November 27, 2007

Brad DeLong versus Charles Murray

Berkeley economist Brad DeLong claims:

"If inherited genetically-based IQ were the source of the extra edge that the children of the rich get in our society, than we would expect a parent with 4 times average lifetime full-time earnings--say $200,000 a year--to have a kid with a lifetime average income of $51,500 instead of the average of $50,000. But it is not $51,500. It is $150,000."

To justify this claim, he cites a Rube Goldberg formula from a multiple regression study that I'll leave to you to ponder.

This is a good example of how complex model-building can wind up in the ditch without the authors or readers noticing where things went wrong. It's reminiscent of the baseball statistical analysis battles of the early 1980s between proponents of all-encompassing models like Pete Palmer and advocates of simpler studies like Bill James.

As a Bill James-style study of the impact of IQ on income, Charles Murray's 1997 follow-up to The Bell Curve remains stunning in its simplicity and power. Murray utilized an insight dreamed up by Sanders Korenman of the City University of New York and Christopher Winship of Harvard: "Compare siblings who have grown up in the same home, and with the same parents, but who have different IQs."

This means that we are comparing pairs with less genetic diversity than typical pairs of strangers, and with very similar "shared environments." (They may well differ in the mysterious "unshared environments" of random bumps on the head or whatever.)

Murray once again used the federal government's National Longitudinal Study of Youth that has been following 12,000 people (and their children!) since 1979. In 1980, the Department of Defense paid to have all of them take the military's AFQT IQ test.

The results look much different than Dr. DeLong would wish you to believe. Murray reported in the London Times in 1997 on his study of sibling pairs in the NLSY:

"To make the analysis as unambiguous as possible, I have limited my sample to brothers and sisters whose parents are in the top 75 per cent of American earners, with a family income in 1978 averaging 40,000 (in today's money) [all figures in the London Times article were in British pounds, but the exchange rate was different then, so just think of the incomes as comparative figures.]

"Families living in poverty, or even close to it, have been excluded. The parents in my sample also stayed together for at least the first seven years of the younger sibling's life."

In other words, this sample represents a utopian America where no child is poor, no child is illegitimate, and even divorce is limited. So, how much income inequality does this model America generate in the next generation?

"Each pair consists of one sibling with an IQ in the normal range of 90-110, a range that includes 50% of the population. I will call this group the normals. The second sibling in each pair had an IQ either higher than 110, putting him in the top quartile of intelligence (the bright) or lower than 90, putting him in the bottom quartile (the dull). These constraints produced a sample of 710 pairs."

So, the Brights had a median IQ of 117, the Normals 100, and the Dulls 83. The differences are a little over one standard deviation (15 points).

"How much difference did IQ make? Earned income is a good place to begin. In 1993, when we took our most recent look at them, members of the sample were aged 28-36. That year, the bright siblings earned almost double the average of the dull: 22,400 compared to 11,800. The normals were in the middle, averaging 16,800."

So, the Brights had incomes 33% higher than the Normals, and the Normals had incomes 42% higher than the Dulls. (It would be interesting to know if Normals with Bright siblings differed from Normals with Dull siblings.)

"These differences are sizable in themselves. They translate into even more drastic differences at the extremes. Suppose we take a salary of 50,000 or more as a sign that someone is an economic success. A bright sibling was six-and-a-half times more likely to have reached that level than one of the dull. Or we may turn to the other extreme, poverty: the dull sibling was five times more likely to fall below the American poverty line than one of the bright.

"Equality of opportunity did not result in anything like equality of outcome. Another poverty statistic should also give egalitarians food for thought: despite being blessed by an abundance of opportunity, 16.3% of the dull siblings were below the poverty line in 1993. This was slightly higher than America's national poverty rate of 15.1%.

"...The young people in our selected sample came from families that were overwhelmingly likely to support college enthusiastically and have the financial means to help. Yet while 56% of the bright obtained university degrees, this was achieved by only 21% of the normals and a minuscule 2% of the dulls. Parents will have been uniformly supportive, but children are not uniformly able.

