Epistemic status: pretty preliminary, not conclusive
Can parenting affect children’s outcomes? Can you raise your child to be better, healthier, smarter, more successful?
There’s a lot of evidence, from twin and adoption studies, that behavioral traits are highly heritable and not much affected by adoptive parents or by the environment shared between siblings.
High heritability does not strictly imply that parenting doesn’t matter, for a few reasons.
- Changes across the entire population don’t affect heritability. For example, heights have risen as nutrition improves, but height remains just as heritable. So if parenting practices have changed over time, heritability won’t show whether those changes helped or hurt children.
- Family environment and genes may be positively correlated. For instance, if a gene for anxiety causes both anxiety in children and harshness in parents, then it may be that the parenting still contributes to the children’s anxiety. If parents who overcome their genetic predispositions are sufficiently rare, it may still be possible that choosing to parent differently can help.
- Rare behaviors won’t necessarily show up at the population level. Extremely unusual parenting practices can still be helpful (or harmful), if they’re rare enough to not be caught in studies. Extremely unusual outcomes in children (like genius-level achievement) might also not be caught in studies.
- Subtle effects don’t show up in studies that easily. A person who has to spend a lot of time in therapy unlearning subtle emotional harms from her home environment won’t necessarily show up as having a negative outcome on a big correlational study.
With those caveats in mind, let’s see what the twin and adoption studies show.
Personality
In a study of 331 pairs of twins reared together and apart, a negligible proportion of the variance in personality was due to shared family environment. About 50% of the variance in personality scores was due to genetics; average heritability was 0.48.[1]
Attachment Style
In a study of 125 early-adopted adolescents, secure-attached infants were more likely to grow into secure-attached teenagers (correlation 0.30, p<0.01), and mothers of secure adolescents were more likely to show “sensitive support” (high relatedness and autonomy in resolving disagreements with children) at age 14. (p < 0.03).[2]
Antisocial/Criminal Behavior
An adoption study found that adolescents whose adoptive parents had high levels of conflict with them (arguments, hitting, criticizing and hurting feelings, etc) were more likely to have conduct problems. Correlations were between 0.574 and 0.696. Effects persisted longitudinally (i.e. past conflict predicted future delinquency).[3]
A meta-analysis of 51 twin and adoption studies found that 32% of the variance in antisocial behavior was due to genetic influences, while 16% was due to shared environment influences.[4]
Drug Abuse
In a Swedish adoption study of 18,115 children, adopted children with biological parents who abused drugs were twice as likely to abuse drugs themselves, while there was no elevated risk for having an adoptive parent who abused drugs. However, adoptive siblings of adopted children with DA were twice as likely to abuse drugs as adoptive siblings of adopted children without DA. This implies that there is both environmental and genetic influence, but suggests that environmental influence may be more about peers than parents.[5]
Psychiatric Disorders
Having a mother (but not a father) with major depression was associated in adoptive children getting major depression, in a study of 1108 adopted and nonadopted adolescents. Odds ratio of having a mother with major depression was 3.61 for nonadopted children and 1.97 for adopted children. Odds ratio of externalizing disorders if you had a mother with depression was 2.23 for nonadopted children and 1.69 for adopted children.[6]
IQ
The Minnesota Study of Twins Reared Apart, which includes more than 100 pairs of twins, and found that 70% of the variance in IQ of monozygotic twins raised apart was genetic. No environmental factor (father’s education, mother’s education, socioeconomic status, physical facilities) contributed more than 3% of the variance between twins. Identical twins correlate about 70% in IQ, 53% on traditionalism, 49% on religiosity, 34% on social attitudes, etc. Identical twins reared apart are roughly as similar as identical twins reared together.[7]
According to a twin study, heritability on PSAT scores was 50-75%, depending on subscore.[8]
Years of Schooling
In the Wisconsin Longitudinal Survey, of 16481 children of which 610 were adopted, finds that adopted parental income has a significant positive effect on years of schooling. Adoptive father’s years of schooling had a significant effect, but adoptive mother’s years of schooling were not significant. In nonadopted families, parental IQ and years of schooling (both mother and father) have a statistically significant effect.[9]
