StatSquatch

"An explosion without a bang"

2011-10-09T19:25:00.000-07:00

The International Journal of Epidemiology has republished a paper on the marginal importance of shared environment by Plomin and Daniels: "Why are children in the same family so different from one another." Neven Sesardic wonders why such a supposedly revolutionary change in thinking has had such little effect:

However astonishing this empirical discovery was (as it definitely was), it did not make a splash. Or, to mix the metaphors, it was like an explosion without a bang. The lack of reaction to such an amazing result is itself amazing. It is not just that this truly remarkable finding was not widely reported in newspapers, magazines or popular science publications.The event was also largely ignored in many relevant parts of psychology. No re-examination there, no questioning of the fundamental presuppositions, no paradigm shift.

Despite mountains of the evidence, juvenile delinquents and their political enablers continue their officer Krupke defense for bad behavior. Be it accelerated evolution or complications in the "Out Africa" model we can expect more quiet explosions.

Leftist Eugenics??????

2011-08-28T07:22:00.000-07:00

Previously, I dismissed the possibility of leftist Eugenics but having just read the retard hating philosopher Peter Singer's A Darwinian Left I am having second thoughts.

Singer writes that the failures of the left are due in part to their ambivalence towards Darwin and this ambivalence was present from the beginning. For example, the Marxist approved of the anti-clerical implication of the theory of evolution but insisted that it was no longer relevant for discussions of human nature. Singer writes,

Plekhanov, the leading nineteenth-century Russian Marxist, followed Engels in holding that 'Marx's inquiry begins precisely where Darwin's inquiry ends' and this became the conventional wisdom of Marxism. Lenin said that 'the transfer of biological concepts into the field of social science is a meaningless phrase." As late as the 1960s, school children in the Soviet Union were still taught the simple slogan: 'Darwinism is the science of biological evolution, Marxism of social evolution.'...It is intriguing how two very different ideologies - Christianity and Marxism - agreed with each other in insisting on a gulf between humans and animals, and therefore that evolutionary theory cannot be applied to human beings.

Singer suggests that the Left leverage human altruism to create a better world. Then he gets a little weird:

We are the first generation to understand not only that we have evolved, but also the mechanisms by which we have evolved...Hegel portrayed the culmination of history as state of Absolute Knowledge, in which Mind knows itself for what it is, and hence achieves its own freedom. We don't have to buy Hegel's metaphysics to see that something similar really has happened in the last fifty years...to those who fear the adding to the power of government and the scientific establishment, this seems more of a danger than a source of freedom. In a more distant future that we can still barely glimpse, it may be turn out to be the prerequisite for a new kind of freedom.

So we have Hegel and more power to government to achieve true freedom. What could go wrong?

Harvard Replicates Sailer's PISA Results

2011-08-24T20:18:00.000-07:00

Eight months later, the Harvard Kennedy School has replicated and expanded on Steve Sailer's PISA analysis ethnicity. The percent of whites who are proficient in math is 41.8 which is better than the scores from France and Iceland and comparable to Australia. New Zealand, and Germany. Reading scores are even better.

They agree with Sailer that Whitey's reading scores are adequate but insist that higher math scores are required since Whitey's percent proficient is "well behind" Germany's (45%) and has as gap of "over 25%" with Finland (57%). Note, in the Finnish comparison they use a simple but effective trick to inflate a difference by ignoring a 15% absolute difference to report a larger relative difference.

They try to make a connection between economic growth and math proficiency and present a graph without Japan. Gee, why wouldn't they want to include Japan?

The data presentation is good but would it be a huge strain on Harvard Kennedy School to calculate a confidence interval or something so we could tell if Ireland is really less proficient than Lichtenstein? Someone may want to know!

Cousin Humping is Good for Science!

2011-08-15T20:36:00.000-07:00

So if Kevin Mitchell is right and rare variants are important for IQ then a good place to look is retards (i.e., Intellectually Disabled). From a review paper by Kaufman (2010) on the causes of non-syndromic intellectually disabled (NS-ID):

Under the rare variants model of causation, large consanguineous families are particularly useful. Rare mutations, which may be identified in such families, can provide us with information about the types of etiological aberrations in NS-ID as well as the genes and relevant pathways that may be essential for normal neuronal functioning.

People with NS-ID have and IQ < 70 without co-morbidities. So HBD chick should lay off the cousin bangers.

Ian Deary's new paper

2011-08-15T20:27:00.000-07:00

There are many comments on Ian Deary's new paper. Among the big time news sites: The Scientist gives the best overview, the Washington Post is a little shallow but balanced while the Atlantic Monthly is the home of diehard liberal creationist.

Razib gives a more in depth review and notes that the paper applies the same method to IQ that Yang (2010) applied to height. The method is a way to show where to look for the missing heritability of complex traits. It fits the phenotype as a function of thousands of SNPs modeled as a random effect thereby yielding an estimate of the lower of heritability with out some exotic data of twins, half sibs, or adoptees. The method must have been controversial since several authors wrote a commentary, Visscher 2010, explaining the method in detail in a separate journal.

In the comments section of the Razib's blog Kevin Mitchel does not agree with the paper's conclusion that the this means the IQ genes are common with small effects and claims, rather, that they could uncommon with large effects. Does it really matter? Either way the relationship between genetics and IQ is going to be complicated and never complete until the biology is understood.

Some look forward to the study will not unleash the HBDapocolypse: that glorious day when the sky parts and irrefutable evidence of inherent human differences spreads throughout the land. Maybe Steve Sailer replaces David Brooks at the NYT or Half Sigma becomes a Federal judge. Do not hold your breath it is going to take many studies.

Oh No...

2011-05-25T22:17:00.000-07:00

Some legitimate blogger is linking and criticizing my post on Duckworth's sloppy meta-analysis and threatening my well-deserved obscurity. Her post was too boring to finish but she seemed to imply that I would not submit my analysis to a peer reviewed journal because of my blog roll of bad guys or something. That is incorrect. As you can tell from this sorry excuse of a blog, I am very lazy.

The only reason I posted on Duckworth's tedious meta-analysis is that the results are driven by a paper written by someone who actually pleaded guilty to a research related crime. That was too hilarious to pass up. That and the fact that you can publish a meta-analysis in PNAS without a forest plot.

A bad apple in Duckworth's IQ-Motivation meta-analysis?

2011-05-08T12:30:00.000-07:00

A new paper by Duckworth et. al. (2011) that links motivation to performance on IQ tests has interested Sailer, Bryan Caplan, and Tyler Cowen. The regular media claimed the study weakens evidence around IQ testing. It does not. The study supports the predictive power of IQ tests but claims that they are a composite measure of intelligence and motivation. If true, if motivation is more malleable, this could help explain the Flynn effect and make it easier to find genes associated with intelligence. But after getting reviewing the paper's meta-analysis of randomized studies of motivation and IQ, I have my doubts about the robustness of "average" 0.64 improvement in SD due to incentives.

The paper's meta-analysis of the 46 studies was very good but they neglected to make a "forest plot" that graphs the effect sizes and confidence intervals from each study. So I did it. The size of the effect circle is proportional to the inverse of the effect's standard error, i.e., bigger studies that get more weight have bigger circles.

Two things are apparent: the study outcomes are highly variable, i.e., they are heterogeneous, and there are only three large experiment (2, 3, and 4 in the graph) that showed motivation leading to a large improvement in IQ score and hence are very influential.

