The Son Becomes The Father

A vigorous discussion has been triggered by the release of Gregory Clark’s The Son Also Rises: Surnames and the History of Social Mobility. In this book, Clark details his work which shows a large transmission of status from generation to generation, all across the world, going back centuries. The discussion has raged on the mode of this transmission. How does it occur? Clark found that when you look at surnames, which trace paternal lineages, you find that the status of families today is related to their status centuries ago. That is, surnames associated with high status in the past are over-represented among high status individuals today, and vice versa. This pattern holds across much of the world, from England, to Sweden, to Japan, to Korea, to China, to Chile. The pattern goes back centuries – Normans surnames are still overrepresented among the English elite. The descendants of the samurai still dominate the Japanese upper classes. The intergenerational correlation of status – a measure that includes wealth, education, occupation, longevity, etc (i.e., the “good stuff” of life), was as high as 0.8 in Clark’s analysis. Clark did find that regression to the mean occurred; high-status families became less high-status with time and vice versa, but it took a very long time – 10 to 15 generations, to for them to get there.

See this talk by Greg Clark, where he explaining his findings:

But why does the pattern that Clark found occur? That remains a key point of debate. Does it occur because of genetic inheritance of traits that lead one to success or failure (which include IQ, determination, cunning, physical health, attractiveness, etc.)? Or does it occur because of the advantage (or lack there of) conferred by one’s family status (e.g., a leg up into prestigious schools, the direct effect of wealth, connections, etc.)?

Greg Cochran and Henry Harpending have both shared their analyses of the situation:

 The Son Also Rises | West Hunter

Simple Mobility Models | West Hunter
Simple Mobility Models II | West Hunter

Cochran on it:

In the short run, from one generation to the next, luck plays a big role. In the longer run, the fact that the subpopulation being examined has a different genotypic average, one more likely to result in high status, means that regression to the mean of the general population is slow for the subgroup, essentially caused by gradual change in its average genotype, change produced by intermarriage with individuals who on average have a less favorable genotype.  Other than high heritability, the other prerequisite for this pattern is highly assortative mating for moxie. If two groups have different average amounts of moxie, complete endogamy (as in Indian castes) would ensure that the between-group difference would continue indefinitely,  disregarding selection.

So is that it? Is this long-term transmission of legacy genetic? It appears that way. The pattern that we see is much what one would expect of a lineage over time if, collectively, the additive genetic components of this success factor was largely passed on from one generation to the next. Indeed, it really shouldn’t be any different. The individual variation is caused by a variety of factors, including environmental “luck”, non-additive genetic effects, developmental noise, and spousal genetic contribution (which may help or hinder). But, the key point, when the whole clan is considered at once, all these sources of variance should more or less cancel out. The only thing that breeds true is the additive genetic variance, and, in any large clan, that should pass on fairly uninterrupted from one generation to the next. The whole clan’s short-term generation-to-generation variance can be caused by variation in local circumstances that may help or hinder the entire lineage. That too should, over the generations, cancel out, in good part. The success of the clan over time is then dictated by its evolutionary fitness and the degree of assortative mating.

Assortative mating is key to perpetuating this process. The more assortative mating there is, the more each clan (which, by the way, is a fitting term in this instance, even for NW European societies, where there are no “clans” in HBD Chick‘s sense) retains the genes for success (or failure) and the slower the regression to mean. Non-assortative mating is the hole through which genes leak out over time, hastening regression and familial turn-over throughout the ages.

So what of these genetic traits that are germane to success? What are they? IQ is definitely one of them, and perhaps the single most important. But there are others, physical health surely, attractiveness likely, certain personality traits, such as conscientiousness, as well as not so admirable traits, such as those on the Dark Triad(Tetrad) (i.e., narcissism, Machiavellianism, psychopathy, sadism), especially Machiavellianism. Maybe we could describe an “m” (moxie) factor (analogous to the g factor) that underlies them all (and maybe this gives credence to the idea that there is a “general factor of personality”). Of course, as we know, a number of these traits (e.g., conscientiousness, attractiveness, health) are positively correlated with IQ, so there’s that.

After all, what gives us this pattern?

Courtesy Steve Hsu

That I will leave to others to hash out. For my contribution, I wanted to zoom in a bit on a more local level and look at our particular snapshot in time, and examine something that should give commenters on this matter (and everyone else with an interest in this) pause.

The idea that this transmission of status over time has been as Clark found it squares well with another facet I discuss frequently on this blog: the fact that parenting doesn’t have much of a lasting effect on children’s outcomes.