Reading Achievement
The Colorado Adoption Study finds that heritability of reading usually explains about 40% of the variance in outcomes in reading achievement, while adoptive-sibling correlations (a measure of shared environment) explain less than 10% of the variance. The rest is non-shared environment. Unrelated sibling correlations are 0.05, while related sibling correlations are 0.26. Genetic correlations rise with age (from 0.34 at age 7 to 0.67 at age 16).[10]
In the Western Reserve Twin Study of 278 twin pairs, ages 6-12, IQ score variance was mostly due to heritability (37%-78%, depending on subscore) and not on shared environment (<8%). However, school achievement was more dependent on shared environment (65-73%) than heritability (19-27%).[11]
In a twin study, spelling ability has a heritability of 0.53.[12]
Language ability in toddlers, in a twin study, was found to be more dependent on shared environment than genetics: 71% of variance explained by shared environment, 28% explained by genetics. This was reversed in the case of reading ability in 7-10-year-olds, where 72% of variance was explained by genetics, while 20% was explained by shared environment. Maybe the effects of home environment fade out with age.[13]
Academic Achievement
A twin study of 2602 twin pairs found that 62% of variance in science test scores at age 9 was explained by heredity, compared to 14% shared environment. There was no difference between boys and girls in heritability.[14]
51-54% of variance in grades, in the Minnesota Twin Study, is due to heredity, in girls and boys respectively. Similar genetic contributions to IQ (52%, 37%), externalizing behavior (45%, 47%) and engagement (54%, 49%). Shared environment mattered less (26%). The majority (55%) of the change in grades after age 11 is due to “nonshared environment.”[15]
Income
The National Longitudinal Study of Youth which included full and half-siblings found IQ was 64% heritable, education was 68% heritable, and income was 42% heritable. Almost all the rest of income variation was non-shared environment (49%), leaving only 9% explained by shared environment.[16]
In a study of Finnish twins, 24% of the variance of women’s lifetime income and 54% of the variance of men’s lifetime income was due to genetic factors, and the contribution of shared environment is negligible.[17]
Corporal Punishment
In laboratory settings, corporal punishment is indeed effective at getting immediate compliance. In a meta-analysis of mostly correlational and longitudinal studies, the weighted mean effect size of corporal punishment was -0.58 on the parent-child relationship, -0.49 on childhood mental health, 0.42 on childhood delinquent and antisocial behavior, 0.36 on childhood aggression, 1.13 on immediate compliance. There were no large adult effects significant at a <0.01 level, but there was an effect size of 0.57 on aggression significant at a <0.05 level.
Bottom line is that corporal punishment is fairly bad for childhood outcomes, but doesn’t usually cause lasting trauma or adult criminal/abusive behavior; still, there are good evidence-based reasons not to do it.
What Parenting Can’t Affect
Personality, IQ, reading ability in teenagers, and income are affected negligibly by the “shared environment” contribution. Drug abuse is also very heritable and not much affected by parenting.
What Parenting Might Affect
Reading ability in children and grades in teenagers have a sizable (but minority) shared environment component; reading ability in toddlers is mostly affected by shared environment. Grades are generally less IQ-correlated than test scores, and are highly affected by school engagement and levels of “externalizing” behavior (disruptive behavior, inattention, criminal/delinquent activity.) Antisocial and criminal behavior has a sizable (but minority) shared environment component. You may be able to influence your kids to behave better and study harder, and you can definitely teach your kids to read younger, though a lot of this may turn out to be a wash by the time your kids reach adulthood.
What Parenting Can Affect
Having a mother — even an adoptive mother — with major depression puts children at risk for major depression, drug abuse, and externalizing behavior. Conflict at home also predicts externalizing behavior in teenagers. Mothers of teenagers who treat them well are more likely to have teenagers who have loving and secure relationships with them. Basically, if I were to draw a conclusion from this, it would be that it’s good to have a peaceful and loving home and a mentally healthy mom.