"The differences among the siblings go far beyond income. Marriage and children offer the most vivid example. Similar proportions of siblings married, whether normal, bright or dull - but the divorce rate was markedly higher among the dull than among the normal or bright, even after taking length of marriage into account. Demographers will find it gloomily interesting that the average age at which women had their first birth was almost four years younger for the dull siblings than for the bright ones, while the number of children born to dull women averaged 1.9, half a child more than for either the normal or the bright.

"Most striking of all were the different illegitimacy rates. Of all the first-born children of the normals, 21% were born out of wedlock , about a third lower than the figure for the United States as a whole, presumably reflecting the advantaged backgrounds from which the sibling sample was drawn. Their bright siblings were much lower still, with less than 10% of their babies born illegitimate. Meanwhile, 45% of the first-born of the dull siblings were born outside of marriage."

This data can be found in tabular form at the bottom of Wikipedia's "Bell Curve" article.

Murray ends his "IQ and Economic Success" paper like this:

"People of different political viewpoints may legitimately respond to this presentation with policy prescriptions that are in polar opposition. In many ways, the Left has the easier task. These data are tailor-made for the conclusion that a Rawlsian redistributive state is the only answer. For its part, the Right must state forthrightly why it thinks that a free society that tolerates large differences in outcomes is preferable to an authoritarian society that reduces them. But though the answers may be different for those of competing political persuasions, the challenge is common to all. It is time for policy analysts to stop avoiding the reality of natural inequality, a reality that neither equalization of opportunity nor a freer society will circumvent."

My published articles are archived at iSteve.com -- Steve Sailer

26 comments:

Anonymous said...

Charles Murray: << How much difference did IQ make? Earned income is a good place to begin.>>

Murray didn't show that variation in IQ made the difference, he showed that it predicted the difference. The difference might have been caused to some extent -- large or small, who knows? -- by other variables correlated with IQ: conscientiousness, time preference, etc., which he didn't even consider. Murray's claim outruns his evidence by far.

Anonymous said...

What I find the most interesting are the Idiocracy-style notes about marriage and children. What's the explanation? I would guess a large part of it is that "dulls" are less likely to use birth control, or perhaps less likely to use it properly. And they might be less likely to get an abortion -- having a child out of wedlock at 19 isn't really getting in the way of anything, since so few of them are going to graduate from college anyway. I certainly don't think it means that dumb people are more attractive to the opposite sex than smart people.

Luke Lea said...

A Rawlsian redistributive state does not have to be authoritarian, with the right tools.

DeLong's faith in fancy statistical analysis reminds me of the Hedge Funds, who seem to think there is no longer any relationship between risk and rate of return. I am waiting for the crash. Hopefully, Harvard's endowment will be reduced by 95%.

Anonymous said...

Steve,

Murray may be right, but as your general anti-plutocrat populism should be sympathetic too, the first guy's analysis aren't for the merely successful, its for the top couple percent in income. It may be that at those levels, parental help does outweigh the meritocracy of IQ. As you tirelessly point out, look at W.

So they both could be right, Murry and the other guy, depending on the shape of the meritocracy curve. It could be that in the middle regions meritocracy reigns, but at the top and bottom privelage or privation overwhelm meritocracy. The one analysis doesn't disprove the other, without further study.

Tino said...

its not just the problem of models. Its the problem of liberal stupidity: idiot delong does not understand that he is comparing the mean of the population (rich) with a single point. Income is very noisy, if you average out one you HAVE to average the second.

Anonymous said...

Ummm, Delong has put his finger on the huge scientific problem with your IQ fixation, and you can't handle it. The problem is that IQ ACCOUNTS FOR A VERY SMALL FRACTION OF ADULT EARNINGS VARIANCE IN A STANDARD REGRESSION. Period. Let me repeat it again so this well-established empirical fact can penetrate your consciousness. IQ ACCOUNTS FOR A VERY SMALL FRACTION OF ADULT EARNINGS VARIANCE.

