The early home environment — what parents can and cannot causally move

Verdict

This is the topic where the database's founding premise does its heaviest lifting, and where the gap between popular belief and causal evidence is widest.

The correlational parenting literature is measuring heredity. Parents who supply books, talk more, and structure homework also supply genes. The clean tests — adoption studies — show how much of the familiar parent-child correlation survives when you break that link:

• A college-educated mother raises an adoptee's college graduation by +7pp, versus +26pp for her biological child. Roughly three-quarters of ordinary transmission is not causal rearing.
• Moving a child out of a disadvantaged home into a screened adoptive home for their entire childhood raises adult IQ by +4.41 points. That is the environmental ceiling.
• Yet adoptees end up indistinguishable on IQ from the biological children of their adoptive families — environment matters at the deprived margin, much less across the normal range.

What is causally movable: attainment (modestly), health behaviours (drinking and smoking transmit environmentally more strongly than attainment does), literacy skills, and IQ at the deprivation margin. What is not: g across the normal range, durably.

And the flagship interventions mostly fail. The most widely disseminated home-visiting programme is null at age 18 on every pre-specified full-sample outcome, including nonverbal intelligence. Shared book reading beats nothing (g=0.23) but is indistinguishable from any other adult-attention activity (g=0.03) — a non-specific attention effect. The one clean positive is the cheapest thing tested: text-message nudges.

What the evidence shows

Source	Design	Grade	Key finding
Sacerdote 2007	quasi-random adoption, n=1,690	B	College grad from educated mother: adoptee +7pp vs biological +26pp
Kendler 2015	adopted-away co-sibling	B	+4.41 IQ points (SE 0.75); replicated +3.18 in half-sibs — the ceiling
van IJzendoorn 2005	62 studies, N=17,767	C	Adoptees gain vs origin peers; do NOT differ from adoptive-family siblings
Haworth 2010	~11,000 twin pairs	B	Shared environment 33% → 16% (ages 9→17); heritability 41% → 66%
Kitzman & Olds 2019 (NFP)	RCT, n=742	B	NULL at 18 on every pre-specified full-sample outcome
Noble 2019	meta-analysis	C	vs passive control g=0.23; vs active control g=0.03 (null)
York, Loeb & Doss 2019 (READY4K)	RCT, active control	B	+0.11 SD literacy; +0.31 for children starting behind
Plomin 2016	synthesis of replicated BG findings	C	"The environment" is itself h² ≈ 0.27 across 35 measures; ~2/3 of parenting↔child correlations are genetically mediated

Added 2026-07-30 — the premise, with numbers on it. This file has always asserted that the correlational parenting literature is measuring heredity. A citation-graph audit surfaced Plomin, DeFries, Knopik & Neiderhiser 2016, which states that claim as two of behavioural genetics' ten most replicated findings and attaches numbers to both. Finding #7: measures of "the environment" are themselves heritable — mean h² = 0.27 across 35 diverse environmental measures in a review of 55 genetic studies. The variables education research treats as inputs are partly phenotypes. Finding #8: most associations between environmental measures and psychological traits are genetically mediated — roughly two-thirds of the correlation between maternal negativity and adolescent antisocial behaviour, for instance. That is the same fraction Sacerdote's adoption data produce from the opposite direction (+7pp adoptee vs +26pp biological), arrived at by an entirely different design. The verdict and confidence do not change — this is a narrative synthesis (grade C), and its finding #5 on rising heritability of intelligence is Haworth 2010, which we already hold, so it is not independent there. What it adds is that the discount this file applies to the parenting literature is not our editorial stance; it is a replicated quantity, and it is about two-thirds.

One caveat this file keeps, from Plomin's own list: finding #2 is that no trait is 100% heritable, and typical psychological-trait heritabilities are 30–50%. The premise licenses heavy discounting of correlational parenting claims. It does not license "parenting does nothing."

On the "30 million word gap." It is among the weakest evidence in this database and among the most policy-influential: 42 families, observed roughly one hour per month, with the headline figure extrapolated rather than measured. Modern all-day recording of 329 families puts the gap at roughly 4 million words — about 7× smaller. There is no genetic control and no intervention arm, so it cannot support the causal claim built on it. (In fairness: the Sperry re-examination is a re-analysis rather than a clean refutation, and a rebuttal exists — but the original should not be cited as causal evidence for anything.)

An important correction the adversarial review forced. This database previously said "nothing durably raises IQ." That is wrong as stated. Schooling itself raises measured IQ by roughly 1-5 points per year, and the gains persist across the lifespan (Ritchie & Tucker-Drob 2018, 600,000+ participants). The defensible claim is narrower and now used throughout the database: no durable gains in g. Education's lasting cognitive benefit is on specific taught skills, not the general factor (Ritchie, Bates & Deary 2015).

Hereditarian-lens assessment

Risk: low — because the verdict deliberately rests on adoption and twin designs plus RCTs, and firewalls the correlational home-literacy literature entirely.

The architecture is the explanation for everything else in this tranche:

• Shared environment for cognitive ability shrinks with age — c² from ~0.33 at 9 to ~0.16 at 17, landing near 0.10 in adulthood, while heritability climbs toward 0.80. The family channel narrows as children grow.
• But c² is ~0.10, not zero. This database does not use the common overstatement "shared environment goes to zero." That residual corresponds to roughly 4-5 IQ points — and that is precisely what full adoption delivers. The theory and the natural experiment agree.
• Shared environment for attainment stays much larger than for ability, which is why home and school reliably move schooling outcomes while barely moving g.
• One honest qualifier: heritability appears somewhat suppressed at low SES in US samples (though not in Western Europe or Australia), so the ceiling may be higher for genuinely deprived children — consistent with adoption's effects being concentrated at the deprived margin.

Practical guidance

• Discount correlational parenting advice by default. "Children whose parents do X turn out better" is, in the typical case, mostly a genetic claim wearing an environmental costume.
• The cheap thing works best. Structured text-message nudges to parents produced +0.11 SD overall and +0.31 SD for children starting behind, against an active control, in a preregistered trial. For a school serving families, this is the highest-value parent intervention in the evidence base. (Caveat: no follow-up beyond 8 months exists, so assume the usual fadeout.)
• Don't fund intensive home visiting for cognitive outcomes. The flagship programme is null at 18 on its own pre-specified outcomes.
• Don't sell shared reading as a language intervention. It beats nothing and ties with any other adult-attention activity. (This is not an argument against reading with children — only against counting it as an evidence-based lever.)
• Target the deprived margin. Every genuine effect in this literature — adoption, Head Start's bottom-quantile gains, READY4K's low-baseline children — is concentrated where the starting environment was worst. Universal provision to already-adequate homes is where returns go to zero.

Open questions

• Whether READY4K-style nudges persist beyond 8 months — no follow-up exists.
• How much higher the environmental ceiling is at genuine deprivation, where G×SES interactions and adoption effects are largest.
• Whether the ~4-5 IQ point ceiling from whole-childhood environmental improvement can be reached by anything cheaper and less total than adoption.