On the effects of culture and education on intelligence.


Varieties of intelligence

Definition of intelligence. Mentions that IQ and g are not the same but doesn’t explain the difference. As far as I can tell, the distinction is that g is an unobserved factor responsible for the high correlation between scores on different areas of IQ tests.

Distinctness of Gf and Gc demonstrated by PFC damage (normal Gc, low Gf) and by autism (normal Gf, low Gc). Also by trajectories over lifespan - Gf begins declining long before Gc (but controversy over exact trajectories).

Executive function and crystallized knowledge are separately inheritable. Wait, Gc is heritable? This argues that this is mediated by culture - educated parents are likely to have educated children. I would also hypothesize that need-for-cognition could be heritable.

In young children, reading and math skills are more correlated with Gf than Gc. In adults, reading and math skills are more correlated with Gc than Gf.

PFC is inhibited by limbic system. Sustained stress may result in permanently reduced Gf.

Robert Sternberg distinguishes:

IQ measures the first. Tests for the latter two are competitive in explaining variation in school grades.

Howard Gardner adds emotional intelligence, musical intelligence, kinesthetic intelligence etc. Uncorrelated with IQ.

Duckworth/Seligman - self-discipline correlates with grades much higher than IQ (.67 vs .32), although difference is much smaller for SAT (.43 vs .36 - not statistically significant). Which makes sense, because SAT is basically an IQ test. Tested at a magnet school, not yet replicated.

Yet another set of evidence that IQ is a very limited measure of effectiveness.

IQ is correlated with many life outcomes, controlling for many variables. But this is shaky MRA - ‘what nature hath joined together, multiple regress cannot put asunder’.

Study of siblings raised together but with different IQs show high correlation with adult income. Still does not control for eg educational opportunities which are gated by IQ.

So the book is basically going to focus on IQ because it’s the only thing that has been heavily studied.

Heritability and mutability

Strong-heriditarian position - at least 3/4 of variation in IQ is genetic, some of the remainder is due to uncontrollable within-family factors (eg birth order) and by adulthood almost none of the variation is due to between-family differences. Based on studies of various relationships (twins, siblings, single parents etc) living alone or apart. Implies not much parents can do to improve IQ of children.

But twins raised apart are not in randomly chosen environments - often raised by eg relatives in similar social class / culture. When environments differ, correlation has been reported as low as .27. So actual environmental effect could be .73 or even higher.

Twins share same uterine environment, and also share same features etc which may result in similar treatment.

Lastly, heritability differs depending on class and twin studies biased towards classes with high heritability.

Scarr, McCartney, Dickens, Flynn - genes and environment also interact eg taller kids more likely to be picked for basketball team, leading to better basketball skills over time. Some environments act as multipliers for small genetic differences.

Stoolmiller - adoptive families tend to have high socioeconomic status and rate much higher than average on measures of Home Observation for Measurement of Environment. Low variation in these important factors means that studies based on adoption underestimate the impact of between-family factors. Estimates that correcting for this could put between-familiy factors as high as .5.

Turkheimer - heritability ~0.7 for upper-middle class and ~0.1 for lower class. Likely because environmental variation is lower in upper classes, so heriditary variation is all that remains.

MRA not good enough. Need natural experiments.

Capron, Duyme - tracked down ‘cross-fostering’ - children fostered into a family of different class. Children born to high-SES parents average 12 IQ points more than children born to low-SES parents (genetics + early env). Children adopted by high-SES parents average 12 IQ points more than children adopted by low-SES parents (late env).

Similar experiment tracked low-SES children adopted into high-SES families, comparing them to their non-adopted siblings. Difference of 12-16 IQ points, depending on test. School failure rate was 13% for adopted vs 56% for non-adopted. Presumably a families that give up a child for adoption are not the best environments to begin with, so this may exaggerate the difference that would be seen from SES alone.