The three experiments were run on special ed kids and written up in one paper by Bruening and Zella (1978) while the first author was at the Oakdale Center for Developmental Disabilities. A few years later Bruening admitted to fabricating data for other experiments on retarded kids at the same center and the case became a text book example of scientific fraud. Although there were no allegations of fraud on the 1978 paper, I re-ran the meta-analysis with out Bruening's data and found the the estimate (using the random effects model) was now 0.48 SD and was no longer significant ( p = .07). This makes me uneasy about accepting Duckworth's results without further replication.

GE interactions will disappoint liberal creationist

2011-05-01T12:05:00.000-07:00

A blogger at the Huffington Post, Scott Berry Kaufman, Ph.D. (h/t Gene Expression), has a "nature-nurture" posts discussing the importance of gene by environment interactions :

Therefore, gene-environment interactions are understood to drive the development of all of our characteristics ... every trait develops through the interplay of genes and the environment. Nature and nurture are complementary, not at odds.

.

The importance of GE interaction is an empirical matter but I think their possible importance serves an emotional need for reformed liberal creationist by making environments important again. The think "complicated" GE interaction will justify their grand interventions in the name of equality? They will be disappointed since important GE interactions can have anti-egalitarian implications.

As an example take a preliminary finding from Robert Plomin's group. Docherty et. al. (2010) defined a SNP score associated with mathematical ability and , although not confirmed by a second study, they (Docherty 2011) evaluated the effect of an interaction of the SNP score with a measurement of "teacher negative" on a mathematics achievement (measured in SDs below):

The result are significant at the 5% level but did not survive the Bonferronni adjustment, but it appears less genetically gifted do not do well with with positive teachers who should be reserved for the genetically gifted. In the future, a GE interaction could justify assigning the dimmer students to meaner teachers. This educational equivalent of personalized medicine may be a good idea but it will disappoint the liberal creationist.

Leftist Eugenics

2011-02-15T20:05:00.000-08:00

Ed West, a blogger for the London Telegraph, is predicting that left will eventually turn to eugenics. He focuses on environmental issues and does not mention a the heritability for favorable traits but he does make one good point

"...we’ve become so completely used to accepting the idea that the state should manage the lives of whole sections of society, it does not take a logical leap to accept eugenics."

The therapeutic state is powerful enough to implement useful eugenic policy. John Glad's pamphlet also makes a good case for an eco-friendly eugenics that encourages the intelligent and altruistic individuals (i.e., liberals) to breed. I predict the left will embrace eugenics right after Half Sigma is is sworn in as a Supreme court justice. They have the way but not the will.

Are you a Bayesian Racist?

2011-02-10T20:24:00.000-08:00

Uhlman et. al. (2010) coin the term "Bayesian Racism" as "the belief that it is rational to discriminate against individuals based on stereotypes about their racial group" and they offer the following test to identify them:

The authors entertain the idea that stereotyping may be "epistemically rational" but then conduct a poll that purports to show that many "Bayesian Racists" are not motivated by accuracy and so are indeed irrational. The description of the poll methodology is remarkably thin. For example, no non-response rate is presented and there appears to be no validation of the Bayesian racist scale. Still, I am relieved to report that not all apparatchiks are as innumerate as Newsweek's Shankar Vedantam.

Update: The Occidenalist has further comment.

Test for test bias biased?

2011-01-31T20:57:00.000-08:00

Jlovborg posted an interesting paper in the comments by Aguinis et.al (2010). The paper challenges the systematic overprediction of minority performance that I mentioned in this post. The paper is technical and poorly presented and I would not normally blog on it. However, the main author is a big shot who co-authored a amici curiae brief in the Ricci case contending that New Haven had not properly validated the fire fighter test. That is, he wrote against Frank Ricci. This paper was well covered in the press since it asserts that the statistical procedures used to assess tests bias may themselves be biased.

The most common method to assess test bias in two populations is to assess via linear regression how well tests scores (e.g., SATs) assess later outcomes (e.g. GPAs). If the relationship between test scores and the later outcomes are the same in the two groups then the test is considered "unbiased." Most tests find the minority overprediction phenomenon shown in the graph below. That is, both groups have the same slope but the low scoring group has a lower intercept indicating a lower level of achievement associated with the same test score.

The paper challenges the overprediction results by claiming that the methods uses to detect differences of slopes and intercepts between the groups are inadequate. Specifically, using both analytical methods and simulations the paper claims the common tests are

Underpowered to detect slope differences
Overestimate differences in the intercepts (m1 - m2).

Point one, if true, has no methodological implications. Tests creators just need larger samples (say n > 500) to rule out significant slope differences. The second point, if true, is a bigger problem since it implies that regardless of sample size you cannot properly estimate the intercept difference.

I am not an expert on psychometrics but I am skeptical of the veracity of 2. Without getting into the technical details, the authors test for the difference in intercepts (i.e., m1 - m2) by looking at difference in R-Squared instead of the coefficient estimates themselves. The simulations are parameterized in terms of R-squared and they provide some perplexing results as shown in their graph below.

Panel C below shows that the type I error increases as the difference in the intercepts increases. Since a type I error is not defined if the intercept difference (m1-m2) is not zero they something else is being tested. Panel D shows that the type I error increases with the sample size which implies that the test used is useless and the authors do not name a replacement.

Overprediction of minority performance

2011-01-24T21:30:00.000-08:00

In the The Affirmative Action Hoax, Steven Farron notes the "universal phenomenon" of aptitude tests over predicting the performance of lower performing groups. That is, members of higher scoring groups have better outcomes than members of lower scoring groups with the same test scores. For example, blacks with a given LSAT will have higher failure rates on the bar exam than equivalent white test takers. The same phenomenon has been observed comparing entrance exams of Arabs and Jews in Israel and whites and Asians on mathematical aptitude tests. Sailer even found a Federal Reserve study that showed that blacks had a higher risk of foreclosures than whites with the same FICO scores.

Borsboom et. al. (2008) showed that this "universal phenomenon" is a statistical artifact dependent on two conditions:

Measurement Invariance: The test is measurement invariant in that the mathematical relationship between the test score and the latent trait
Differential distribution: the mean of the distribution of the latent trait that the tests is trying to estimates differs between the two groups.

La Griffe du Lion had a similar post about this but I recommend Borsboom's more traditional treatment. The paper has a graph similar to that below that shows a stylized example of a test that is a simple function of the latent trait and that a test score of 50 is required for acceptance into a program while those with a latent trait level above 50 are suitable.

The ellipses represent the joint distribution of the test score and the latent attribute for low and high achieving groups. Due to measurement invariance the distributions only differ in the means. Obviously more members of the high scoring group will be true positives and more of the low scoring group true negatives but the two error types are more important since the accepted / not suitable candidates are at risk for failure and any rational institution tries to minimize this type of error.

Borsboom et. al. prove that the conditional probability

P(accepted | not suitable, low achieving group ) > P(accepted | not suitable, high group).

So using the same the cutoff for both the high and low scoring groups leads to disproportionate rate of failure among the low scoring groups. Of course, with affirmative action a lower criterion is used for low scoring groups with predictable results.