That is, even though IQ is certainly correlated with adult earnings, a SD change in IQ has only a small effect on adult earnings, especially once family background and education are taken into account. See e.g. Zax and Rees, November 2002 Review of Economics and Statistics, although this finding has been established hundreds of times in numerous data sets.

Against this, you put a ridiculous and unscientific exercise by Murray, in which he distorts data until he gets the finding he wants. The "Brights" and the "Dulls" are not in the same families at all -- nothing about them is necessarily comparable. "Dulls" could all be from families with HS graduate parents and "Brights" could all be from families who have parents with advanced degrees. Their IQ is measured in their late teens, after all kinds of things are different about them. For all you know, the "Dulls" all have criminal records already! There's a reason you have to go to a London semi-tabloid newspaper to find a discussion of Murray's, ummm, "findings".

Sorry, dude, but I've been waiting for you to grapple with this issue for years and this just doesn't cut it. Science isn't so much fun when it contradicts your preconceptions, is it? Basically, at this point it's you who is the one who is running away from science. Maybe the experience will help you understand how PC types are tempted into science denial as well.

MQ

P.S. like most good social science findings, the limited IQ/earnings relationships accords with common sense. Steve, I'm guessing you have a pretty high IQ. Ever met anybody who made significantly more money than you but seemed, well, not the sharpest tool in the shed? I thought so. It's a common experience, because once you reach a reasonable level of intelligence a lot, and I mean a lot, of other personality characteristics have a bigger influence on earnings.

Unknown said...

especially once family background and education are taken into account.

I love this, as if it were perfectly reasonable to "control for" a proxy of the (heritable) trait you're studying; "see, if you control for the length of the pants in people's closets, all races are the same height!" (sorry for robbing you there Steve...)

Anonymous said...

It's a rule that nearly never fails: the more complicated the analysis, the weaker the argument.

Anonymous said...

Anon snarkily says:


Ummm, Delong has put his finger on the huge scientific problem with your IQ fixation, and you can't handle it. The problem is that IQ ACCOUNTS FOR A VERY SMALL FRACTION OF ADULT EARNINGS VARIANCE IN A STANDARD REGRESSION. Period. Let me repeat it again so this well-established empirical fact can penetrate your consciousness. IQ ACCOUNTS FOR A VERY SMALL FRACTION OF ADULT EARNINGS VARIANCE.

That is, even though IQ is certainly correlated with adult earnings, a SD change in IQ has only a small effect on adult earnings, especially once family background and education are taken into account. See e.g. Zax and Rees, November 2002 Review of Economics and Statistics, although this finding has been established hundreds of times in numerous data sets.


Delong seems to have claimed only a 3% increase for however many SDs he is claiming ...

But the paper you referred to seems to say:


For the United States, Zax and Rees (2002) estimate conservatively that an increase in IQ of 1 SD is associated with an increase in wages of more than 11 percent, falling to 6 percent when controlling for other covariates.


Found here.

So, was Delong only talking about a half SD or what?

Anonymous said...

Steve: don't refer people to Wikipedia, it can be edited faster than you can say 'political correctness'.

Seriously, you really think IQ is responsible for most of the earnings gaps? I believe you've blogged about ethnic mafias and family connections quite a few times. Not that it doesn't help but there are many other factors.

Anonymous said...

Dull women are more willing to have children because having children is unpleasant and career-destroying, as I pointed out in the other thread, and they are less able to predict the first and less affected by the second.

Normal societies afford status to a woman fulfilling her reproductive function, high status to a woman doing so in a way that will further the claims and existence of the patriarchy. Ours does not. We have an inverted version, however, whereby a woman who reproduces and rears children in such a way as to attack the patriarchy (a single mother living on the tax revenue created by productive men, a divorced mother destroying the wealth created by her husband) is given a sort of status.

The number of normal or bright women who have cultural values that defeat this is going to be low.

