlled panels give opposite signs on adult volunteering. Reconciling them — cohort, era, outcome definition, conditioning on prior volunteering — is the single most valuable piece of work available in this topic, and the data already exist.
• Has anyone repeated the interaction diary? The most decision-relevant finding in the topic rests on 41 adolescents in central Oklahoma in 1994, in a homeschooling population that no longer resembles today's. A modern replication with contact logs would be cheap and would settle more than another survey.
• Do social outcomes differ across homeschooling arrangements? Most homeschoolers now use co-ops, tutors or part-time enrolment; essentially all the socialisation evidence treats them as one group.
• How much of the mental-health picture is detection rather than prevalence? Separating the two requires linkage to clinical records rather than self-report, and would change the interpretation of one of the most-cited favourable findings.
• What happens to the reactive-withdrawal subgroup? Families who withdraw a struggling, bullied or investigated child are pooled with families who never enrolled. They plausibly drive the PTSD signal, the attainment penalty and the maltreatment association simultaneously, and no study separates them.