That’s right, all the things parents do for their children, beyond the incredibly demanding task of keeping children alive and healthy, doesn’t count for much in the long run. We know this because of the absence of shared environment effects on children’s outcomes: children do not resemble the people they grow up with once you subtract the effects of heredity. Yes, the similarity between parents and child relative to the environment as a whole is completely due to shared genes. See my posts All Human Behavioral Traits are Heritable and Taming the “Tiger Mom” and Tackling the Parenting Myth for more on the mechanics of this.

The interesting thing is that even the people who take me seriously on this point still believe that there’s something their efforts can do, beyond keeping their children fed, clothed, clean, and cognizant of the basic ways of the world. Steven Sailer frequently suggests that the outcome of poorer children, especially those of color (mostly Hispanics) would improve if they had fewer of them, and hence could afford to invest more in each, despite the fact that this doesn’t hold up in adoption studies.

And to be sure, parents play and have played an important role in passing on skills and knowledge to their children. This is often a long and highly involved process, but, as most any parent knows, a rewarding one. But though parents play this vital role, when it comes to how our kids turn out, it’s best to think of it as “you can lead a horse to water….” The child’s innate abilities and proclivities, plus whatever developmental luck he or she possesses, will guide his or her path through the world.

However, a lot of parents – especially in the West today – feel one of their biggest goals is to see that their children receive the best education possible. Education unlocks many of the goodies in society today, and as such, it’s typically best that a child maximize his educational attainment to do best in life.

This is something where parents – even those aware of the non-effect of parenting – often innately feel that they have a role to play. And they may be justified in thinking so.

A large meta-analysis of behavioral genetic studies of educational attainment (Branigan, McCallum, and Freese, 2013) performed across much of the developed world (Australia, Britain, Germany, Italy, Denmark, Finland, Norway, Sweden, Spain, and the United States) found a significant and fairly substantial shared environment component to adult educational attainment. They found that the shared environment accounted for, all told, 36% of the variance in adult educational attainment, and 24% when only people born after 1950 were considered.

Interesting, isn’t it? Very interesting, you might think. Some may be quick to declare that this invalidates all that have been saying about the non-effect of parenting. Since education is something that clearly “matters” so much in life, then parents should redouble their efforts, right?

Well, it actually turns out that it doesn’t quite work that way.

There are actually quite a few complications to this, and quite likely this doesn’t mean what it appears to mean. For one, “educational attainment” is often measured in these studies as either a quantitative variable (years of schooling completed) or a categorical one (degree attained). It doesn’t add in the obvious complication of where and how. A quick look at the previous chart of transmission of income by occupation should be one clue. Another clue is this chart by Razib Khan, that makes it abundantly clear not every bachelor’s degree is created equal:

See also this SAT to IQ conversion of various college majors. We see a big difference between physics, astronomy, mathematics, and engineering majors on one end, with average IQs in the 125-135 range, and social work, early childhood, student counseling on the other, with average IQs barely above 100. I’d wager that if you were to decompose the “rigor” of the degree attained by major and by prestige of the institution that awarded it, the shared environment component would disappear. I’ll leave this for a future project.

It’s also worth mentioning also that “shared environment”, while it would be where you’d look to find the effects of parents and upbringing, doesn’t necessarily mean parents. We see this with studies of smoking initiation (which often occurs in teenage years). The extended twin family design, which looks not just at twins, but their parents, grandparents, non-twin siblings, avuncular relatives, etc., can decompose the “parental” shared environment from the “non-parental” shared environment.

Edit, 1/22/16: [Here something I wrote on the matter previously (on a now defunct site) that I now quote here:

The “shared environment” can mean a lot of things, and in fact, it can be confounded by quite a few things. Normally, this is not much of a problem, because the shared environment is reliably zero for adult outcomes.

When we find a non-zero shared environment, then we have a problem. Beyond the normal “ACE” components of behavioral genetics (additive genetic; common, or shared environment; and unique environment), there’s other things, like non-additive genetic variance (D, or “dominance”, in behavioral genetic equations), and assortative mating (which would erroneously inflate the shared environment measure at the expense of the additive genetic measure).

And of course, “environment” in the shared environment could come from several sources. It could come from parents. Or it could be from peers (or “not parental shared environment”, broadly), as I have noted.

Fortunately, there’s a way of breaking all of these apart. The extended twin design looks not just twins, but their relatives, parents, children, grandparents, uncles, nieces/nephews, cousins, etc. This type of study is pretty hard to do right, and in order to get reliable measurements, especially when the trait in question is on the uncommon side, you need huge samples, ideally >20,000.

