NotPoliticallyCorrect

Home » IQ » Test Validity, Test Bias, Test Construction, and Item Selection

Test Validity, Test Bias, Test Construction, and Item Selection

Enter your email address to follow this blog and receive notifications of new posts by email.

Join 223 other followers

Follow me on Twitter

Charles Darwin

Denis Noble

JP Rushton

Richard Lynn

L:inda Gottfredson

Goodreads

3400 words

Validity for IQ tests is fleeting. IQ tests are said to be “validated” on the basis of performance with other IQ tests and that of job performance (see Richardson and Norgate, 2015). Further, IQ tests are claimed to not be biased against social class or racial group. Finally, through the process of “item selection”, test constructors make the types of distributions they want (normal) and get the results the want through the subjective procedure of removing items that don’t agree with their pre-conceived notions on who is or is not “intelligent.” Lastly, “intelligence” is descriptive measure, not an explanatory concept, and treating it like an explanatory measure can—and does—lead to circularity (of which is rife in the subject of IQ testing; see Richardson, 2017b and Taleb’s article IQ is largely a psuedoscientific swindle). This article will show that, on the basis of test construction, item analysis (selection and deselection of items) and the fact that there is no theory of what is being measured in so-called intelligence tests that they, in fact, do not test what they purport to.

Richardson (1991: 17) states that “To measure is to give … a more reliable sense of quantity than our senses alone can provide”, and that “sensed intelligence is not an objective quantity in the sense that the same hotness of a body will be felt by the same humans everywhere (given a few simple conditions); what, in experience, we choose to call ‘more’ intelligence, and what ‘less’ a social judgement that varies from people to people, employing different criteria or signals.Richardson (1991: 17-18) goes on to say that:

Even if we arrive at a reliable instrument to parallel the experience of our senses, we can claim no more for it than that, without any underlying theory which relates differences in the measure to differences in some other, unobserved, phenomena responsible for those differences. Without such a theory we can never be sure that differences in the measure correspond with our sensed intelligence aren’t due to something else, perhaps something completely different. The phenomenon we at first imagine may not even exist. Instead, such verification most inventors and users of measures of intelligence … have simply constructed the source of differences in sensed intelligence as an underlying entity or force, rather in the way that children and naïve adults perceive hotness as a substance, or attribute the motion of objects to a fictitious impetus. What we have in cases like temperature, of course, are collateral criteria and measures that validate the theory, and thus the original measures. Without these, the assumed entity remains a fiction. This proved to be the case with impetus, and with many other naïve conceptions of nature, such as phlogiston (thought to account for differences in health and disease). How much greater such fictions are likely to be unobserved, dynamic and socially judged concepts like intelligence.

Richardson (1991: 32-35) then goes on to critique many of the old IQ tests, in that they had no way of being construct valid, and that the manuals did not even discuss the validity of the test—it was just assumed.

If we do not know what exactly is being measured when test constructors make and administer these tests, then how can we logically state that “IQ tests test intelligence”? Even Arthur Jensen admitted that psychometricians can create any type of distribution they please (1980: 71); he tacitly admits that tests are devised through the selection and deselection of items on IQ tests that correspond to the test constructors preconceived notions on what “intelligence” is. This, again, is even admitted by Jensen (1980: 147-148) who writes “The items must simply emerge arbitrarily from the heads of test constructors.

We know, to build on Richardson’s temperature example, that we know exactly is what being measured when we look at the amount of mercury in a thermometer. That is, the concept of “temperature” and the instrument to measure it (the thermometer) were verified independently, without circular reliance on the thermometer itself (see Hasok Chang’s 2007 book Inventing Temperature). IQ tests, on the other hand, are, supposedly, “validated” through measures of job performance and correlations with other, previous tests assumed to be (construct) valid—but they were, of course, just assumed to be valid, it was never shown.