Another looked at low-IQ abused children who were tested at adoption and again 10 years later. No correlation between IQ at adoption and SES. Gains on later test were low SES +8 IQ, middle +16 IQ, upper-middle +20 IQ. Also underestimates effects because even low-SES adoptive families have better than average environments for their class.

Similar findings from meta-analysis of well-designed adoption studies.

Takeaway - environmental factors characteristic of low-SES families reduce IQ 12-18 points below it’s potential. Plus whatever confounding effects adoption has!

Degree of heritability places no constraints on mutability anyway eg heritability of height is ~.85 but several countries have seen gains in average height of >1SD in the last generation. This is a fantastic intuition pump.

Getting smarter

Education improves IQ: Reduced growth or even drop in IQ over summer holidays (worse for low-SES children - explains much of the academic gap between low-SES and high-SES). Many natural experiments show that missing school years results in steady decline in IQ. Using school cutoff dates shows that a year of age is worth ~1/2 as much as a year of school in terms of IQ points.

Western-style schooling improves results on memory and spatial perception tasks that are supposed to only test raw intelligence. This is usually interpreted as showing that IQ tests have a cultural bias. Not sure how one would separate the two.

Flynn effect - gain of ~1/3 IQ point per year. Can’t be explained by familiarity with tests - steady increase of IQ vs rapid explosion in testing, plus repeated exposure to tests does not increase scores enough to account for the effect. Can’t be explained by nutrition - US and Western Europe have not seen significant nutrition deficits during that period (except for two world wars - how did rationing affect IQ?), increases are across the board rather than only at the (presumably nutritionally deficient) lower end.

Taking this literally, the average IQ in 1917 would have been around 73 - retarded by modern judgment. This seems unlikely. Worth comparing this to Stanovich’ work - it makes sense that massive increase in IQ doesn’t yield massive increase in intelligence because IQ is a very narrow measure of intelligence. Having better abstract reasoning skills doesn’t help you much if you don’t notice when you need to apply them.

Increase of 2 SD on Raven matrices. Blair, Williams etc note that early schooling today includes much more abstract and visual reasoning, and that math problems have more steps and require more working memory. Similarly, controlled experiments show that certain kinds of video games increase attentional control. Rosario Rueda - used computer games to train attention management and executive control in 4-year olds - showed dramatic improvements on Raven matrices after only five days. Adele Diamond - used structured play that involved memory aids and inhibition control - showed improvement in executive-function tests.

Increase of 1.6 SD on similarities test. Notes that although it’s seen as test of crystallized intelligence, it also has significant fluid component in determining which similarity is maximally informative.

Increase of .25 SD on vocab test and 1/3 SD on comprehension test.

Little increase on information test - no extra time spent memorizing.

Little increase on arithmetic test, which is strange because material is successfully taught at earlier and earlier ages and scores on standardized tests have increased substantially. Tentatively notes that arithmetic test mostly measures rote application of skills, whereas standardized tests also stress reasoning and working memory.

Improving the schools

Studies showing that voucher programs work are flawed - parents self-select. Intent-to-treat experiments show improvement of .10 SD (in grades?). Similarly weak results of randomized entry into charter schools.

MRA on class size usually reveals little effect, but a randomized controlled experiment in Tennessee found that smaller classes (average 15 students) had results .19 to .28 SD better than larger classes (average 22 students), and effects persisted for at least three more years afterwards and were greated for low-SES and minority children.

Teacher experience matters, but most improvements happen in the first year of experience. Estimates of impact of teacher quality (as defined by increase in their class scores vs average increase) is .27 SD (and measured at .20 SD within schools).

Unions oppose rewarding teachers based on measures other than certification or seniority.

Educational research is poor. Mostly anecdotal. Actively opposed to experimental method (reminds me of the situation in medicine previously, were doctors were opposed to experiments on the grounds that denying treatment to the control group is unethical).

Studies of whole school interventions are not conclusive or convincing.