"Diversity" and the higher education bubble

2011-01-22T15:59:00.000-08:00

In Steven Farron's excellent book The Affirmative Action Hoax he speculates that colleges made ridiculous court arguments to allow them to apply lower standards to blacks and Hispanics in the name of diversity (e.g., the Bakke case) out of fear that the courts would force them to abandoned achievement tests as the courts have forced employers to do (e.g. the Griggs case). This means that universities and the military are the only entities in the US that can use g-loaded standardized tests which vary be demographic groups. This is a huge economic advantage since admittance to a selective college is a reliable and legal method that employers can use to discriminate between applicants. This is why I do not expect the higher education bubble to pop completely anytime soon. A cheaper way to do this sorting would be to make junior colleges more selective, e.g., only allowing whites with an IQ higher than 100 to enter while making allowances for blacks and Hispanics.

The War on Thai children and Bayes' Theorem

2011-01-17T19:33:00.000-08:00

Last year Sailer commented on an article from Slate's Shankar Vedantam who wrote this curious paragraph

"People in Thailand will associate white American tourists with pedophilia even though many more acts of pedophilia are committed by Thais. But white Americans are a minority in Thailand, as are acts of pedophilia. So you will hear Thai people shout until they are blue in the face about individual anecdotes showing white Americans who are pedophiles. (The same is true of gay men and pedophilia in the United States.) "

Thais may know Bayes' theorem and that the facts "that many more acts of pedophilia are committed by Thais" and "Americans are minority in Thailand" do not imply that nationality is irrelevant to gauging the risk of bad touching. Let P(BT|W) and P(BT|T) be the probability of being a bad toucher given that an individual is white or Thai respectively. Bayes' theorem gives

and

where P(W|BT) and P(T|BT) are the probabilities of being white or Thai respectively given that you are a bad toucher and P(T) and P(W) are the probabilities of being white and Thai. Using basic algebra you get

So Thai children should be more wary of whites than Thais if the proportion of whites among bad touchers is higher than their proportion in the general population. I do not know if this is true but it seems reasonable. So Thais may be perfectly rationale by stereotyping whites and by ignoring this possibility Shankar Vedantam appears to be either innumerate or a NAMBLA shill.

Neven Sesardic gives a more academic treatment of Bayes rule and rational stereotyping in Making Sense of Heritability (page 218) but his example is not as much fun.

Charles Murray interview

2011-01-12T11:06:00.000-08:00

Here is Charles Murray speaking at the International Conference of Intelligence Research in 2009. About 40 49 minutes in he advises the researchers to write up their results in obscure technical language to avoid controversy. Research obfuscation will not help the cause of good governance in multiethinic societies but it does mean more fun for the Statsquatch.

ISIR 2009 Distinguished Interview: Charles Murray from Timothy Bates on Vimeo.

The neocons are linking the Arizona massacre to ... Philip K. Dick?

2011-01-10T10:15:00.000-08:00

On the Commentary blog John Podhertz is linking the Arizona massacre to the late science fiction writer Philip K. Dick . Podhertz opines

"He may, in other words, have found his intellectual solace not in political ideology of any sort but rather in the false-reality fantasies of writers like Philip K. Dick, who all but invented a science-fiction genre about how the powerful have the rest of us living in a dream world in which we are manipulated. The most commercially popular version of this worldview is The Matrix.."

Like Loughner, PKD had some mental health issues, although it is not clear he was schizophrenic. However, he never would have had anything to do with dreck like The Matrix if only because he knew the second law of thermodynamics. The Matrix was inspired by continental philosophers like Jean Baudrillard. So Heidegger should take the blame for the shooting.

Stylized examples of departures from measurement invariance

2011-01-08T23:21:00.000-08:00

I had a hard time of finding stylized examples of departures MI probably since it requires the conceptually slippery latent variables. So I made up some hypothetical examples inspired by some graphs in Jelte Wicherts' thesis. The graphs make the simplifying assumption that a given test score is a linear function of the single underlying latent trait. The latent trait is not observed but the slope and intercept are derived from factor analysis. If MI held in these examples the lines would be equal for the two groups.

No Comprende: Suppose a "culturally unfair" test of verbal intelligence is given to two groups: native speakers and foreigners. The confused foreigners answer at random so that test cannot distinguish the bright from the dim and measurement invariance does not hold.

Privileged: A test is given to the oppressed and the privileged. Because of test anxiety caused by the menace of the privileged the oppressed student at all levels of latent ability perform exactly 15 points less than the privileged students of the same level of ability. The fair thing to do is to give the oppressed a 15 point boost.

Differential: In this case the test have different characteristics for each set of subjects. The test is biased in favor of A (B) for those with a latent ability less (greater) than 30.

Stereotype threat and measurement invariance

2011-01-06T16:50:00.000-08:00

Stereotype threat (ST) is a popular explanation for the observed black - white (B-W) IQ and achievement of gaps.  Although I think the chance of ST explaining a large magnitude gap is low,  pigeon hole menace advocates have some interesting studies and investigating them is worth the work. This post will discuss the relationship of ST to measurement invariance (MI).   Briefly, MI implies that for two groups the functional relationship between a test score (e.g. IQ ) and the the latent variable the score measures (e.g. intelligence) is the same. It is the most important psychometric property without a Wikipedia page so this is the simplest introduction I can find.

ST Studies: The hundreds of ST studies (e.g., Steel and Aronson 1995)  test whether black (or women or white) test takers feel more performance pressure than whites (or men or Asians) because of anxiety about being pigeonholed.   The experiments all have more or less the same fairly rigorous (and absolutely hilarious) design. Black and white subjects are randomly assigned to two test taking groups. The experimental group is subjected to an ST (e.g., the proctor unfurls the confederate battle flag) that should only affect blacks. The gap between the two groups is compared using standard methods (e.g. ANCOVA) and more often than not the gap is larger in the experimental group.

Errors in Analysis: Wicherts (2005) criticizes the use of ANCOVA models that test for the ST effect while controlling for the students individual characteristics (e.g. SAT scores).  His primary concern is that the effect of ST can vary by the student's ability so the the assumptions of equal regression weights may not be met and is not tested.  He then suggests using Multigroup Confirmatory Factor Analysis (MGCFA) to test for MI. Wicherts et al (2005) applied MGCFA to three ST experiments and found that ST could lead to he loss of MI even when the ANCOVA showed little difference between group averages. These results have been (correctly) interpreted that ST can lead to biased test results.   So how much, if any, of the B-W gap observed in non-experimental conditions can they explain?

Magnitude:  In a meta-analysis, Walton and Spencer (2009) found that ST caused a -0.2 SD difference between black and white test scores. However they used some questionable modeling techniques (e.g. comparing the black experimental patients to white control patients). Wicherts and de Han (2009) presented at a meeting the results of a meta-analysis where they found publication bias and concluded that "Stereotype threat cannot explain the difference in mean cognitive test performance between African Americans and European Americans".  This has not yet been published though.

Generalizability: ST study authors extrapolate the results to the the general setting (e.g., SAT). ST critics counter with the general validity of the tests and note that black SATs under predict academic performance for blacks (Sackett 2008) implying that latent intelligence could not be underestimated. Wicherts and Millsap (2009) challenge this conclusion and note that under prediction can still occur even if measurement variance does not hold.