For another example (as I’ve shown with IQ many times) of a psychological construct that is not valid is ASD (autism spectrum disorder). Waterhouse, London, and Gilliberg (2016) write that “14 groups of findings reviewed in this paper that together argue that ASD lacks neurobiological and construct validity. No unitary ASD brain impairment or replicated unitary model of ASD brain impairment exists.” That a construct is valid—that is, it tests what it purports to, is of utmost importance to test measurement. Without it, we don’t know if we’re measuring something else completely different from what we hope—or purport—to.

There is another problem: the fact that, for one of the most-used IQ tests that there is no underlying theory of item selection, as seen in John Raven’s personal notes (see Carpenter, Just, and Shell, 1990). Items on the Raven were selected based on Raven’s intuition, and not any formal theory—the same can be said about, of course, modern-day IQ tests. Carpenter, Just, and Shell (1990) write that John Raven “used his intuition and clinical experience to rank order the difficulty of the six problem types . . . without regard to any underlying processing theory.”

These preconceived notions on what “intelligence” is, though, fail without (1) a theory of what intelligence is (which, as admitted by Ian Deary (2001), there is no theory of human intelligence like the way physics has theories); and (2) what ultimately is termed “construct validity”—that a test measures what it purports to. There are a few kinds of validity: and what IQ-ists claim the most is that IQ tests have predictive validity—that is, they can predict an individual’s outcome in life, and job performance (it is claimed). However, “intelligence” is “a descriptive measure, not an explanatory concept … [so] measures of intelligence level have little or no predictive value” (Howe, 1988).

Howe (1997: ix) also tells us that “Intelligence is … an outcome … not a cause. … Even the most confidently stated assertions about intelligence are often wrong, and the inferences that people have drawn from those assertions are unjustified.

The correlation between IQ and school performance, according to Richardson (1991: 34)may be a necessary aspect of the validity of tests, but is not a sufficient one. Such evidence, as already mentioned, requires a clear connection between a theory (a model of intelligence), and the values on the measure.” But, as Richardson (2017: 85) notes:

… it should come as no surprise that performance on them [IQ tests] is associated with school performance. As Robert L. Thorndike and Elizabeth P. Hagen explained in their leading textbook, Educational and Psychological Measurement, “From the very way in which the tests were assembled [such correlation] could hardly be otherwise.”

Gottfredson (2009) claims that the construct validity argument against IQ is “fallacious”, noting it as one of her “fallacies” on intelligence testing (one of her “fallacies” was the “interactionism fallacy”, which I have previously discussed). However, unfortunately for Gottfredson (2009), “the phenomena that testers aim to capture” are built into the test and, as noted here numerous times, preconceived by the constructors of the test. So, Gottfredson’s (2009) claim fails.

Such kinds of construction, too, come into the claim of a “normal distribution.” Just like with preconceptions of who is or is not “intelligent” on the basis of preconceived notions, the normal distribution, too, is an artifact of test construction, along the selection and deselection of items to conform with the test constructors’ presuppositions; the “bell curve” of IQ is created by the presuppositions that the test constructors have about people and society (Simon, 1997).

Charles Spearman, in the early 1900s, claims to have found a “general factor” that explains correlations between different tests. This positive manifold he termed “g” for “general intelligence.” Spearman stated “The (g) factor was taken, pending further information, to consist in something of the nature of an ‘energy’ or ‘power’…” (quoted in Richardson, 1991: 38). The refutation of “g” is a simple, logical, one: While a correlation between performances “may be a necessary requirement for a general factor … it is not a sufficient one.” This is because “it is quite possible for quite independent factors to produce a hierarchy of correlations without the existence of any underlying ‘general’ factor (Fancer, 1985a; Richardson and Bynner, 1984)” (Richardson, 1991: 38). The fact of the matter is, Spearman’s “g” has been refuted for decades (and was shown to be reified by Gould (1981), and further defenses of his concepts on “general intelligence”, like by Jensen (1998) have been refuted, most forcefully by Peter Schonemann. Though, “g” is something built into the test by way of test construction (Richardson, 2002).