In this type of study, you can pull out all of these possible influences. You can measure non-additive genetic variance by seeing if the correlation between more closely related relatives reliably exceeds what you’d expect from their level of genetic relationship alone (MZ twins being much more correlated than DZ twins, for example). You can measure the effect of non-parental shared influence (e.g., peers) by seeing if the correlation between siblings is greater than the correlations between parents and children (it shouldn’t be). You can look at the effect of assortative mating by looking at the correlation between spouses of twins and their co-twins.

And there’s a paper that did just that for smoking initiation. See here [Maes, Neale, Kendler, et al, 2006]. The paper explains many of these analyses in detail.

And what did they find? The lion’s share of the variance comes from additive genetic factors (as usual), of which about 10% resulted from assortative mating (slicing that much off the shared environment). No non-additive genetic effects were found. As for the shared environment, most of it was from “non-parental sources”. Some of it was indeed twin-specific; the twin correlation was higher than the non-twin sibling correlation (as well, opposite sex DZ twin correlation was lower than same sex DZ twin correlation). These shine the spotlight squarely on peers as a key factor.

The idea of “cultural” transmission from parents doesn’t hold here, because, the correlation (less genetics) was actually considerably negative! Children tended to be considerably less like their parents than you’d expect from shared genes, especially men. Part of that is no doubt cohort effects (the enormous decline in smoking over time). But, those that like to think children rebel against their parents may have something to munch on here. ***End Edit***]

This study found that “cultural transmission” (i.e., from parents) couldn’t explain the pattern seen in children (indeed, the parent-child correlation was negative once you removed heredity). The non-parental environment explained the variance, suggesting that other influences, such as peers, likely explain the results.

Another major issue with the finding of a nonzero shared environment effect exists. This issue squares the matter with Gregory Clark’s results. That is, when you consider other facets, education per se doesn’t seem to mean much in the end. Apparently, you can’t teach moxie. This is revealed by the fact that every trait “going in” that shapes a person (and should be relevant to educational attainment) reliably shows absolutely no shared environment impact. This includes not just the most well-known example, IQ…

…but personality traits, and not just some “broad major personality dimensions”, I mean highly specific behaviors, and including one’s work preferences and interests, the presence or absence of mental disorders, and including the features of a person we think of as “character.” Parents leave no lasting effect on any of it, aside from what they bequeath to their children genetically. See this review of the behavioral genetic evidence by Thomas Bouchard and Matt McGue [2002], as well as this one by Bouchard (2004). The high heritability and zero shared environment is also seen in the Dark Triad(Tetrad) traits as well, as seen in this meta-analysis [Evertsson & Meehan 2012] of the heritability psychopathic traits.

It’s worth mentioning that many of these studies don’t partition out measurement error, that is, inaccuracies in the assessment of the traits to be analyzed. This has the effect of attenuating the heritability estimate. Other studies, which use methods to get around that problem [Riemann, Angleitner, and Strelau, 1997], find heritabilities for personality traits in the 0.7-0.8 range, as is found for IQ, mental disorders, and physiological variables like height and BMI.

The significance of measurement error brings me to another thing. Sure, parenting might have no effect on intelligence or behavioral characteristics, but what about “values” and “beliefs,” the things many parents hope to instill in their children? Well, there’s no effect of the shared environment there, either. We see this for overall religiosity and religious values. (A note here, a review of behavioral genetic studies [Koenig & McGue, 2011] on this found that there was a non-insignificant shared environment component to religiosity in adults. However, religious beliefs and convictions are traits for which assortative mating [Watson et al, 2004] is very strong [Zietsch et al 2011] – indeed, my wife and I are both atheists, and so are my wife’s sister and her husband. This would have the effect of erroneously inflating the shared environment estimate at the expense of the heritability estimate in MZ twin-DZ twin studies. One study in the review compared MZ twins raised apart with those raised together. It found a zero shared environment, making it likely that assortative mating is behind the non-trivial shared environment finding in MZT/DZT studies). We also see this with political beliefs, as found by Peter Hatemi et al’s massive meta-analysis across five countries [Hatemi et al, 2014], and with their “extended family design” twin study [Hatemi et al 2010], which included a longitudinal component, allowing for both the partitioning of any shared environment findings and accommodation of the effects of assortative mating and measurement error:

Indeed, when we consider the effect of measurement error (adding it to the heritability estimate and to the somewhat nonsensical negative gene-environment correlation values), the heritability of political attitudes and social values skyrockets, being upwards of 85% (74%) for views towards pornography in women (men). The heritability of overall political orientation, when accounting for measurement error, teeters on 100%!

Liberals and conservatives will be battling for a long time to come.

