Amerika

Furthest Right

Refuting Gould

“Science” is just a word. Anyone can claim to be a scientist, and getting a degree in the sciences doesn’t guarantee one knows more than lab theory. Scientists lack an essential tool: critical thinking, specifically the ability to analyze an argument with the ability of a philosopher.

As a result, we have a constant drama where a scientist presents a single detail and makes a huge abstract conclusion from it, and then those pesky philosophy course types point out that while the detail is true, the conclusions are not. They are promptly accused of being “anti-Science,” with the same lofty tones previously reserved for burning witches and heretics.

Stephen Jay Gould, a man who clearly knew where an audience could be found and that they’d make him a household name for saying things they agreed with, attempted a debunking of Herrnstein and Murray’s The Bell Curve, which included among its many arguments observations about race and IQ.

Gould then embarks on a process that is sure to hoodwink fools by sending them on a moral pretense and intellectual superiority trip, all by giving them a true detail and then extracting false conclusions from it:

The central fallacy in using the substantial heritability of within–group IQ (among whites, for example) as an explanation of average differences between groups (whites versus blacks, for example) is now well known and acknowledged by all, including Herrnstein and Murray, but deserves a restatement by example. Take a trait that is far more heritable than anyone has ever claimed IQ to be but is politically uncontroversial—body height.

Suppose that I measured the heights of adult males in a poor Indian village beset with nutritional deprivation, and suppose the average height of adult males is five feet six inches. Heritability within the village is high, which is to say that tall fathers (they may average five feet eight inches) tend to have tall sons, while short fathers (five feet four inches on average) tend to have short sons. But this high heritability within the village does not mean that better nutrition might not raise average height to five feet ten inches in a few generations.

Similarly, the well–documented fifteen–point average difference in IQ between blacks and whites in America, with substantial heritability of IQ in family lines within each group, permits no automatic conclusion that truly equal opportunity might not raise the black average enough to equal or surpass the white mean.

Gould argues that genetics plus nutrition could affect height. Although he has noted that possible polycause, he doesn’t catch the next level: while nutrition could add on to height, the base level of height — determined by genetics — remains the same.

Amazing how many people are fooled by this simple technique.

Furthermore, Herrnstein and Murray know and acknowledge the critique of extending the substantial heritability of within–group IQ to explain differences between groups, so they must construct an admittedly circumstantial case for attributing most of the black–white mean difference to irrevocable genetics—while properly stressing that the average difference doesn’t help in judging any particular person, because so many individual blacks score above the white mean in IQ.

He’s implying a subtle variation on Lewontin’s fallacy here, which is that because individual differences seem greater than differences between groups, group differences are presumed not to exist — when once again, both conditions can be true with different starting levels.

This case includes such evidence as impressive IQ scores for poor black children adopted into affluent and intellectual homes; average IQ increases in some nations since the Second World War equal to the entire fifteen–point difference now separating blacks and whites in America; and failure to find any cognitive differences between two cohorts of children born out of wedlock to German women, reared in Germany as Germans, but fathered by black and white American soldiers.

We cover these surveys elsewhere, but here’s the basic gist: cherry-picked data can result in any outcome, and if you take whites approximate to blacks in IQ, and compare the two, you’ll come out with equality — but from an unrepresentative sample.

Nothing in The Bell Curve angered me more than the authors’ failure to supply any justification for their central claim, the sine qua non of their entire argument: that the number known as g, the celebrated “general factor” of intelligence, first identified by British psychologist Charles Spearman, in 1904, captures a real property in the head.

Here Gould, who we think should know better, uses an incompletion argument: “because it is not written in the heavens that g equals actual intelligence, it must be 100% wrong.” The fact of the matter is that g like any other assessment open to us is a measurement of intelligence potential. A drugged or dead high g individual may not perform intelligently, or may have test-taking anxiety. But without high g, we don’t have the ability to have high intelligence. Gould somehow sidesteps that necessary realization.

The next one’s a good laugh:

This is scarcely surprising, and is subject to interpretation that is either purely genetic (that an innate thing in the head boosts all performances); the positive correlations in themselves say nothing about causes. The results of these tests can be plotted on a multidimensional graph with an axis for each test. Spearman used factor analysis to find a single dimension—which he called g—that best identifies the common factor behind positive correlations among the tests. But Thurstone later showed that g could be made to disappear by simply rotating the dimensions to different positions. In one rotation Thurstone placed the dimensions near the most widely separated attributes among the tests, thus giving rise to the theory of multiple intelligences (verbal, mathematical, spatial, etc., with no overarching g). This theory (which I support)

Positive correlations don’t imply causation, he says; and then without noticing he’s crossing his own T, goes on to support one of those positive correlations made from skewing the data from a relatively straightforward position to one that will create radically different results. Yes, any data can be manipulated; that doesn’t mean the manipulations are valid. He has confused manipulating the instrument of measurement with getting different measurements.

As for Kaus’s second issue, cultural bias, the presentation of it in The Bell Curve matches Arthur Jensen’s and that of other hereditarians, in confusing a technical (and proper) meaning of “bias” (I call is “S–bias,” for “statistical”) with the entirely different vernacular concept (I call it “V–bias”) that provokes popular debate. All these authors swear up and down (and I agree with them completely) that the tests are not biased—in the statistician’s definition. Lack of S–bias means that the same score, when it is achieved by members of different groups, predicts the same thing; that is, a black person and a white person with identical scores will have the same probabilities for doing anything that IQ is supposed to predict.

But V–bias, the source of public concern, embodies an entirely different issue, which, unfortunately, uses the same word. The public wants to know whether blacks average 85 and whites 100 because society treats blacks unfairly—that is, whether lower black scores record biases in this social sense. And this crucial question (to which we do not know the answer) cannot be addressed by a demonstration that S–bias doesn’t exist, which is the only issues analyzed, however correctly, in The Bell Curve

Another clever sidestep. If the test represents basic attributes of intelligence, and treats both groups the same way, it is not biased unless it is specifically designed to take advantage of a trait of one group and not another — unless Gould is arguing that logic and nerve impulse speed are somehow a white thing (or a black thing), his argument is ludicrous here, but it will fool the sophomoric.

The book is also suspect in its use of statistics. As I mentioned, virtually all its data derive from one analysis—a plotting, by a technique called multiple regression, of social behaviors that agitate us, such as crime, unemployment, and births out of wedlock (known as dependent variables), against both IQ and parental sociometric status (known as independent variables).

This is some ugly linear thinking. Gould fails to acknowledge that if multiple types of measurement show a single pattern, there is validity and more likely a causal relationship. As pointed out elsewhere, “correlation does not equal causation” has a caveat which is “correlation does not equal not causation, either.”

My charge of disingenuousness receives its strongest affirmation in a sentence tucked away on the first page of Appendix 4, page 593: the authors state, “In the text, we do not refer to the usual measure of goodness of fit for multiple regressions, R2, but they are presented here for the cross–sectional analyses.” Now, why would they exclude from the text, and relegate to an appendix that very few people will read, or even consult, a number that, by their own admission, is “the usual measure of goodness of fit”? I can only conclude that they did not choose to admit in the main text the extreme weakness of their vaunted relationships.

Although low figures are not atypical for large social–science surveys involving many variables, most of Herrnstein and Murray’s correlations are very weak—often in the 0.2 to 0.4 range. Now, 0.4 may sound respectably strong, but—and this is the key point—R2 is the square of the correlation coefficient, and the square of a number between zero and one is less than the number itself, so a 0.4 correlation yields an R–squared of only .16. In Appendix 4, then, one discovers that the vast majority of the conventional measures of R2, excluded from the main body of the text, are less than 0.1.

Attack a detail, claim it’s the lynchpin of the whole argument. While his argument makes sense for statistical arguments that claim a single factor correlates exclusively to a single result, that’s not what Herrnstein and Murray are arguing. Instead, they’re arguing that a complex set of genetic traits leads to similar conclusions as a design, and that we can see this through multiple factors compared, because at the time our genetic science did not allow us to identify genes with traits to the level needed.

Even more, Gould seems to have forgotten that he first argued “correlation is not causation,” then argued that no amount of linear data could prove a point, and now is trying to do the inverse — using the linear data he claims is incorrect, show a low correlation score. Mind-numbing but it’ll impress the sophomores.

Having worked himself up to a point of sorts, Gould concludes with a quotation:

The tendency has always been strong to believe that whatever received a name must be an entity or being, having an independent existence of its own, and if no real entity answering to the name could be found, men did not for that reason suppose that none existed, but imagined that it was something particularly abstruse and mysterious.

Of course, he forgets the reflexive argument, which is that any bias must be analyzed, including one that purports to be the status quo, as his does. He offers us, for example, no evidence that IQ testing is invalid or that IQ does not correlate to intelligence or performance, but merely implies that in his wisdom he has seen such things.

So instead we are to take his assumption that we are all equal, which has a name as well, and although we cannot find proof for it or understand a mechanism by which diversely evolved organisms have identical abilities, we are to view it as Gospel.

Unimpressive.

See also: The Mismeasures of Gould, by J. Phillipe Rushton

|
Share on FacebookShare on RedditTweet about this on TwitterShare on LinkedIn