In order to show that the observed gap is due to stereotype threat you have to show MI does not hold. Wicherts et.al. (2005) note, "If a certain test score gap is accompanied by measurement invariance (and power is not an issue), stereotype threat is not likely to play a differential role in those particular group differences." The few tests of MI for blacks and whites (Dolan 2004) have detected no violations. Lack of MI must be observed in an important test of achievement for ST to have any practical importance.

Accidental Evidence of HH: Regression to the mean in MR rates

2010-12-29T21:51:00.000-08:00

My current obsession with the mentally retarded has led me to some further evidence for the herditarian hypothesis of the black-white IQ difference to add to Chuck's magnum opus. Drews et. al. (1995) conducted a thorough case-control study of retarded children in Atlanta.

There is some great data in the paper that is well analyzed and presented. The author's interpretation of the results would have been improved if they recognized that IQ was heritable which simply explains the higher number of MR children born to women with lower educational status. What is more interesting though is that having a black mother is a risk factor for MR after correcting for educational attainment. Other factors being equal, a college educated black mother has the same risk of having a retarded child as a white mother with a high school education.

This is analagous to the famous SAT - Race - SES graph and what you would expect to observe if the B-W IQ difference is heritable. The probability of a black woman having an MR child would be higher than that of a white women with the same IQ. Of course there are other explanations. Chapman (2008) suggest intergenerational risk factors

To fully address this problem, we may need to consider intergenerational risk factors, which involve the mother’s own developmental history. Maternal intergenerational factors clearly play a role in low birthweight ..., and it is likely that other aspects of development, including cognitive development, also have an intergenerational component .

Intergenerational risk factors that affect cognitive development. What could those be?

Evolution did not stop at the neck: PKU

2010-12-26T12:27:00.000-08:00

Phenylketonuria is a rare inherited genetic disease (incidence ~ 1/15000) that left untreated leads to retardation and seizures. It can be treated and and most first world newborns are tested. With a special diet most suffers are spared a ride on the short bus although a meta analysis (Moyle 2007) showed PKU patients to have average IQs approximately 0.5 of an SD lower than those of controls. So is the incidence of PKU distributed evenly throughout human sub-populations/races? Probably not, Hardelid (2007) found the incidence among Europeans about 10 times that of individuals of African descent.

The Age of the Renowned Cryptozoologist

2010-12-22T21:27:00.000-08:00

History Channel’s Monsterquest is one of the most thought provoking shows on TV. Each episode begins with a description of a putative monster from and “experts” in the field, say the swimming Sasquatch of Prince Edward Island. Then the episodes documents a well financed aunt in the wilderness to find scientifically in convertible evidence of the creature’s existence. Despite the inclusion of drone aircraft, night vision equipment and ground penetrating radar the investigators fail to find the swimming Sasquatch. But does repeatedly failing to find the Sasquatch support its non-existence or do we just need to wait for the falsification of the no Sasquatch hypothesis? What would Karl Popper say?

The show finds pro Sasquatch experts such as Lauren Coleman a “renowned cryptozoologist." The term “renowned cryptozoologist” is troublesome. If Coleman was truly renowned he would have to find something and he would become an orthodox zoologist. But today you can get for famous for repeatedly failing to prove the right hypothesis. The country is full of educrats doggedly trying to establish that the black-white IQ gap is not inherent and can be bridged by innovative pedagogy. Indeed renowned cryptozoologists are in the highly esteemed company of Geoffrey Canada and Michelle Rhee. We all await the day when Sasquatch will swim into New York harbor and the KIPP academies will close the gap.

HBD - Race not Required

2010-12-16T21:14:00.000-08:00

Many people use HBD as a synonym for race realism when a broader definition is more is more useful. Half Sigma defines HBD as a superset of race realism:

"Human biodiversity is an acknowledgment that humans differ from each other in various ways because of our different genotypes. Differences include, but are not limited to, physical appearance, athletic ability, personality, and cognitive abilities."

To deny HBD is to embrace blank slatism which is easy to refute and more importantly puts the deniers on the moral low ground. Like opponents of stem cell research HBD deniers prevent the sufferers genetic diseases possible cures.

Take fragile X, a hereditary genetic condition that presents its sufferers with a one way ticket to retard town. Should we deny HBD in a blank slate fog or should we join Doris Buffett and try to save the children from a lifetime of free bus passes and dish washing jobs. HBD is for the children!

Evolution did not stop at the neck: APOE4

2010-11-24T19:31:00.000-08:00

Apolipoprotein E (APOE) has a well established link to a general neuro-psychiatric disorder: Alzheimer’s disease. Studies show that subjects with one copy of the e4 APOE4 allele have 4 times the odds of getting AD and those with two copies have 20 times the risk. A diagnosis of Alzheimer's dementia is not subject to the vagaries of psychometrics. APOE e4 is also associated with increased cognitive decline in non-demented adults (Fan 2008) and even some decrease in the cognition of children has been observed (Bloss 2008).

So is the incidence of the e4 allele the same in all human populations? Probably not, Eisenberg (2010) has the latest paper to find that the incidence varies by latitude. Populations near the equator and the Arctic have a higher incidence than those at mid latitudes. They found this curvilinear relationship even after adjusting for population structure supporting their hypothesis that the differences are driven by the natural selection instead of genetic drift.

APOE4 is associated with high cholesterol levels and Eisenberg's hypothesis is that this was beneficial in warm or cold climates where humans had high metabolic rates. Fortunately, human equality is a contingent fact of history so we can ignore these findings.

The weird thing about Neven Sesardic's ariticle

2010-11-07T13:55:00.000-08:00

After I recommended it to jewamongyou, I reread Neven Sesardic's excellent article "Race: A Social Destruction of a Biological Concept" and there is something weird the acknowledgements. Sesardic article gives a great overview and critique of the philosophical discussion of the "race" concept claiming that current philosophers of science have ignored a great deal of scientific of evidence to argue away the biological reality of race. The article is accessible to non-philosophers like myself and the references alone make it worth the effort. I missed the acknowledgement where Sesardic thanks Alex Rosenberg the chairman of the Duke Philosophy department. Rosenberg is also a signer of the liberal Group of 88 statement on the Duke Lacrosse unpleasantness. So, a big shot liberal professor of the humanities gave helpful comments on a paper that discusses Lewontin's fallacy and favorably references Jensen and Rushton. That is weird.

The underwhelming "liberal" Gene

2010-10-31T11:00:00.000-07:00

I saw this from Jim Manzi on the liberal gene paper:

"As you may remember, I wrote a long piece for the magazine in 2008 that described why we should be very skeptical of assertions of causality that are derived from the kind of study that you reference. The basic reason is that, while these kinds of studies have remarkable rhetorical force because their purported subject is biology, if you look under the skin at the bones of the analysis, the core method is traditional social science. The article you cite is an almost perfect illustration of this."

I appreciate the paleo-Fisherian (e.g., randomization is the only way to establish causality) viewpoint but there is utility to this sort of observational study and relating liberalism to biology gives you more than just rhetorical force. Neither the association of smoking and lung cancer nor the association of the BRCA gene and breast cancer was established via a randomized experiment in humans. The overwhelming weight of evidence came from observational studies AND in-vitro and animal studies, i.e., results from the wet lab. In-vivo and in-vitro expreiments are amenable to randomization so the "totality" of the data convinced people, correctly I think, to stop smoking or, in the case of the BRCA gene, have prophylactic mastectomies. A relation between a gene and behavior established in an observational study can be supported with evidence from the lab where causal relationships can be established.