So as we see, the heritability of everything that goes into forming a person is high, the shared environment, which represents the effects of parents, is zero. (It’s worth mentioning for those who are unfamiliar with these terms that there is also a “unique environment” term, which tends to be somewhere between 50-20% or so, typically lower once you account for measurement error. Hence the “shared environment” ≠ “all environment.”)

By the way, these findings don’t just hold in Western countries. The high heritability of and nil shared environment impact on behavioral traits are also found in Japan and South Korea [Hur et al 2013]. These East Asian cultures – with vastly different attitudes towards parenting – show the same pattern of heritable and shared environment influence as do Westerners.

But that’s all OK, yes? The whole point of education is to “shape” the raw individual beyond his/her genetic predilection, right? Wrong.

The problem is that everything that comes out, the adult outcomes, shows a shared environment impact that is also zero. These include:

• Criminality (massive meta-analysis [Rhee & Waldman 2002] of twin and adoption studies shows insignificant shared environment effect in adults)
• Marital stability/divorce risk (WW II [Trumbetta, Markowitz, and Gottesman, 2007] and Vietnam era twin registry [Jerskey et al, 2011] and Minnesota twins study [Jocklin, McGue, and Lykken, 1996])
• Substance abuse (review of meta analyses [Agrawal & Lynskey 2007] – high heritabilities in the 0.6-0.75 range, 0 shared environment for alcoholism, cannabis use, cocaine, and heroin)
• Tobacco smoking (meta analysis [Li et al 2003] – shared environment impact on initiation, as explained above; no impact on persistence)
• “Sociosexuality” (promiscuity) (Australian twin registry study [Bailey et al, 2000])

All the major outcomes don’t seem to show any lasting impact from whatever the shared environment impact on educational attainment is. But, most damning of all, a large meta-analysis covering [Hyytinen et al 2013] the U.S., Australia, Finland, and Sweden has found that the shared environment impact on lifetime income is also zero! The very thing most hope education will translate into appears to depend more on the individual’s innate traits – “moxie“ – and luck – than any special benefit conferred by mere degree. Whatever shared environment influence there is on educational attainment, like so many other things, it doesn’t seem to matter in the long run.

OK, so you might be willing to accept that you can’t shape your child’s personality or values. You can’t control his major life outcomes. You can’t even control how much money he will go on to earn. But surely you can do something useful, like leave your children a lifetime of happiness, right? After all, I believe, and advise, that a parent’s key duty, after ensuring that their children grow up healthy and safe, is to ensure that each has a happy childhood. Surely that must count for something, too,? It does, in the form of fond memories of childhood.

For it turns out that overall life satisfaction as an adult has a high heritability and a shared environment impact that is also zero, as found in a Dutch [Stubbe et al 2005] twin study and in a Norwegian [Røysamb et al 2003] one, together with a combined sample size of over 12,000. One’s lifetime of happiness boils down to genes and to the fickleness of luck. Edit, 3/3/15: [A new meta-analysis of subjective well-being and life satisfaction twin studies with a combined sample of nearly 48,000 finds a significant (>35%) heritability and a zero shared environment. This study included twins reared apart, finding similar results in that sample. See Bartels 2015. ***End Edit***]

Some of you might wonder how I could be a parent and believe that my efforts in raising my child will not impact who he goes on to become. Well, I’ve long since known that it was out of my hands. He will be who he will be. It’s only my job to help him get there, and pass on the legacies of all those who came before him. I did all I could do: I married well. Beyond that it’s in the hands of “fate”.

The failure of parents to appreciably affect the outcomes of their children affirms Gregory Clark’s findings, and indicates that much of the transmission of status from one generation to the next is ultimately genetic in origin. Clark’s studies used several measures of status, and I haven’t covered them all here. Perhaps something reliably affected by the shared environment might yet turn up. I’m not betting on it though.

Almost certainly, throughout history, and across the diverse societies, there has been a huge amount of “noise” in the transmission of status, especially on the individual level and in the short run. The vagaries of the circumstances no doubt imbued good fortune on some and dashed the success of many others. But through it all, the thing that is at the root of continuity – DNA – remained the active ingredient to propagate lineages in their respective places through out the ages.

It is as it was said in the Richard Donner Superman films: “The son becomes the father, and father becomes the son.” That encapsulates the essence of the reality here. Underneath all the variability (much of which is driven by more or less random forces), there is a fundamental truth in those words.

As for the theme for this post, I can’t think of anything more appropriate than “Jaga’s Theme.” This tune symbolizes both survival and the passage of knowledge and a legacy from one generation to the next. It is perfectly fitting.