Castles (2013: 93) notes that “Spearman did not simply discover g lurking in his data. Instead, he chose one peculiar interpretation of the relationships to demonstrate something in which he already believed—unitary, biologically based intelligence.”

So what explains differences in “g”? The same test construction noted above along with differences in social class, due to stress, self-confidence, test preparedness and other factors correlated with social class, termed the “sociocognitive-affective nexus” (Richardson, 2002).

Constance Hilliard, in her book Straightening the Bell Curve (Hilliard, 2012), notes that there were differences in IQ between rural and urban white South Africans. She notes that differences between those who spoke Afrikaans and those who spoke another language were completely removed through test construction (Hilliard, 2012: 116). Hilliard (2012) notes that if the tests that the constructors formulate don’t agree with their preconceived notions, they are then thrown out:

If the individuals who were supposed to come out on top didn’t score highly or, conversely, if the individuals who were assumed would be at the bottom of the scores didn’t end up there, then the test designers scrapped the test.

Sex differences in “intelligence” (IQ) have been the subject of some debate in the early-to-mid-1900s. Test constructors debated amongst themselves what to do about such differences between the sexes. Hilliard (2012) quotes Harrington (1984; in Perspectives on Bias in Mental Testing) who writes about normalizing test scores between men and women:

It was decided [by IQ test writers] a priori that the distribution of intelligence-test scores would be normal with a mean (X=100) and a standard deviation (SD=15), also that both sexes would have the same mean and distribution. To ensure the absence of sex differences, it was arranged to discard items on which the sexes differed. Then, if not enough items remained, when discarded items were reintroduced, they were balanced, i.e., for every item favoring males, another one favoring females was also introduced.

While Richardson (1998: 114) notes that test constructors had two choices when looking at sex differences in the items they administered to the sexes:

One who would construct a test for intellectual capacity has two possible methods of handling the problem of sex differences.
1 He may assume that all the sex differences yielded by his test items are about equally indicative of sex differences in native ability.
2 He may proceed on the hypothesis that large sex differences on items of the Binet type are likely to be factitious in the sense that they reflect sex differences in experience or training. To the extent that this assumption is valid, he will be justified in eliminating from his battery test items which yield large sex differences.
The authors of the New Revision have chosen the second of these alternatives and sought to avoid using test items showing large differences in percents passing. (McNemar 1942:56)

Change “sex differences” to “race” or “social class” differences and we can, too, change the distribution of the curve, along with notions of who is or is not “intelligent.” Previously low scorers can, by way of test construction, become high scorers, vice-versa for high scorers being made into low scorers. There is no logical—or empirical—justification for the inclusion of specific items on whatever IQ test is in question. That is, to put it another way, the inclusion of items on a test is subjective, which comes down to the test designers’ preconceived notions, and not an objective measure of what types of items should be on the test—as Raven stated, there is no type of underlying theory for the inclusion of items in the test, it is based on “intuition” (which is the same thing that modern-day test constructors do). These two quotes from IQ-ists in the early 20th century are paramount in the attack on the validity of IQ tests—and the causes for differences in scores between groups.

He and van de Vijver (2012: 7) write that “An item is biased when it has a different psychological meaning across cultures. More precisely, an item of a scale (e.g., measuring anxiety) is said to be biased if persons with the same trait, but coming from different cultures, are not equally likely to endorse the item (Van de Vijver & Leung, 1997).” Indeed, Reynolds and Suzuki (2012: 83) write that “Item bias due to“:

… “poor item translation, ambiguities in the original item, low familiarity/appropriateness of the item content in certain cultures, or influence of culture specifics such as nuisance factors or connotations associated with the item wording” (p. 127) (van de Vijver and Tanzer, 2004)

Drame and Ferguson (2017) note that their “Results indicate that use of the Ravens may substantially underestimate the intelligence of children in Mali” while the cause may be due to the fact that:

European and North American children may spend more time with play tasks such as jigsaw puzzles or connect the dots that have similarities with the Ravens and, thus, train on similar tasks more than do African children. If African children spend less time on similar tasks, they would have fewer opportunities to train for the Ravens (however unintentionally) reflecting in poorer scores. In this sense, verbal ability need not be the only pitfall in selecting culturally sensitive IQ testing approaches. Thus, differences in Ravens scores may be a cultural artifact rather than an indication of true intelligence differences. [Similar arguments can be found in Richardson, 2002: 291-293]

The same was also found by Dutton et al (2017) who write that “It is argued that the undeveloped nature of South Sudan means that a test based around shapes and analytic thinking is unsuitable. It is likely to heavily under-estimate their average intelligence.” So if the Raven has these problems cross-culturally (country), then it SHOULD have such biases within, say, America.

It is also true that the types of items on IQ tests are not as complex as everyday life (see Richardson and Norgate, 2014). Types of questions on IQ tests are, in effect, ones of middle-class knowledge and skills and, knowing how IQ tests are structured will make this claim clear (along with knowing the types of items that eventually make it onto the particular IQ test itself). Richardson (2002) has a few questions on modern-day IQ tests whereas Castles (2013), too, has a few questions from the Stanford-Binet. This, of course, is due to the social class of the test constructors. Some examples of some questions can be seen here:

‘What is the boiling point of water?’ ‘Who wrote Hamlet?’ ‘In what continent is Egypt?’ (Richardson, 2002: 289)

and

‘When anyone has offended you and asks you to excuse him—what ought you do?’ ‘What is the difference between esteem and affection?’ [this is from the Binet Scales, but “It is interesting to note that similar items are still found on most modern intelligence tests” (Castles, 2013).]]

Castles (2013: 150) further notes made-up examples of what is on the WAIS (since she cannot legally give questions away since she is a licensed psychologist), and she writes:

One section of the WAIS-III, for example, consists of arithmetic problems that the respondent must solve in his or her head. Others require test-takers to define a series of vocabulary words (many of which would be familiar only to skilled-readers), to answer school-related factual questions (e.g., “Who was the first president of the United States?” or “Who wrote the Canterbury Tales?”), and to recognize and endorse common cultural norms and values (e.g., “What should you do it a sale clerk accidentally gives you too much change?” or “Why does our Constitution call for division of powers?”). True, respondents are also given a few opportunities to solve novel problems (e.g., copying a series of abstract designs with colored blocks). But even these supposedly culture-fair items require an understanding of social conventions, familiarity with objects specific to American culture, and/or experience working with geometric shapes or symbols.

All of these factors coalesce into forming the claim—and the argument—that IQ tests are one of middle-class knowledge and skills. The thing is, contrary to the claims of IQ-ists, there is no such thing as a culture-free IQ test. Richardson (2002: 293) notes that “Since all human cognition takes place through the medium of cultural/psychological tools, the very idea of a culture-free test is, as Cole (1999) notes, ‘a contradiction in terms . . . by its very nature, IQ testing is culture bound’ (p. 646). Individuals are simply more or less prepared for dealing with the cognitive and linguistic structures built in to the particular items.

Cole (1981) notes that “that the notion of a culture free IQ test is an absurdity” because “all higher psychological processes are shaped by our experiences and these experiences are culturally organized” (this is a point that Richardson has driven home for decades) while also—rightly—stating that “IQ tests sample school activities, and therefore, indirectly, valued social activities, in our culture.

One of the last stands for the IQ-ist is to claim that IQ tests are useful for identifying at-risk individuals for learning disabilities (as Binet originally created the first IQ tests for). However, it is noted that IQ tests are not necessary—nor sufficient—for the identification of those with learning disabilities. Siegal (1989) states that “On logical and empirical grounds, IQ test scores are not necessary for the definition of learning disabilities.