I doubt that will happen with the liberal gene though. I read the paper and the results are underwhelming. The paper fails to show that liberalism is related to a gene. It purports to show a gene (the novelty seeking DRD4) by environment (number of friends in school) interaction is related to your adult ideology. The actual magnitude of the interaction (p = .02) is unexciting. If you have two copies of the R7 allele and you have four friends you become on average, 0.16 (on a five point scale) more liberal than a person with two friends. If y ou have no copies then friendship has no effect. It is weird that there is no main effect of genes or number of friends on liberalism and makes me doubt the results can be replicated. Their table 1 sums it up: there will be no selective breeding of NPR listeners any time soon.

The Utility of Studying the Genetic Basis for Intelligence

2009-09-21T18:07:00.000-07:00

A recent article in Bloomberg shows points out the benefits of studying the genetic basis of intelligence. Four Bio-Pharma companies are developing treatments for Fragile X syndrome, a very rare genetic that leads to retardation. Unlike, the more common genes found to be weakly associated with intelligence, presence of the mutation in boys is a one way ticket to tard town. Given the glacial pace of medical research these days, I would not rush to disband the Special Olympics but one sunny academic says:

This medicine “could be the first to reverse mental retardation,” Randi Hagerman, a professor at the University of California at Davis who’s working with Roche on the clinical trials, said in an interview. “What everyone is hoping is that the causes of autism may converge on a final common pathway that may be sensitive to this drug. The secret hope is that we might find it can be used much more broadly than for Fragile X.”

There is a lot of hype in the early phases of drug development but the development strategy of “find the gene, find the mechanism, make the drug” has worked for some conditions. So, the small amount of resources put into researching the “nature” side of intelligence has yielded a few drug candidates for a debilitating condition. What exactly are we getting for the resources put into the “nurture” side?

The Overprediction of NAM Achievement

2009-07-05T18:55:00.000-07:00

This graph shows three hypothetical regression lines relating one measurement of academic achievement (e.g., test score) and an outcome (e.g., grade point average, another test). The center line represents a regression for the entire population, the top line the regression line for the higher achieving population where the test underpredicts the outcome, and the bottom for a low achieving population where the test overpredicts the outcome. For example, the relationship between SAT scores and first year college GPA for NAMS and the Whites and Asians majority (WHAM). The SAT is said to overpredict minority first year grades compared to the entire population and underpredict WHAM grades. For example, a NAM with a SAT of 2100 will on average have a lower GPA than a WHAM with the same score.

Many explanations are offered for this phenomenon. When the y-axis represents the California High School Exit Exam ( CASHEE) and the x-axis the California Standardized Test ( CST ) Sean Reardon, professor of Education at Stanford University, speculates that the phenomenon is caused by stereotype threat. As Steve Sailer ridiculed, Reardon’s theory is that the NAMs are afraid of being pigeonholed as high school drop outs so they choke on the exit exam and become high school drop outs. When the y-axis represents bar exam scores and the x-axis and LSAT scores, Guy White, noted HBD blogger, speculates that the ETS has modified the LSATs to be biased against the WHAMs.

I cannot dismiss Guy White’s or Prof. Reardon’s explanations out of hand, but I can offer a simpler one: measurement error. Assume that tests of academic achievement are actually imperfect but unbiased measures of intelligence, and suppose, just for fun, that this intelligence is distributed differently between the two different populations. This would imply that a NAM student who obtains a score higher than their group mean on a test is likely to score closer to their group mean on their next test of academic achievement. The student, like us all, regresses to the mean. When this occurs you will observe the graph above without test bias and no difference in test anxiety. I show a simple example below.

Others have pointed this out. La Griffe du Lion discuss this phenomenon and coldly states that if a college wants to maximize first year GPA it should require that NAMs have higher SAT scores than WHAMs. On the other side of the debate, anti-HBD blogger Three Toed Sloth complains about the failure of Herrnstein and Murray to account for this type of error in their analyses. Sloth links to the work of psychometrician Roger Millsap who says

The direction of the intercept difference [in the graph above] is determined by the factor mean difference: the group with the larger factor mean will have the larger intercept. If a common regression line is imposed on the two groups, the group with the higher intercept will show systematic underprediction via the common line. In many applications, this group is the majority or reference group.

Millsap, the current editor of Psychomterika, gives conditions where the test is “measurement invariant” across population but it is not “prediction invariant” across populations. That is, the test is “fair” but you still overpredict NAM performance.

Here is a simple example to illustrate how over or underprediction can occur even on a fair test. Define M_1N = mean NAM score on test 1 and and M_1W = mean WHAM score on test 1. Define M_2N and M_2W similarly, assume that test 1 and test 2 are distributed bivariate normal in both populations with a standard deviations and a correlation of R. Finally, assume that both means are one standard deviation apart (i.e., M_1W – M_1N = S = M_2W – M_2N).

Suppose a NAM and a WHAM both receive a score of X on this test 1. Who will score higher on test 2? For those of you lucky enough to have avoided an introduction to mathematical statistics class, the expected scores for the NAM and WHAM are

(1) E(Test 2 NAM) = M_2N – R*(M_1N – X) and
(2) E(Test 2 WHAM) = M_2W – R*(M_1W – X)

respectively. Under the one standard deviation assumption the expected difference in the scores will be S(1 – R) in favor of the WHAMS. As noted more generally by Robert Miller, unless the tests are perfectly correlated (i.e., R=1), you cannot be “color blind” if you want to calculate the expectation.

The Smart Money is on at least a little cognitive HBD

2009-04-15T19:48:00.000-07:00

I have been enjoying the HBD discussions hosted by Feminstx and in one exchange she discussed the possibility that there is at least one gene that is associated with cognition that is not evenly distributed between the races. If such a gene could be found, feministx agreed “I will believe that the specific trait that that gene codes for is different between races on average due to heredity.”

That is, there is cognitive HBD. So, is it reasonably likely that such a gene exist? I think so and I have constructed my own subjective prior probability for this event and Statsquatch’s Subjective Prior Probability (SSPP) gives 99 to 1 that such a gene exists.

Mark Watham at Congenial Times had a posts on five genes that maybe associated with IQ and are definitely distributed differently between the races. Now what are the odds that any one of these genes is definitively associated with IQ? I know biologist hype their results so I put the chance at 1 in 10 for each gene. The chance that none of these 5 genes will be associated with IQ is 59%. Put another way, SSPP gives a 41% chance that one of these five genes will demonstrate cognitive HBD. These are not overwhelming odds but this is not a Black Swan event either.

The genes he listed are genes that are common enough to find in the Hapmap but are believed to have a small effect on IQ. The genes responsible for mental retardation are uncommon but have a larger effect on IQ and this paper mentions 282. What is known about the racial distribution of these genes?

I could only find information on two genetic MR conditions. PKU deficiency appears to be over five times more prevalent in Caucasian than African Americans while one study has Fragile X 1.5 times more prevalent in African Americans than Caucasian males . I assign a 95% probability that at least one of these (90% chance for PKU, 50% for Fragile X) genes is differentially distributed between the races. Combining that with five more common genes the SSPP becomes 97% .

What about the other 280 genes associated with MR? I will give each a 99% chance of not being associated with decreases in IQ or not being differentially distributed gives a 6% chance that none of these genes will provide evidence for cognitive HBD. All of these genes give a SSPP > 99%. These genes, particularly the MR genes, may explain very little of the group differences in IQ but they still point to some cognitive HBD.

More Spanking Research Still Desperately Needed

2009-03-22T09:33:00.000-07:00

TGGP pointed out this spanking study by Lynch et. al. where the authors analyze the relationship between bad outcomes (drug abuse, externalizing behaviors, and internalizing behaviors) and corporal punishment in the children of a fairly large sample of Australian twins. By applying a hierarchical linear model (HLM) to this natural experiment they try to control for genotype while looking at the association of punishment and bad outcomes. Their conclusion is that severe (e.g., beating) but not moderate (e.g., spanking) corporal punishment leads to increased drug and alcohol use and bad externalizing behavior regardless of genetics.

An HLM is type of random effects model that is used to analyze data where the experimental units can be organized into different levels. In this case there are three levels: the family the twins are raised in, the family of the twins, and the individual children of the twins. This model allows for testing the effect of punishment at the child level while at least partially controlling for genotype at the twin parent level.

Unfortunately, the authors spoil the analysis by making some poor analysis choices. The twins are a subset of larger Australian twin study including subjects who reported having problems with alcohol, being divorced, conduct disorder and chosen controls. To control for this selection bias the authors use the residuals from a simple regression at the twin level model as the dependent variable in the HLM. I find this an odd choice. They could easily have added these factors to the twin level of the HLM where these effects could be examined directly.

More suspicious is the fact that their conclusion that mild physical is equivalent to mild non-physical punishment is not based on the HLM results. Rather, after inspection of mean outcomes at the child level the authors collapse the mild physical and the mild non-physical punishment groups in the HLM and compare this pooled group to the harsh punishment group. They could instead have tested the mild non-physical vs. mild physical punishment effect in the HLM and determined if after controlling for genotype these groups are similar and then compared each group to the severely punished. Could it be that the dutifully spanked children have better behavior than the non-spanked? Did they dare not offend the powerful anti-spanking lobby? More likely, the clear difference in behavior found between the severe and the mildly punished children is muddied once you fit a more complicated model. Maybe I am cynical, but I doubt you could publish a paper that did not find some kind of association between severe punishment and bad outcomes. Despite the failure of the editors to demanded a more thorough analysis, the study does indicate that a little swat may be ok. Claeralry, more spanking research is still desprately needed.

Fun with Fertility...Data

2009-02-16T20:08:00.000-08:00

One TGGP’s commentators, Jason Malloy, linked to this article on sex, status and reproductive success in the US by sociologist Rosemary Hopcroft. She finds that fertility generally declines as income, education and IQ increase but that the relationship is different for men and women. She reports that male fertility increases with income while female fertility decreases. Since income is usually positively correlated with IQ this association would be evidence against the “only stupid people are breeding” theory.

She used GSS data but modeled the relationship between fertility and other factors using an ordinary least squares regression. I think a generalized linear model with a Poisson distribution and a log link is more appropriate for count data so I down loaded the GSS data and repeated her analysis. I confirmed that the association of income and fertility is different in men and women (there is an interaction, p < 0.001) but I think the relationship is more complicated. A GLM was fit separately for each gender and had a main effects of age, age squared, WORDSUM score and income and there appears to be an interaction between WORDSUM and income. Low income is an effective birth control for smarter men but not for the dimmer (see graph). Intelligent poor men report the fewest children but the fertility of men with lower scores on the WORDSUM does not vary as much by income. This reminds me of a result in cognitive epidemiology. Having children when you have a low income may be viewed as a “disease” that men with higher intelligence avoid.

GSS variables used: INCOME (Family income in 1996 dollars), CHILDS, WORDSUM, SEX, AGE

Interesting NLSY Analysis

2009-02-01T10:35:00.000-08:00

The Inductivist links to this analysis of National Longitudinal Survey of Youth(NLSY) data that analyzes income, wealth, and education by race, sex and IQ. The authors of this simple analysis have a particular point of view but their results are consistent with some more complex analyses performed by researchers with a different point of view that I discussed earlier .

Of course, I have a few technical quibbles. They analyze a Jewish cohort separately from other whites but they do not give sample sizes or standard errors. Since I doubt there are more than 500 Jews in the NLSY data set I can not make much of these results.

Their analysis on black/white differences, however, replicate the findings that blacks enjoy a higher return on IQ than whites and that this difference varies by gender. Technically, there is a gender by race by IQ interaction. Unlike the earlier analyses, the authors show that in males the black vs. white return on IQ varies by IQ level. Smart black males considerably out earn equivalent gentile white males but whites with average intelligence earn about the same as average black males. This variation in return to IQ supports the conclusions that this difference is due to affirmative action and not to test bias or measurement error if you would expect the bias to be the same at all IQ levels. This interaction is not present in females; black women enjoy a consistently higher return on IQ than do white females. This may be confounded by hours worked or other variables.

There are also some interesting analyses on wealth and education. Read the whole thing.

Smart Sperm Theory

2009-01-25T15:18:00.000-08:00

This latest contribution to cognitive epidemiology by Arden et. al finds a correlation between intellignece as measured by “g” and sperm quality and brings up many sticky issues.

Cognitive epidemiology investigates correlations between health outcomes and intelligence and has found them for the most important health outcome mortality . Linda Gottfredson speculated that differences in IQ are the ultimate cause of the observed differences in health outcome by subjects in different socio economic stratum.

The problem is determining causality, is it poor implies stupid and fat that together imply early death or stupid implies fat and poor that both imply early death? Studies look at the relationship between IQ and specific events like accidents or heart attacks and then try to control for other factors. These epidemiological versions of The Bell Curve analyses suffer from the usual problems of untangling correlation from causality. This smart sperm paper is unique because the authors are not looking for causation, only correlation.

Arden et. al. reanalyzed some crusty data from a Vietnam Vet study done in 1985. They had intelligence and sperm data on 425 veterans and a found a small but significant correlation (0.14 – 0.19) between sperm motility, concentration and count and intelligence as measured by a “g” factor derived from a principle component analysis of four IQ test given around the same time as the semen collection. Further analyses controlled for a few variables known to be related to sperm quality (Body mass index, smoking history, age and abstinence) and”g” was still a significant predictor of sperm quality. So the correlation between “g” and sperm quality may not be mediated by the factors. That is, stupid does not imply fat which implies lazy sperm. In spite of assessment of homo and heteroscedasticity the analysis appears a bit perfunctory to me since there is no accounting for measurement error and other factors such as race or SES.

Regardless, the results lead to an explosive stream of evolutionary speculations from the authors about a general fitness factor: the “f” factor. If intelligence does not cause good sperm then their association may have a direct genetic basis like the correlation of intelligence and height. Unlike, height though, this association is probably not the direct result of assortive mating. Instead, it may be an indirect result of assortive mating through detrimental “mutation” load. Smarties have lower overall mutation loads and preferentially mate with other people with lower loads of mutations. Obviously this needs replication, but the Mensa guys should start chatting up the ladies with this news about mutation loads.

More Spanking Research Needed

2009-01-10T20:34:00.000-08:00

Somebody needs to do some research on spanking so the Brahmins will try a little tougher love on their kids. Read this sad tale of Peter Morse. Raised on “Peter, Paul and Marry” in an upscale neighborhood, Peter became a community organizer with a Ph.D. in History (he wrote his thesis on the Wobblies) and ran a clean needle exchange in Frisco. The poor bastard found do-gooding rather stressful and ironically overdosed himself on heroin. Could this tragedy have been avoided if his dad had given him a good thrashing once in a while?

Hard to say, but I spent five minutes on Google Scholar and found this spanking meta-analysis and the the anti-spanking data a little thin. Just like The Bell Curve only observational studies are used and no pseudo-causal methods (e.g., path analysis, propensity scores) are employed to account for other factors like SES. So correlation does not imply causality. Most studies used the Conflict Tactic Scale to measure behavior which I am pretty sure was developed useing the rubbishy method of factor analysis behind the “G-Factor.” Unlike the data in the The Bell Curve the studies are contradictory and the correlations are pretty low. So until there is further research, Statsquatch says spank away.

College graduates need to spank their kids

2008-11-09T22:42:00.000-08:00

One analysis of the NLSY in Intelligence, Genes, and Success is of particular interest since it questions Herrnstein and Murray’s conclusion that criminal activity is associated with low IQ. Sociologist Lucinda Manolakes reanalyzes of criminal activity as a function of IQ, race, parental education and some other variables. She claims that there is an interaction between race and IQ that implies the chance of being criminally active increases with IQ for black makes, the opposite of the trend for white males.

There are a few problems though. She added some novel variables to her analysis including environment (urban, Rural, etc) and she did not state how the final results depend on these factors. This could be an old data “dredging” trick: add and subtract background variables and interactions from a model until you get the “right” answer for the variables you are interested in, here an IQ and race interaction. I have done this way too much myself. Also, from her equations I obtained the following logistic regression coefficients for IQ .0233 (SE=.0065) for whites and 0.0167 (SE=.0066) for black. So both coefficients are positive, significant (p < 0.01), and going in the same direction indicating that the relationship between IQ and crime is not reversed for whites and blacks. I might be missing something though but I can not tell unless I replicate the analysis.

Her analysis also has an interesting non-racial result: an interaction between IQ and parental education. The likelihood that a dumb child (lowest third of the IQ distribution) is criminally active increases with parental educational., That is, dumb children of college graduates are more criminally active then dumb children of high school drop outs. Maybe it is because college graduates do not spank.

Wages: Sucks to be poor and white, good to be smart and black, but stupidity is always a problem

2008-11-01T11:00:00.000-07:00

Herrnstein and Murray look at wages directly and indirectly by looking at prestige of occupation in The Bell Curve and claim that holding IQ steady that blacks are more likely to be in prestigious occupations and blacks make similar wages to whites by occupation groups. There models are not very complicated, however, and they do not test for differences in return for IQ and other factors nor do they use the standard method of modeling wages with the log transformation. In Intelligence, Genes, and Success, Cavello, Elabaddi, and Heeb (CEH) and Cawley, Connerly, Heckman, and Vytalacil (CCHV) look at wages directly to test Herrnstein and Murray conclusions with more complicated models that use the standard log transformation of wages.

CCHV fit a separate regressions for each gender-race combination (black-males, black females, Hispanic males, etc) on log wages over time as a function of “g” and the nine other principle components of the intelligence test and some background variables (experience, experience squared, unemployment rate). They conclude that “g” exists, has the same attributes in all the race gender combinations, and is a very good predictor of wages, though not the only good predictor. This is all consistent with the Bell Curve although you can quibble about the size of “g” predictive effect. They then make a more startling claim. After, rejecting the null hypothesis that all the coefficients for all the gender-race combinations are the same they then declare “payment is not made for ‘ability’ alone”, that is, there is no meritocracy. Besides the fact that there model building is not transparent (why not unemployment rate squared) they have a technical problem that with a large data set (over 20,000 observations) it is easy to reject null hypotheses (get small p-values) for a very small differences that may not be very meaningful. You have to actually look at the way and the size the coefficients differ. When you actually look at the coefficients for IQ (“g” is this case) you see that the return for IQ is always higher for black males and females than white males and females. This means that in many cases blacks with a given IQ are making more than whites with the same IQ. The authors do not tell us what to make of this evidence but they do quote Ecclesiastes so I am sure they felt pretty smug.

CER do a better job of detangling the interactions between gender, race, and IQ. They refit Herrnstein and Murray’s model but find a fairly convincing age, race, and IQ interaction. Specifically, the return on IQ for black females is much higher than that for white females or either white or black makes for that matter. They also model the log wages and find the same thing as CCHV, higher returns on IQ for black then for whites, particularly black females. They conclude by testing certain linear combinations that the “average” black male does make %6 less then the comparable white man, but for women this is reversed with “average” black women making 15% more then the comparable white women. The male result does not contradict the differences in return of IQ, it just means that whites with black like characteristics (below white average IQ, low SES, low education attainment, average black age) make more than blacks. The black advantage on the return of IQ is made up somewhat by the higher white male return for age. The opposite is true for females. Another result is that white males have a higher return for socio economic status than black males: poor whites have a greater disadvantage to rich whites then poor blacks do rich blacks. CER do not seem smug, but they also do not discuss these differences. The obvious answer is reverse discrimination in both in males (supported by the SES findings but contradicted by the age finding) and females (supported by all variables), but it could also be an artifact of measurement error.

None of these authors ask a related but non-racial question. Is the return on IQ similar in all occupation groups or for all levels of education? Does a dumb high school dropout garbage man make less than a brainiac drop out garbage man? Maybe the later is better at fighting off the Raccoons. Herrnstein and Murray cite two economists that do look at this with the NLSY: Blackburn and Neumark. You have to wade through a lot of weird econometrics to get at it, but their provocative answer appears to be no. They write, “The increase in the return for education has occurred largely for workers of higher levels of ‘academic ability’[IQ]”. The conclusion is that for some of us spending 8 years at community college has no advantage over spending 8 years working at Burger King. If they are right than maybe we should leave a few children behind.

Fun with narrow sense heritability (part 1)

2008-10-26T13:59:00.000-07:00

In chapter 3 of Intelligence, Genes and Success Daniels and others perform a Bayesian meta-analysis of IQ heritability papers and estimate the narrow sense heritability to be "only" 0.34. Such a small number, they claim, invalidates The Bell Curves prediction of a hereditaryelite.

Is this a small number? Let's have some fun and imagine a virus that turns all males with an IQ over 100 impotent, gay, or otherwise reproductively unfit. Also, women mate randomly. It is paradise for the male riders on the short bus and reminds me of high school.

Supposing you start with an average IQ of 100 and and SD of 15. What happens in 22 generations? I ran a simulation with many simple and unreasonable assumptions and found a mean of IQ of 79 and, more importantly, only 0.21% of the population has an IQ over 120. No more smarties! Unless, Gustafason's banana peel just so story is true?

#shortbus party simulation in R assuming stable population and random mating. heridest_function(mp1,mp2,male,female,narrow){ meanof_((mp1+mp2)/2) + narrow*((male+female)/2 - (mp1+mp2)/2) } mcrit_100 male_rnorm(5000,100,15) female_rnorm(5000,100,15) narrow_.34 male.mate_sample(male[male"<"mcrit],length(female),replace=t) est_heridest(100,100,male.mate,female,narrow)
male_est+rnorm(length(est),0,sd)
female_est+rnorm(5000,0,sd)
newgen_c(male,female)

for(i in 1:21){
mp1_mean(newgen)
male.mate_sample(male[male"<"mcrit],length(female),replace=T) est_heridest(mp1,mp1,male.mate,female,narrow)
male_est+rnorm(5000,0,sd)
female_est+rnorm(5000,0,sd)
newgen_c(male,female)
print(mean(newgen))
print(sum(newgen>120)) print(i) }

Best "just so story" ever

2008-10-19T12:46:00.000-07:00

Linda Gottfredson's reprint site never disappoints. She has her own 'just so' theory of the evolution of intelligence in the 2007 paper "Innovation, Fatal accidents, and the Evolution of General Intelligence".

Extending here research results that people with "low g" have more accidents than those with higher "g", she conjectures that this differential worked in pre-history weeding out the short bus riders. She claims that technological innovation is a double edged sword, lowering the over all age adjusted mortality but introducing "novel risks" (e.g., fire, automatic weapons, meth) that disproportionally affect the dimmer among us. For those worried about a future idocracy where the dims out breed the smarties, she assures us: "Fatal accidents are still a major cause of death in all societies, so they provide continuing opportunity for natural selection." Whew, what a relief.

Maybe the would be sterilizers over at half sigma need to change tactics. Instead of snipping the short bus riders they just need to litter their neighborhoods with hazards (e.g., down power lines, banana peels, refrigerator boxes, crack cocaine, cheap firearms) to provide an opportunity for natural selection.

Bell Curves

2008-10-17T15:47:00.000-07:00

If you like Bell Curves as much as I do you should read Intelligence,Genes, and Success:Scientists Respond to the Bell Curve. This is a 10 year old rigorous critique of The Bell Curve. It give a lot of technical back ground into factor analysis, intelligence testing, and the statistics needed to detect discrimination. The tone is measured. It appears that the authors would like to dismiss some of the conclusion but they can not land many blows on the main points of the book. I think the conclusions hold up under their reanalyses.

I guess they are on crack.

2008-10-12T15:31:00.001-07:00

With the market meltdown it appears that many of Americas elite inferers are indeed on crack. According to the Black Swan Guy they are all doped out of their minds. Mencius Moldbug says we should go long on gold. But how do I short crack?

Does being fair pay? Not always

2008-09-28T14:31:00.000-07:00

There is a lot of discussion on Sailer's Blog about "the diversity recession." This reminds me about this post which references this post. Both conclude that in some cases it is unprofitable to treat people equally.

Suppose you have a measure (e.g., SATs, FICO score) that are suppose to be a proxy measured with error for some other quality (e.g., "intelligence", "credit worthiness"). Your job is to pick a single cutoff of the score to distribute some good (college admission, mortgage) that will minimize something bad (drop outs, defaults). Suppose that

1) Your proxy is a good predictor of the bad thing

2) There are two populations with different normal distributions of your proxy measure: one with a lower mean.

Should you be fair and use the same cut off for both populations or should you be mean and demand a higher score for the group with lower average? The second link above uses Bayes rule to claim that being mean is the most profitable. I was surprised at this until I wrote out Bayes theorem. Suppose FICO scores measure credit worthiness. If you use the same cut off for creditworthiness in both populations you will be more likely to incorrectly categorize the lower scoring group as creditworthy.

If you do not believe me, run the following simple simulation in R. Suppose you have two populations with different distribution of creditworthiness (a distributed like a fico score):

Group 1: sampg1_rnorm(1000,680,100)
Group 2: sampg2_rnorm(1000,600,100).

Suppose Fico are an unbiased measure of the above but are measured with error (SD=50)

coef_1
true_rbind(sampg1,sampg2)
fico.est_coef*true + rnorm(length(samp),0,50)

You set up a cut off of FICO 700 as a "good risk". How many people are you going to incorrectly categorize as "good risk" in group 1?

crit_700
clt1_(true[1,]<crit)*(fico.est[1,]>crit)
denom1_(fico.est[1,]>crit)
c(sum(clt1),sum(denom1),sum(clt1)/sum(denom1))

About 17%. How about group 2?

clt2_(true[2,]<crit)*(fico.est[2,]>crit)
denom2_(fico.est[2,]>crit)
c(sum(clt2),sum(denom2),sum(clt2)/sum(denom2))

About 30%. So, if the real world matched this simple example you should probably discriminate. No wonder they made it illegal, this crime could pay.

g-Factor criticism - Scary but Hilarious

2008-09-23T12:20:00.000-07:00

I was rereading this discussion of the "g factor" from "Three-toed sloth" and found a link to this excellent paper by "Glymour".

He spends most of the paper criticizing the statistical methods (multiple regression and factor analysis) used in the "Bell Curve". The later has always seemed a little odd to me with it's "latent variables" but it is (or at least was) the foundation of psychometrics and is used extensively in the real world to construct questionnaires. For instance, all the anti-depressants on the market today were approved based on a scale (usually the HAM-D) that was developed with factor analysis. Multiple regression applied to non-randomized experiments is the "back bone" of all social science statistics (e.g., economics, finance, sociology) and a lot of "biological" science (e.g., ecology and epidemiology).

Of both methods Glymour concludes:
"Hernstein and Murray use the tools of their profession, and social science
statistics generally, gave them. The tools were incompetent for the use
Hernstein and Murray put them to, but what were they to do?"

This statement will not surprise anyone who has been taught or has practiced the perils unscrambling the eggs of observational (e.g., non-randomized) data with linear or generalized linear models. But if we can ignore Hernstein and Murray because they use flawed methods who else can I ignore: the federal reserve, the USDA, the CDC, the guys who came up with my FICO score? Is there a crack in the inferential foundations of society? Are the inferers of my society on crack? Are we being led by court astrologers like Moldbug says?

That would be hilarious. Scary, but hilarious.

G-Factor and Spooky Theories and Theorems

2008-09-22T17:39:00.001-07:00

I found this Linda S. Gottfredson link on the inductivist: http://www.udel.edu/educ/gottfredson/reprints/. She is an intelligence researcher who has done work finding variables that correlate with "g". She claims "g" correlates positively to various measures of health including sperm count. I am not an expert on this "g" phenomenon but if half of what these "g-men" say is true, then it is a little spooky.

But then I got to thinking, the normal distribution itself is a little spooky. Not all variables in nature are normally distributed like height or “g”. Mortality is a good counter example. However, in most cases the averages of most random variables will be normally distributed. That is if you took the age of death for 1000 sets of 1000 people (1,000,000) the distribution of the averages would be bell shaped.

That, the results of the central limit theorem, is spooky if you ask me. If you have the misfortune of ever taking a math stat class you will have to prove the central limit theorem for various cases: independent random variables, weakly dependent random variables, identical distributed random variables, etc. What do you always get: the bell curve. This seems a lot spookier to me than the golden ratio.

National Longitudinal Survey of Youth Home Page

2008-09-21T15:07:00.000-07:00

Has anyone every worked with the National Longitudinal Survet of Youth? This is the primary data set that Murray and Herrnstein use in 'The Bell Curve."

They say it is available here: http://www.nlsinfo.org/ordering/display_db.php3.