Kulik - across 16 controlled studies of computerized teaching of maths (description sounds similar to Khan Academy), found median effect size of .40 SD. Similarly dramatic results for writing and natural/social sciences.

Student Teams Achievement Divisions - students work/study together in heterogenous groups of 4 - controlled experiments show effect size of .30 SD. Structured dyad - students alternating role of tutor/tutee - similarly impressive effect sizes.

Richard Herrnstein - taught [long list of techniques] (much overlap with list of rationality mindware in Stanovich’ work - very exciting) - .62 SD for language comprehension, .46 SD on representing ‘problem spaces’ (what are this?), .77 SD on decision making, .50 SD on inventive thinking, .35 SD on general-abilities test, .43 SD on IQ test, .11 SD on Cattell Culture Fair Intelligence Test (similar to Raven matrices). Then Venezuela government changed and program was cancelled!

Lepper gives advice on effective tutoring:

Social class and cognitive culture

Biological influences associated with low-SES: Some evidence that supplementing micro-nutrients in low-SES children increases IQ. Lead exposure very strongly linked to IQ loss, and exposure is somewhat correlated with class eg inner-city low-SES kids. Even small amounts of alcohol (< 2 ounces) in pregnant women affects IQ, attention and memory, and drinking during pregnancy is more common in low-SES mothers. Lower health in general in low-SES children eg dental health, exposure to pollution, vision and hearing - compounded by worse medical care. Some pesticides still common in low-SES households are associated with lower IQ. Low-SES mothers less likely to breast-feed, which may impact IQ - still debated.

Environmental influences associated with low-SES: Less stable environment eg move homes more frequently - extra stress plus disruption to schooling. Behavioral problems more common. McLoyd - low-SES parents less likely to be warm and supportive, more likely to use strict punishment. Early emotional trauma damages the PFC (why?). These influences don’t have to be that widespread to have a large impact - knock-on effects on low-SES classes, neighborhoods and social peers.

Skill disparity between US social classes - math, lit, sci as measured by OECD show almost 1 SD difference between bottom class quartile and top class quartile. In Scandinavia only 2/3 SD, mostly bringing up the bottom. In South Kora only 1/2 SD. US pattern is closer to developing countries.

US economic gap also bigger. Would raising income reduce IQ gap? Maybe, but cognitive culture is also a problem.

Hart + Risley - verbal behavior towards kids across classes. Working class parents converse less and demand more, include kids in conversation less. 1300 words / hour vs 200 words / hour => 50% more vocab. Ratio of reprimand to encouragement 6:1 vs 2:1.

Heath, and later Lareau. More reading to kids. More books. Begin reading earlier. Relate content to outside world. Encourage analysis (why did X, what if X etc, valuation). More active conversation vs reactive. => Better prepared for early school, less likely to be left behind.

Activity instruction more verbal, less physical. Low SES children do find with ‘what’ questions, but struggle with non- known-answer questions - categorization, analysis, evaluation, counterfactuals.

IQ drops over sumer.

Gains from income improvement don’t show up in first generation.

IQ in black and white

Black/white IQ gap in US was 15 points in 80s ie 1 SD! Gap exists even controlling for SES and stereotype threat.

Default belief is black/white gap is genetic. Have to break taboo on discussion in order to change minds. Strong ethical argument - refusing to acknowledge black/white gap is preventing understanding which prevents intervention.

Rush - smaller brains in blacks - brain size to IQ correlates at 0.4. But brain size not causal eg in siblings. Brain size gap exists between male/female but no IQ gap. Exists small brained community in Ecuador with normal IQ. Evolutionary direction is towards smaller brains anyway.

Genetic mixing as natural experiment. Skin color and Caucasian features each correlate with IQ ~0.1, despite social advantage. In WW2, black/white soldiers raising children in Germany - IQ gap in children is only 0.5 points. IQ in black kids in Chicago not correlated with number of European ancestors. Similarly for blood groups found more often in whites.

B/W adoption studies show gap but many flawed eg different average adoption ages, which is known to affect IQ. Moore adoption study accounts for this and found no difference.

Eagon + Holland - test B/W community college students. More vocab for W but no difference in ability to learn new vocab. Better comprehension and verbal scores, but no difference when tests are restricted to common vocab. So difference in prior knowledge but not in raw learning ability. Learning ability predicts existing knowledge within B/W but not across. Not entirely sure how to parse the stats here. Needs replication though.

Flynn effect. B IQ now > W IQ in 50s. B/W IQ gap dropped from 15 in 80s to 7.5 now. More improvement for B than W in NAEP long-term trend for reading and maths.

Gap has huge effect at tail - 15 pt difference means 18x as many W at IQ 130+, 7.5 pt difference means 6x as many W as IQ 130+. Same effect for dropouts at other tail.

B:W income 67% but B:W wealth only 12%. Redlining - keeping B out of desirable neighborhoods.

Single-parent rates higher in B.

Negative beliefs about employability of poor, young, black men. Application experiment - W resumes with felonies got better responses than otherwise identical B resumes with clean record for entry level jobs. (Might lower incentive to finish education/qualification if it doesn’t have as much value due to prejudice).

Black females 30% more likely to graduate than males (in 1965). 2x more likely to have IQ 120+. Is B/W IQ gap lower between females?

Similar effects for other low-status groups worldwide eg untouchables in India.

Involuntary minority vs autonomous minority. Argues involuntary minorities lack ‘effort optimism’ - even anti-effort eg ‘acting white’.

Early US history. ‘Colored men preferred, no Irish need apply.’ Higher occupational status and higher income for black vs irish (evidence presented is thin).

19th cen Boston blacks had 3/4 college attendance. B high school scored higher than all other schools on city-wide standard tests. School produced many elite members eg first black general.

Post- Civil War migration overwhelmed existing Northern culture. Slavery, sharecropping, Jim Crow, denied entry to most unions. No reward for work so no effort optimism in culture.

Similar effects for Irish under English landlords. But Catholic church put heavy weight behind education and opportunity.

Whole historical section is interesting but weak evidence. Skipped section on West Indies culture which was especially weak.

2/3 black mothers are single earners => lower IQ kids, more dropouts => next generation of men don’t have work => more single earners.

Heath again (in 60s/70s). Similar parenting differences between B/W with same SES. B welfare kids - 600 words / day (previous were per hour, which is correct?). 3 yr children talking to professional parents show wider vocab than black mothers on welfare talking to children!

Don’t talk to children directly, or interpret their early noises as words. No manipulative toys/puzzles. No books. No known-answer questions.

Uses very absolute terms here - what were effect sizes? What about effects of bias in observers?

Philips + Brooks-Gun (80s/90s). CNLSY and IHDD both confirm that B families score lower on HOME. Within B families, HOME scores correlate highly with IQ.

Mind the gap

Schooling alone can’t fix gap, and even schooling can’t be fixed by mandate.

Head Start. 3-4yo poor kids. 5 half-days per week. Has almost closed SES mortality gap. Initial gain of ~.35 SD on cog tests but fades over time, gone by late elem school. (Not well-controlled - groups outside program also went to some pre-kinder). Very little long-term evaluation, but 2-5% higher HS completion and 3-6% higher college attendance.

Meta-review of best (best by performance? that seems like a poor way to select data) small-scale interventions focusing on black children shows larger effect sizes but still fading to nothing.

Which programs produce long-term gains?

Perry Preschool Program. 1962-7 in Michigan. Selected black kids in poverty with low IQ mothers. Start at 3-4yo, 2yr program. Focused on cognitive and social development. 6:1 teacher ratio. Weekly 90min home visit to involve mother in education. Controlled randomized! Mean IQ 95 vs 83 at end, reverting to mean by 10yo. But huge gains in achievement eg at 14yo reading, math, language tests .50-.75 SD higher, HS grades .57 SD higher. Higher income, home ownership, marriage rates. Lower welfare, arrests. Improvement presumably due to temperament/motivation rather than raw intelligence.