Father’s income and family income, but not mother’s income, predicts years of schooling; I’m guessing that this is because richer families can afford to send their kids to school for longer. You can, obviously, help your kids go to college by paying for it.
References
[1]Tellegen, Auke, et al. “Personality similarity in twins reared apart and together.” Journal of personality and social psychology 54.6 (1988): 1031.
[2]Klahr, Ashlea M., et al. “The association between parent–child conflict and adolescent conduct problems over time: Results from a longitudinal adoption study.” Journal of Abnormal Psychology 120.1 (2011): 46.
[3]Klahr, Ashlea M., et al. “The association between parent–child conflict and adolescent conduct problems over time: Results from a longitudinal adoption study.” Journal of Abnormal Psychology 120.1 (2011): 46.
[4]Rhee, Soo Hyun, and Irwin D. Waldman. “Genetic and environmental influences on antisocial behavior: a meta-analysis of twin and adoption studies.” Psychological bulletin 128.3 (2002): 490.
[5]Kendler, Kenneth S., et al. “Genetic and familial environmental influences on the risk for drug abuse: a national Swedish adoption study.” Archives of general psychiatry 69.7 (2012): 690-697.
[6]Tully, Erin C., William G. Iacono, and Matt McGue. “An adoption study of parental depression as an environmental liability for adolescent depression and childhood disruptive disorders.” American Journal of Psychiatry 165.9 (2008): 1148-1154.
[7]Bouchard, T., et al. “Sources of human psychological differences: The Minnesota study of twins reared apart.” (1990).
[8]Nichols, Robert C. “The national merit twin study.” Methods and goals in human behavior genetic (1965): 231-244.
[9]Plug, Erik, and Wim Vijverberg. “Does family income matter for schooling outcomes? Using adoptees as a natural experiment.” The Economic Journal 115.506 (2005): 879-906.
[10]Wadsworth, Sally J., et al. “Genetic and environmental influences on continuity and change in reading achievement in the Colorado Adoption Project.” Developmental contexts of middle childhood: Bridges to adolescence and adulthood (2006): 87-106.
[11]Thompson, Lee Anne, Douglas K. Detterman, and Robert Plomin. “Associations between cognitive abilities and scholastic achievement: Genetic overlap but environmental differences.” Psychological Science 2.3 (1991): 158-165.
[12]Stevenson, Jim, et al. “A twin study of genetic influences on reading and spelling ability and disability.” Journal of child psychology and psychiatry 28.2 (1987): 229-247.
[13]Harlaar, Nicole, et al. “Why do preschool language abilities correlate with later reading? A twin study.” Journal of Speech, Language, and Hearing Research 51.3 (2008): 688-705.
[14]Haworth, Claire MA, Philip Dale, and Robert Plomin. “A twin study into the genetic and environmental influences on academic performance in science in nine‐year‐old boys and girls.” International Journal of Science Education 30.8 (2008): 1003-1025.
[15]Johnson, Wendy, Matt McGue, and William G. Iacono. “Genetic and environmental influences on academic achievement trajectories during adolescence.” Developmental psychology 42.3 (2006): 514.
[16]Rowe, David C., Wendy J. Vesterdal, and Joseph L. Rodgers. “Herrnstein’s syllogism: Genetic and shared environmental influences on IQ, education, and income.” Intelligence 26.4 (1998): 405-423.
[17]Hyytinen, Ari, et al. “Heritability of lifetime income.” (2013).
[18]Gershoff, Elizabeth Thompson. “Corporal punishment by parents and associated child behaviors and experiences: a meta-analytic and theoretical review.” Psychological bulletin 128.4 (2002): 539.
This is interesting, thank you.
Could you define what “personality” means here?
I know for me some things I think of as part of my personality feel like an inherent part of me and some parts feel like a consequence of family experiences (and some are a mix or don’t feel strongly either way). It would be interesting to know if this intuitive feeling of causal relation bears much relation to reality.
More generally, the discourse of “don’t worry about your parenting, everything is genetic anyway” sometimes makes me feel kind of defensive and invalidated because I do feel like aspects of how I was parented significantly affect my current mental health. As you mention, some of these effects might be too subtle to show up in studies, though I think normally I’d be skeptical of “this important effect is undetectable but believe me it’s there” and probably the only reason I’m inclined to believe that in this case is that it feels like the best explanation of my experience.
Personality means OCEAN traits and some others (like religiosity).
I also have the introspective sense that parenting affects subtle psychological stuff, which is why I put that caveat in there.
I think this misses a couple of important things. First, non-shared environment can be events that occur in utero. These won’t be picked up as heritable, but aren’t the sorts of things that people generally think of as environmental factors.
On the other hand, I believe these studies assume that siblings completely share an environment, but doesn’t account for differences in parenting between children, which most parents I know say occurs. Children have different temperaments and abilities and parents gain knowledge and experience over time, or the whole environment may shift due to outside factors at different points in this siblings’ lives.
>Language ability in toddlers, in a twin study, was found to be more dependent on shared environment than genetics: 71% of variance explained by shared environment, 28% explained by genetics. This was reversed in the case of reading ability in 7-10-year-olds, where 72% of variance was explained by genetics, while 20% was explained by shared environment.
It’s incredible that more than 90% of language ability variation is explained – in toddlers, 99% explained by either shared environment or genetics. For 7-10 year olds it’s *only* 92%, which is not much more believable.
It’s a formalization. All variance is said to be explained by heredity, shared environment, or non-shared environment. (Noise, among other things, goes in non-shared.)
It’s still incredible to me that only 1% of variance is due to noise for toddlers, and only a bit less incredible that only 8% is for older kids. Are toddlers really so remarkably predictable? What’s the test-retest or interrater reliability? The paper doesn’t address this. Seems to me like the conclusion from that section should be “this study is obviously an exercise in overfitting so we can’t draw any conclusions from it”.
ok, you’re probably right
And of course if there’s any true effect from nonshared environment then the implied “noise” is even lower.
On corporal punishment I’m curious what you think of SlateStarScratchpad’s claim (unfortunately on Tumblr) that it’s mainly evidence that parents who like hitting their kids or beat them severely do poorly, and that if you control for things like heredity or harsh beatings it’s not obvious that mild corporal punishment is more harmful than other common punishments: https://slatestarscratchpad.tumblr.com/post/155249630246/ilzolende-ambivalencerelations
(A relevant reply and reply-to-reply here: https://slatestarscratchpad.tumblr.com/post/155260523561/i-looked-at-the-second-study-you-linked-because-it/embed )
I checked the review article you cited (footnote’s missing from the text, but I could guess which one you meant) and it doesn’t include either of the studies Scott cited in the bibliography.
I feel like this was addressed in a coded message. Whether the twin studies were included or not in the meta-analysis doesn’t matter if it consists of averaging together a lot of correlational studies.
Nisbett has an interesting argument that adoptive and twin studies underestimate environmental effects. It goes like this: If you compare a bunch of kids in completely identical environments, the statistical tests would (correctly) tell you that 100% of the observed variation was genetic. If you compare a bunch of kids in similar environments, the statistical tests will give a higher estimate of variation due to genetics than if you compared kids across the full range of environmental variation that occurs in the wild. Adoptive families are screened for good home environments, and twins raised apart tend to be raised by families with similar socioeconomic levels, so it’s plausible that both are producing lower estimates of environmental effects than exist across the full spectrum in the wild.
“Twin studies” doesn’t mean twins raised apart. In theory, that sounds like the ideal study, but the actual studies are lousy, in part because there are so few samples.
“Twin studies” means comparing identical twins to fraternal twins. This avoids the problem of restriction of range of adoptive families. There is still restriction of range in getting people to participate in studies at all. In theory you could correct for that: if you have a really well-characterized instrument, you could observe that the sample in the twin study had less variance than the population; but I don’t think people do.
The persistent pattern across many domains at this point seems to me that you can cut off the bad tail but can’t do much to the upside tail. To the point that it is now my prior in new domains. Insert the bit about many things going right vs one or two things going wrong. But other than that, there’s no obvious candidates that I know of for what could be going on. I am somewhat suspicious about how we choose to bucket things. See agency debates wrt the DSM etc.
> Subtle effects don’t show up in studies that easily. A person who has to spend a lot of time in therapy unlearning subtle emotional harms from her home environment won’t necessarily show up as having a negative outcome on a big correlational study.
It would greatly confuse me if subtle emotional harms that mattered enough for someone to spend a lot of time in therapy unlearning them didn’t show up in self-reported happiness, mental illness, relationship/marital outcomes, career outcomes, substance abuse, or something easy to measure and easy to think of measuring. (It would also greatly confuse me if parents didn’t cause those harms non-negligibly often, but this isn’t necessarily a confusion about the studies because they might still be fairly rare, and/or not inflicting some smaller amount of them might be fairly rare, and/or their presence or absence might be correlated with genotype. And ‘depressed mother’ and ‘conflict at home’ do affect some of the outcomes that emotional damage would be expected to affect.)
FYI, there was a *truly epic* meta-analysis of *all twin studies* done back in 2015. Here is a very brief summary of what it found, with links to the study and a data visualization tool to dig in for yourself:
http://becomingeden.com/the-heritability-of-everything/
I have some criticisms of this post, most of which likely apply to your facts posts more broadly in some form. This one was close enough to my central areas of interest so that I could go into a high level detail at relatively low cost.
=====
(1) The paper “Personality similarity in twins reared apart and together” is almost 30 years old, and some of the studies since then have yielded different conclusions.
The same research group that published the paper has more recent study “Shared environmental influences on personality: A combined twin and adoption approach “https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3868213/pdf/nihms-537602.pdf with significantly larger samples (984 MZ twin pairs, 545 DZ twin pairs, 405 pairs of genetically unrelated siblings, and 204 genetically related sibling pairs) finding significant shared environment effects on 4 of the 11 personality scales reported on in “Personality similarity in twins reared apart and together.”
The authors make the important point that studies of twin pairs raised together can’t separate out the effect of shared environment and the effect of nonadditive genetic variance (the presence of the latter depresses estimates of the former if one isn’t modeling it), so that one needs other types of data such as data from adoption studies. The same remarks apply to twin studies of everything (not just personality).
Even with the larger sample sizes, the error bars around estimates of the effect of shared environment are in some cases as large as 9 absolute percentage (with one of the scales for which a significant effect wasn’t found having error bars from 0% to 18%). For only one of the 11 traits was the upper end of the 95% confidence interval less than 8%. One really needs thousands of sibling pairs to get estimates with tight error bars, and there are very few (if any) studies of this size.
(2) Related to the above point, the post has essentially no discussion of the assumptions made in twin studies and adoption studies, the biases that they may have, and what empirical evidence is available on the quantitative effect sizes of the biases, leaving open the question of the amount by which model uncertainty should lead one to lower one’s credence in the ostensible results.
(3) The quantitative “percent variance explained” figures are not really meaningful in absence of knowledge of the distributions of the underlying variables (educational attainment, antisocial behavior, income, etc.). For example, in the statement “Almost all the rest of income variation was non-shared environment (49%), leaving only 9% explained by shared environment” it will be unclear to most people much variance 9% actually is in dollar terms. Similarly, the odds ratio associated with having depression contingent on having a mother with depression is only meaningful with knowledge of baseline rates of depression and the level of severity of the condition reported on.