Anonymous said...

I'm gonna do the Sancho Panza thing here and stick with the low-level knowledge that I have a practical familiarity with. Murray has just described how 2% of the "Dulls", despite having an IQ of 90 or less, nonetheless got college degrees. IOW, readers, you can get a college degree today with an IQ of 90. Now it so happens that given the experience I have, I already knew that; however, a lot of people, especially people who aren't Americans, have trouble believing it. Is this a waste of time and resources or what? If you're not sure what kind of a degree those 2% can get, two words -- teachers college.

Luke Lea said...

If I am not mistaken, anonymous MQ in his snotty way is confusing the predictive power of a single individual's IQ for his individual future income. If the correlation if .40 (as best as I can remember?), well, a scatter plot of a .40 correlation is pretty spread out. It tells you almost nothing.

On the other hand, for group of a thousand individuals of a given IQ, a .4 correlation can tell you a lot about their average income, when you compare to other groups who differ by one standard deviation.

It is all about averages over groups, not single individuals.

This is an enlightening distinction Steve made in one of his posts a year or so ago. It was new to me then, but once you make the point it is easy to grasp. It bears repeating.

Anonymous said...

anonymous, Murray's study shows large achievement gaps between the dull and normal siblings of the same family, and between the normal and bright siblings. And this even when the poorest 25% of families are excluded, as are single parent families to an extent. It means that IQ makes a big difference.

A comparison between dull and bright siblings of the same family would be interesting, of course, although there may not be a whole lot of such pairs.

the other anonymous: if Latinos had an average IQ of 120, the US would long ago have been conquered by them.

Anonymous said...

Anon argues that opportunity costs for dull women are lower wrt childbirth than smarter women. Very likely so. Also over-estimation of how long fertility can be extended (for smart women) and another important factor:

Smart women know having someone else's kid will repel the A-lister they want.

Whereas dull women often try to snag their own A-lister (lower down on the socio-economic scale) by having a kid with them, overestimating the degree of control that will give them. Laci Peterson, etc.

Social conformity is also higher among women, even smart women. Thus the need to conform to group standards of delayed and fewer children.

Dull women for example, will pass by a perfectly good plumber or carpenter in favor of the local bad boy (say a fertilizer salesman like Scott Peterson). While smart women endlessly pursue Mr. Big. This means more than just IQ is being selected out.

1. There is a genetic bottleneck -- only a relatively few men are impregnating women.
2. Other heritable traits like cooperation, trust, reliability, loyalty are being selected out in favor of aggression and dominance through sexual selection (women's choice).

Ron Guhname said...

"The problem is that IQ ACCOUNTS FOR A VERY SMALL FRACTION OF ADULT EARNINGS VARIANCE IN A STANDARD REGRESSION."

In addition to the other criticisms, liberals shout from the rooftops when they find, for example, a statistically significant relationship between race and some form of discrimination, even though the variance explained is 1 or 2%. Even if it were true that IQ did not explain a large amount of variation, it impacts so many areas of life, it still would be very important. Plus, you can't whine about unimpressive effects cross-nationally (they're huge), so you're the one who has to deal with unwanted facts.

Anonymous said...

In 1977, the wealthiest 1 percent of Americans owned 26% of all of the country's financial assets. By the early nineties, the top 1% Owned more than 48% percent of these assets. Did the ultra-wealthy become astronomically smarter during this decade? Or perhaps the increasing stratification had more to do with Carter, Reagan, and Bush administration policies of cutting social services for the poor and middle class, while at the same time cutting taxes for the wealthy and drastically increasing military spending, which of course benefits wealthy individuals and corporations (also, incidentally, the primary contributors to the campaigns of these administrations; funny how these things work out)?

On a similar note, let's just take a look at blacks as a group for a second. From 1960 to 1973 (during the expansion of the "welfare state" to blacks, after it had been previously restricted to whites), black poverty was cut in half. After 1973 (when the "welfare state" programs failed to keep pace with inflation), the drop in black poverty and unemployment stopped falling. Did blacks become much, much smarter between 1960-1973, and then get dumber after that? But blacks have been doing better on standardized intelligence tests over the past 20 years, yet black unemployment and poverty rates have fluctuated during this period with out any apparent regard to changes in intelligence. This is also true with regards to out-of-wedlock births, crime, and many other things said (or implied) to be strictly determined by intelligence. Black crime increased in the mid-60's, then started falling in the late 70's up to today. Illegitmacy rates have been dropping since the early 90's. Black and white fertility rates have dropped so much in the past 20 years that they are now pratically identical. I really don't see any evidence that "dulls" are having more children today than "brights," as some commenters have suggested.
(And in fact, the rise in out-of-wedlock births among blacks from the 60's to the 90's had less to do with unmarried black women having more children than with married black women having LESS children, so that the total share of out-of-wedlock births increased relative to the total number of black births, without increasing very much in absolute terms.)

Finally, poverty rates among the elderly dropped dramatically since the institution of the New Deal in the 1930's. Is that because old people have been getting smarter, or does it have more to do with the New Deal?

IQ may be correlated with economic success and failure, but, contra Murray and others, social and economic policy can make a huge difference.

Anonymous said...

What exactly is the point of comparing IQ and economic success? To prove that blacks' lack of success is due to blacks' low IQ?

Aside from that, the more relevant question for the welfare of the community is the correlation between IQ and productivity.

Anonymous said...

IOW, readers, you can get a college degree today with an IQ of 90.

A hundred years ago, Harvard and Yale awarded plenty of degrees to men with IQ's of 90 or below.

Anonymous said...

Folks, all you are demonstrating is either your inability or unwillingness to understand social science. The evidence on IQ and earnings is every bit as strong as the evidence on race and IQ. Yet when it does not satisfy your preconceptions you run away from it just as fast as liberals run away from racial IQ differences. Not to mention totally ignoring the obviously cooked nature of the Murray example.

To take the "criticisms" one by one:

I love this, as if it were perfectly reasonable to "control for" a proxy of the (heritable) trait you're studying

Regression analysis is specifically designed to handle this objection, in fact it is the whole reason for the existence of regression. Regression is probably the most widely used tool of the social sciences, and one of the most basic (taught in early undergrad stats). If you have to reject or misunderstand it to make your point, then why should anyone take you seriously?

Also, it is plainly ridiculous to claim that family background and education have no other role in human life as anything other than as a "proxy" for IQ. Finally, as the next commenter points out...

For the United States, Zax and Rees (2002) estimate conservatively that an increase in IQ of 1 SD is associated with an increase in wages of more than 11 percent, falling to 6 percent when controlling for other covariates.

Exactly my point, thank you for making it. Delong was doing his own calculation of the role of IQ in inherited earnings, not replicating Zax and Rees or the mountain of other evidence on this matter. But the reason Delong's calculation shows a small effect is that *the underlying relation is not strong*. The correct regression controls for other covariates, but even when you don't a 1 SD increase leading to an 11 percent growth in earnings is not a strong relationship. It implies that if people with an average IQ of 100 average earnings of $50,000 per year, people with an IQ at the 90th percentile will average earnings of about $58,500. Does that seem like a big difference to you? And of course, the correct difference once covariates are accounted for is only about half that.

If I am not mistaken, anonymous MQ in his snotty way is confusing the predictive power of a single individual's IQ for his individual future income....It is all about averages over groups, not single individuals.

Quite so, I'm far from confused about that. All of these methods use average values of IQ and other predictors to forecast average group values. It's still quite important to know that in general knowing someone's IQ alone tells you very little about what their earnings will be (perhaps unless they are far out at the tails).

I should note, though, that IQ alone clearly cannot explain group average differences. Black males earn an average of $32,000 per year, white males earn an average of $48,000 per year. That's a 50 percent difference, as opposed to the 6 to 11 percent per SD difference cited above. And that's likely a significant underestimate of the difference, since the many black males in prison aren't counted the true difference is over 50 percent.

liberals shout from the rooftops when they find, for example, a statistically significant relationship between race and some form of discrimination, even though the variance explained is 1 or 2%.

I have no problem criticizing "liberals" when they do that. For instance, the liberal fuss about the male/female earnings differential is less relevant once you realize that this differential becomes quite small (less than 10 percent) once other relevant covariates such as experience are controlled for. Incidentally, I'm guessing the people above who thought that it was ridiculous to control for other influences on earnings in the case of IQ would find it very correct to control for these influences when it comes to the effect of sex on earnings. Just another example of how peoples' attitudes toward science change depending on what political belief is at stake.

Plus, you can't whine about unimpressive effects cross-nationally (they're huge)

That's another debate, where I'm admittedly not as familiar with the literature. What I'm pretty sure is happening there is that IQ is proxying for a ton of other differences between the societies that are more important; when those differences don't exist (as in within-country earnings comparisons) then you see how good IQ actually is as a predictor of success in life...it tells you something, but not really very much. If I wanted to lay a bet on how well someone would do as an adult based on just one piece of information, I'd much rather know how much their parents earn than their IQ.

All of this doesn't even get to the more fundamental problems with IQ as a construct. Truth is, there are plenty of other psychometric constructs and personality measurements that are just as useful predictors as IQ.

OK, it's been fun, thanks to all.

MQ

Anonymous said...

evilneocon, you continue to be wrongy mcwrongerson on your bizarre sexual fantasy-theory about a small number of men impregnating harems. They're only banging the ladies, not leaving buns in their ovens.

Anonymous said...

This is hardly a "Rube Goldberg" formula. It's simple analysis of correlations. Steve should really really learn basic statistics.

Anonymous said...

a Rube Goldberg formula from a multiple regression study that I'll leave to you to ponder.
Come on now Steve, you're just gonna dismiss something because you don't understand it?

Anonymous said...

IQ does a much better job of explaining differences of average earnings between races than it does between individuals.

Low IQ predicts poverty much better than high IQ predicts wealth.

Murray, who is an advocate for the importance of IQ (and I certainly agree that it's important), has unfortunately exaggerated the extent to which IQ predicts income.

A slightly above average IQ (say 105) is sufficient for earning a high salary if you have family to place you on a high paying career track.

On the other hand, a person with genius level IQ will often wind up being an unemployable person with a PhD in History.

Anonymous said...

MQ

Could you explain your objection to Murray's "obviously" cooked books please. I admit, I'm a layman, but I try to be an honest one and am open to your critique if you'll explain it. I am guessing its that Murray takes pairs with shared environments, but then compares the averages of the whole groups, who don't share environments beyond the pairs. Maybe if he compared mean and median differance in income for each dull/norm or norm/bright pair?

Anonymous said...

th said

You are missing the point of my thought experiment.So I'll try and be clearer this to around to avoid any confusion.

During the next BIG DEBATE on the MSM, it would be very easy to sidetrack a discussion about the harmful effects of non-white immigration-asian included-on the majority Euro-American population.
If certain immigration reformers raise this issue, the debate will degenerate into a vapid discussion of the finer points of psychometrics. Only very small fraction of a percent of the American population would understand and be intested in the psychometric debate.

However, the Euro-American viewing audience would be very interested hearing a 40 year Euro-American passed over for promotion for a 25 year old legal immgrant from hindu.

On the last Phil Donhaue show this happened. Only a few minutes after a 40 year old Euro-American male complained on national tv of being passed over for promotion for a 25 year legal immgrant from India, Jared Taylor sidetracked the discussion by raising the Race and IQ score issue. Then he bragged how he wasn't a racist because he thinks asians are smarter than Europeans. What garbage. You know the saying with friends like this who needs....Much better to say to argue against immigration in terms of the racial transformation of American which goes-hand in hand with the economic dispossession of Euro-America.