Milwaukee project. Same conditions, except lasting from 6mo to first grade. 7 hrs/day, 5 days/week. Tried to reproduce middle class environment. Provided food and medical care for kids and training for mothers. Mean IQ 124 vs 94 (at end of program?) (vs 113 for not-at-risk kids)! In grades 1-4 had higher test scores at ~.75 SD (but p ~ 0.1).

Abecadarian Program. Poor, mostly black, low IQ mothers, other risk factors for retardation. Two interventions, four groups.

Kinder intervention. Starts at 6mo, year-round, full-day. 6:1 teacher ratio. (Note control group also had some form of daycare.)

School intervention. Home-school teacher, biweekly parent-teacher meetings, assist parents with employment, paperwork and doctors visits.

All groups were moved to good schools.

Not much effect from school intervention.

At 21yo, still a 4.5 IQ gap from kinder intervention. Huge achievement change. At 15yo, reading 1.4 SD, math .86 SD. 2/3 vs 1/2 HS graduation. 1/3 vs 12% college admission. At 21yo, 2/3 vs 40% in higher ed or skilled job.

Several partial replications. Project Care paired kids with identical scores on Bayley Mental Development at 6mo. By 12mo, test group scored 11pts higher (what is SD?). Another replication only ran 1yo to 3yo but still showed 3.8-5.3 IQ effect at 18yo (but achievement difference was minimal and only significant in very undersized babies). Are these deliberate replications, or just similar studies?

Not much research on home interventions. Anecdotally, seem to work best when coaching specific behaviors.

Laundry. 10-20 1.5hr sessions. Coaching response patterns to infants. High effect size, including on emotional/behavioral measures.

Charter schools make no measurable difference. Similarly for most whole-school interventions.

Project SEED - hire STEM professionals to teach poor minority kids. Socratic questioning, dialogue, debate. Test in Dallas on CAT scores showed .37 SD over comparable schools, or .19 SD over non-SEED students in same school.

Reading Recovery. Low-performing first-graders. Daily 1:1 30 min lessons for 12-20 weeks. Randomized found effect sizes of .57-72 SD across different measures, fading to .20 SD by third-grade.

Knowledge Is Power Program. 7:30am - 5:00pm weekdays. Three extra weeks in summer. Home visits. Set social/behavioral expectations. Rewards and penalties for achievement eg points redeemable for trips and snacks. Teacher burnout from high hours / effort. Evidence is mixed. Independent study by SRI international showed large improvements in number of kids scoring above median on SAT over years of schooling (this is a weird measure to report - is the mean effect less impressive?). Similarly, large improvements on number of ‘proficient’ scores on CAT. Self-selected application to school though and attrition rate is 9% / year, but at least effects are seen within-subject. Randomized controlled replication is needed.

Dweck and mindset.

Aronson et al. Poor Texas HS kids assigned mentors. Control group mentors focus on drugs. Test group mentors focus on change mindset. Reinforced by web site presenting same message, and by assignment where students represent the message. TAAS math scores .64 SD higher for males, 1.13 SD higher for females. TAAS reading scores .52 SD higher.

Oyserman et al. Poor minority HS kids. Several sessions on life planning, predicted difficulties, planning to counter difficulties and thinking which friends could best help. Small group discussions on dealing with social and academic problems. What did control group do? Improved .23 SD for GPA, .36 SD for test scores, .60 SD for retention odds. SD seems like the wrong measure for retention odds - not interested in a significant improvement if it’s still small in absolute terms.

Walton and Cohen. Intervention designed to convince participants that worries over social acceptance are normal and not a sign of racial acceptance ie countering stereotype threat for social acceptance. Grades in black students show ~1 SD improvement. Minimal effects for white students.

NLSY shows B/W AFQT (comparable to IQ) gap grows throughout HS, from ~3/5 SD to ~1 SD. Hereditarian argument is that genes have more and more effect throughout development. If the gap stayed the same, they could argue it was because education has no effect. If it was reduced, which way would we fall? Myerson et al found that B/Q IQ gap shrinks in college though. Could be differences in school quality, peer pressure, stereotype threat / self-identity. Surprised the author doesn’t suggest self-selection eg kids who don’t self-identify as academic/intelligent are probably less likely to apply to college.

Societal cost-benefit ratios of these programs. Heckman estimates 8:1 for Perry Preschool Program in less special ed classes, less extra schooling, lower crime, lower welfare, higher income. Similarly, Abecedarian program estimated at 3.78:1.

Total cost of PPP for poorest 1/3 children ~$105b. But existing public pre-kinder care is already $20b, not to mention private programs and the income gains for mothers. By comparison, 2001 tax cuts for richest 1% estimated to cost $94b.

KIPP much cheaper, but free teacher labor not sustainable. If it paid public school rates, would cost an extra $35b over existing schools.

Author not advocating for installing specific programs, just for further and better research into the area.

Generally concerned with jelly-bean effects in the mentioned studies - mentioned that many interventions fail, cherry-picked those with good results and then told a story about them. I need to learn how to correct significance testing for multiple post-hoc hypotheses. The hypothesis does at least have plausible causal mechanism based on eg known correlations between HOME and IQ, and discipline/conscientiousness predicting achievement, and poor home environment predicting poor delay of gratification.

Advantage Asia?

Flynn. Meta-review with corrected methodology finds mean East Asia IQ slightly lower than US. But huge achievement gap eg on TIMS US are .75-1.0 SD lower on math and .33-.50 SD lower on science than Japan, Korea, Taiwan, Singapore, Hong Kong. This gap starts small and grows throughout schooling. (Also incidentally demonstrates that achievement tests are not just disguised IQ tests.)

Asian children are more likely than American children to attribute success to hard work over natural gifts. American-Chinese children fall somewhere in the middle.

In repeated tests, Asian college students were more likely to respond to failure on the first test by working even harder, whereas Canadian students were more likely to work harder if they did well on the first test.

In the case of Chinese culture, the author attributes this to a long history of tying success to academic achievement and to a strong sense of familial obligation.

Compared to American students, Asian students pay more attention to environment in pictures, are more likely to attribute behavior to environment, are less prone to FAE, are less likely to attribute the confuse the contents of a forced essay with the opinions of the writer, more likely to group objects by relationship rather than category, less able to separate logical statements from the concrete grounding. Some of these are also corroborated by brain scans showing more/less activity in the corresponding areas.

Asian-Americans and residents of Hong Kong fall in the middle, implying that these differences are social rather than genetics.

Like the previous section, there is are also many cultural just-so stories that are interesting but largely unsupported.

People of the book

Jews were the first national group to achieve universal male literacy, by about 1700 years. I looked this up and the very first result argues that this was a no-true-scotsman kind of affair - 2/3 of the population simply stopped being Jewish rather than learning to read.

Ashkenazi Jews are massively over-represented among Nobel Prize winners (15-200x, depending on how you calculate it), and similarly for Turing award and Fields medal. Similarly over-represented among Ivy League students. In 1931 Polish census, Jews were 9.8% of the population but owned 22.4% of the wealth. In 1938 they owned 55% of medium-large businesses. Average IQ is 2/3 - 1.0 SD above average for white population.

Notes that the over-representation in various fields is much larger than would be naively expected from IQ alone.

Average IQ of Sephardic and Oriental Jews is no higher than average, even in Israel. Suggests a genetic source, but the various theories are problematic:

Have to put this in the context of many cultures which have similar levels of literacy and wealth but radically different levels of achievement. Similarly, various cultures have dominated in science and technology at different periods in time with no accompanying massive genetic shift.