(4) Giving the estimates to 2 places past the decimal point without error bars gives an impression of a level of precision that’s not present. Surely it’s more useful to know that the study of association between maternal depression and child depression found an odds ratio of ~3.6 (c.i. 1.6−8.2) than it is to know that the study found an odds ratio of 3.61!
(5) Some of the results that you cite are in tension with each other, even when they’re not about the exact same topic. For example, it’s *a priori* unlikely that there would be a familial environment effect on antisocial behavior but no familial environment effect on personality (maybe no effect on *some* personality traits, but not *all* of them). Similarly, one expects that if maternal depression is an environmental factor for children developing depression, the same would be true of fraternal depression, even if the effect size is substantially smaller, such that the low effect of paternal depression is evidence against the high effect of maternal depression, and vice versa.
You generally simply state the effects found on different things in different studies, without noting the tension and offering thoughts on how it might be resolved. This makes it hard to form takeaway conclusions from the results laid out in your posts.
(6) In some cases, the studies that you cite themselves offer context relevant to (5). For example, in study “An adoption study of parental depression as an environmental liability for adolescent depression and childhood disruptive disorders” that you cited, the authors write
>>Adolescents have experienced only a portion of the risk period for the onset of many of the studied disorders, especially substance use disorders, and we can expect many incident cases of mental health problems in this cohort as it ages into adulthood. Previous research on twins suggests that genetic influences increase while the importance of the family environment diminishes during the transition from adolescence to early adulthood (35). Consequently, it will be especially important to follow the SIBS participants into early adulthood not only to characterize the emergence of genetic effects but also to determine whether parental depression has an enduring environmentally mediated effect on adult children’s mental health.
which suggests that the result should be thought of as being apples-to-apples comparable only with studies of heritability of adolescents.
You seem to be operating on a model that there’s value in laying out a bunch of numbers from papers in juxtaposition with one another without giving a cohesive sense of how the different pieces of evidence fit together, when the latter is most of where the value is.
(7) In general, your coverage of the literature is far short of comprehensive, and the decision making process that goes into the selections that you cite it is highly nontransparent.
You cite some individual studies on a given topic without citing others, even when there are review articles that discuss the studies that you cite in the context of others (for example, “An adoption study of parental depression as an environmental liability for adolescent depression and childhood disruptive disorders”, is discussed along side others in “Raised by Depressed Parents: Is it an Environmental Risk?” (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4224978/pdf/nihms590981.pdf).
The same is true of your choice of outcomes to examine. For example, the Swedish adoption study “Genetic and Familial Environmental Influences on the Risk for Drug Abuse” that finds an insignificant effect of having an adoptive parent who abuses drugs excludes alcohol abuse from consideration, contrasting with “The heritability of alcohol use disorders: a meta-analysis of twin and adoption studies” (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4345133/pdf/nihms-625385.pdf) which does find a familial environment effect. There’s ambiguity around whether (a) it didn’t occur to you to mention alcoholism, (b) you think that alcoholism is less important to discuss than abuse of other drugs, (c) you didn’t notice that alcoholism was excluded from the Swedish study (d) you considered discussing studies about alcoholism, but decided that the research literature is weak and that they shouldn’t be included.
The effect of this is that when I read your facts posts, I don’t know how representative what I’m seeing is of the literature on a subject. It’s not so much that I suspect systematic bias as much as that it seems a report on a random slice, and correspondingly plausibly misses some of the more important results. This substantially degrades the value of reading your fact posts relative to conducting an independent literature search (especially because the studies that you cite are not particularly difficult to find).
====
My basic take on the situation is that you mistakenly believe that it’s possible to usefully share knowledge of the scientific research literature ~15% of the work that would be required to do so (exposition of scientific material and forming a cohesive picture of the literature are both very hard and seldom done well!), and that you get positive feedback anyway because people have a applause lights around science, and are impressed by you being very smart, without actually deriving value for them. I think that it would be more useful to go into depth on a smaller number of topics, something that would both boost the post’s intelligibility and the quality of content.