The Implicit Association Test
It's an interesting exercise in cognitive psychology, to be sure, but when it comes to interpreting the results of the test there seem to be several pretty basic unstated and unexamined assumptions at play:
- That performance of a task at this level of abstraction says anything significant about real-world attitudes or behavior (and if there is a correlation, have the causal order considerations and other statistical and methodological issues been worked out?);
- That differences in performance times are material for tasks where all performance times are very short;
- That the test measures closeness of cognitive association and not the relative speed at which the subject sorts things of greater and lesser familiarity;
- That a closer cognitive association, if that is what is being measured, reflects a preference;
- That the way the white and black faces are shown (as close-cropped squares showing little but the main features of eyes, nose and mouth) and the tonalities of the faces presented (they seem to be presented to minimize tonal contrast, probably so it is harder for a subject to key on skin tone alone) will not influence how strong an effect the test indicates in some or many instances;
- That if slower performance times mean a subject is using "higher" cognitive functions to perform part of the sorting task, the "higher cognitive functions" are overriding some "baser instinct" in the opposite direction instead of just performing an extra check--even if both sets of mental processes arrived at the same end result, it would presumably still take longer;
- That the subject's knowledge of what is purported to be tested does not affect the subject's performance (there seems to be an assumption that because it's hard to "beat" the test, consciousness of what is being tested does not matter--but every subject, in the online version anyway, is conscious of what is being tested), for example by setting the threshold for reverting to "higher cognitive functions" higher than it would otherwise be. A related idea, with apologies to Justice Stone, is that parts of our brains recognize what might be called "suspect classifications" and automatically apply "strict scrutiny" to whatever is bubbling up from the nether reaches of the psyche. This is not a bad thing, nor is it a controversial notion in other connections.
I think there is a lot to be said for the idea that what they are really testing is how quickly your unconscious brain sorts items in categories of greater or lesser familiarity. Most people taking the test in North America are likely to have seen a lot more white people than black people, and they probably have encountered the "good" words more regularly than the "bad" ones (speculation on my part, but I think it's reasonable--to keep it that way, limit your viewing of this). It would be interesting to test this theory by replacing the "good" and "bad" word category with some other value-neutral one. There is probably a way it could be done with the names of colors, where I would expect to find more familiar names like green and red being sorted faster than words like chartreuse and magenta. Would an IAT with those words show that people have an implicit association between white people and common names for colors? I think it probably would. By choosing value-laden words and assuming test subjects are equally accustomed to encountering the "good" ones and the "bad" ones, I think the authors of the IAT may be missing an important potential flaw in their methodology.
Here was the most interesting part of the online IAT's FAQ section:
When will implicit attitudes agree with explicit attitudes?Only two reasons? Isn't it begging the question even to call the implicit association an "attitude"? Isn't there an assumption here that the implicit association (if that is what is really being tested) is somehow always the more authentic one? Just because it's possible to be deceptive about some things and not others, it does not follow that people don't usually tell the truth about things they could lie about, or that the things you can't lie about represent a more fundamental truth merely because they can't be falsified. Also, isn't it possible--for some or all of the reasons listed above and others you may think of--that the implicit association data is just meaningless? At least when someone says "I like Turks," you know that they are either being honest or dishonest. When an IAT tells you something about someone and Turks, there seems to me almost no telling what that might be.
Answer: There are two reasons why direct (explicit) and indirect (implicit) attitudes may not be the same. The simpler explanation is that a person may be unwilling to accurately report some attitude. For example, if a professor asks a student "Do you like soap operas?" a student who is fully aware of spending two hours each day watching soap operas may nevertheless say "no" because of being embarrassed (unwilling) to reveal this fondness. The second explanation for explicit-implicit disagreement is that a person may be unable to accurately report an attitude. For example, if asked "Do you like Turks?" many Germans will respond "yes" because they regard themselves as unprejudiced. However, an IAT may reveal that these same Germans have automatic negative associations toward Turks. (This IAT result has been demonstrated quite clearly in Germany.) Germans who show such a response are unaware of their implicit negativity and are therefore unable to report it explicitly. The unwilling-unable distinction is like the difference between hiding something from others and something being hidden from you. So, implicit and explicit attitude measures should agree when people are both aware of and willing to report their automatic associations. For example, in the experiment that measures attitudes toward flowers and insects, respondents are both able and willing to report these attitudes. In that experiment, implicit and explicit measures generally reveal the same attitudes. Likewise, college students are able and willing to report whether they like science or arts. In such a case also, implicit and explicit attitudes reveal the same preferences.
The author of the Slate article, through statements like "The elegance of Banaji's test is that it doesn't let you lie" and "The IAT, then, is an objective measure of bias" shows that he didn't give much thought to the kind of considerations I've raised in this post. As I recall, Malcolm Gladwell had a more thoughtful treatment of the same subject in "Blink," informed in part by getting a "biased" score on the IAT (his mother is black).
I should not in closing that the authors of the IAT make a large number of research papers available though a page of their website, and those papers quite possibly address some or all of the concerns I've expressed here.
posted at 8:19 PM
0 Comments:
Post a Comment
<< Home