So, what could contribute to a cultural explanation? The author gives a pile of anecdotes, acknowledges that that’s all he has, and quickly moves on the next chapter.

Raising your child’s intelligence… and your own

Can draw some obvious conclusions about how to raise your child’s intelligence from the various correlations on previous chapters.

No evidence that particular toys or music will help. Studies that concluded that stimulation is important for neuron growth compared rats locked in dark cages to rats allowed to play outside with other rats. Similar gains are observed for children moving from drastically poor environments to good environments, but no data on whether stimulation past that point is beneficial.

Exercise is plausibly beneficial. Bigger, healthier babies are more intelligent later in life. Increases growth of neurons and blood supply to brain. Staves off age-related decline. Consistent strong correlations between daily exercise and fluid intelligence.

Breast-feeding for the first nine months is beneficial for most populations, especially for premature babies.

Anticipation exercises, stimulus discrimination, conflict resolution and inhibitory control exercises all seem to have an effect, and can be found at

Working memory and attention exercises have improved scores on Raven IQ tests.

Meditation has improved scores on Raven Progressive Matrices.

Self-control (tested by delayed gratification) is a better predictor of achievement than IQ. Don’t know if it is causal, but seems plausible. No strong evidence on how to improve self-control, but children appear to learn from the examples of adults around them. Growth vs fixed mindset also appears to matter.

Giving rewards reduces intrinsic motivation. Probably better to cultivate the latter where possible rather than using bribery, where possible.

Tutoring: control, challenge, confidence, curiosity, contextualize.

Try to get children into classrooms with experience, effective teachers. Certain computer programs for teaching reading, math and science are strongly supported by the evidence and should be used. Similarly for cooperative learning tasks. Visit the What Works Clearinghouse for evidence.

No evidence that teachers with certificates or higher degrees are any better.

What we now know about intelligence and academic achievement

Heritability of intelligence is not a fixed value. It’s high (up to 70%) for high-SES families where relevant environmental variation is lower. I don’t actually recall any direct evidence in the book that environmental variation is lower in high-SES families.

Heritability doesn’t place limits on mutability - it’s only measured within a particular set of environmental variations.

Variation between social classes is a result of environmental differences more than genetics.

The evidence for a genetic role in IQ difference between races is strongly refuted by more careful attention to environmental variation in adoption studies, by the shrinking IQ gap between in the US and by Flynn effect lifting up all races past previous highs.

Differences between races begin in the home and in early childhood and so are unlikely to be erased by changes in schools alone.

There are promising interventions for low-SES and minority children, some of which are practically free.


By far the most useful part of this book is the comparison to height, which is ~90% heritable in any given study and yet which has seen huge environmental effects in practice. It’s a fantastic intuition pump for what would otherwise be a subtle point of statistics.

The observation that cross-race and cross-class adoption studies neglect to account for lower environmental variation in adoptive families is fairly convincing, especially when combined with the fact that the few studies that control for environment find stronger environmental effects.

The general lesson that observed variation limits observed correlation is well-worth remembering. I wonder how many places I will notice this mistake now that I have been given such a good intuition pump.

The overall case that intelligence is more mutable than generally believed seems pretty strong, but I haven’t read anything from the opposing side. The section on Ashkenazi Jews seemed weaker than the others - the criticism of the genetic theories wasn’t matched by much in the way of evidence and the cultural explanation was even weaker.

The chapter ‘Raising your child’s intelligence… and your own’ did not in fact discuss raising intelligence in adults. Come to think of it, I don’t think there was any discussion in the entire book of whether intelligence is mutable in adults.

I am certainly convinced now that much more research into early life interventions is warranted. If the examples here aren’t just cherry-picked flukes then the potential knock-on benefits for society far outweigh the costs.