When Goddard brought the first IQ tests to America and translated them into English from French is when the IQ testing conglomerate really took off (see Zenderland, 1998 for a review). These tests were used to justify current social ranks. As Richardson (1991: 44) notes, “The measurement of intelligence in the twentieth century arose partly out of attempts to ‘prove’ or justify a particular world view, and partly for purposes of screening and social selection. It is hardly surprising that its subsequent fate has been one of uncertainty and controversy, nor that it has raised so many social and political issues (see, for example, Joynson 1989 for discussion of such issues).” So, what actual attempts at validation did the constructors of such tests need in the 20th century when they knew full-well what they wanted to show and, unsurprisingly, they observed it (since it was already going to happen since they construct the test to be that way)?

The conceptual arguments just given here point to a few things:

(1) IQ tests are not construct valid because there is no theory of intelligence, nor is there an underlying theory which relates differences in IQ (the unseen function) to, for example, a physiological variable. (See Uttal, 2012; 2014 for arguments against fMRI studies that purport to show differences in physiological variables cognition.)

(2) The fact that items on the tests are biased against certain classes/cultures; this obviously matters since, as noted above, there is no theory for the inclusion of items, it comes down to the subjective choice of the test designers, as noted by Jensen.

(3) ‘g’ is a reified mathematical abstraction; Spearman “discovered” nothing, he just chose the interpretation that, of course, went with his preconceived notion.

(4) The fact that sex differences in IQ scores were seen as a problem and, through item analysis, made to go away. This tells us that we can do the same for class/race differences in intelligence. Score differences are a function of test construction.

(5) The fact that the Raven has been shown to be biased in two African countries lends credence to the claims here.

So this then brings us to the ultimate claim of this article: IQ tests don’t test intelligence; they test middle-class knowledge and skills. Therefore, the scores on IQ tests are not that of intelligence, but of an index of one’s cultural knowledge of the middle class and its knowledge structure. This, IQ scores are, in actuality, “middle-class knowledge and skills” scores. So, contra Jensen (1980), there is bias in mental testing due to the items chosen for inclusion on the test (we have admission that score variances and distributions can change from IQ-ists themselves)


3 Comments

  1. dealwithit says:

    what, in experience, we choose to call ‘more’ intelligence, and what ‘less’ a social judgement that varies from people to people, employing different criteria or signals.

    that’s a LIE. it’s no more subjective than beauty, which is NOT subjective.

    rr is just sounding dumb. he’s resting at camp 3 not knowing there’s more mountain to climb.

    Like

  2. dealwithit says:

    this article is so tiresome i stopped half way through. the argument that IQ tests don’t measure intelligence or that intelligence is subjective are juvenile and totally irrelevant.

    Previously low scorers can, by way of test construction, become high scorers, vice-versa for high scorers being made into low scorers.

    FALSE.

    There is no logical—or empirical—justification for the inclusion of specific items on whatever IQ test is in question.

    THERE DOESN’T NEED TO BE.

    rr’s “arguments” apply to all psychology and psychiatry and the use of any psychological vocabulary at all. “happy” vs “suicidal” is meaningless because can’t be measured by thermometer. fucking retarded.

    how many times do i have to say it? the fact that psychiatry isn’t a science doesn’t mean there aren’t crazy people. the fact that IQ isn’t a thing doesn’t mean there’s no such thing as intelligence.

    intelligence and lack of it are everyday phenomena.
    if one wished to operationalize it the large and diverse test battery (not the ravens) is the best way to do it.

    Like

  3. dealwithit says:

    italian women are no more beautiful than pygmy women because can’t be measured by thermometer and not construct valid [richarsdon 2007]

    arnold not better bodybuilder than modern bubblegutbuilders because can’t be measured by scale [richardson 2008]

    good and bad taste are not construct valid because no theory. ice cream and feces are equally delicious. [richardson 2009]

    sad.

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Please keep comments on topic.

Jean Baptiste Lamarck

Eva Jablonka

Charles Murray

Arthur Jensen

Blog Stats

  • 599,387 hits
Follow NotPoliticallyCorrect on WordPress.com

suggestions, praises, criticisms

If you have any suggestions for future posts, criticisms or praises for me, email me at RaceRealist88@gmail.com
%d bloggers like this: