Implicit Association Test

Discovered an interesting psychological test called the IAT (Implicit Association Test).

The claim is that even if we exercise a strong control over conscious thoughts and attitudes, the truth is that we have subconscious biases that this test is able to ferret out.

It requires you to quickly sort through a variety of items and assign them to one of two columns.

The first set might have the columns "Male" and "Female" and you will be asked to sort male and female nouns and pronouns.

The next round may have the columns "Career" and "Family". Again you will be asked to sort items.

Finally, each column will get two categories, for example, "Male or Career" and "Female or Family" and you will have to sort 4 sets of items into those two columns.

The theory is that we subconsciously associate certain concepts, say male and career or female and family. When we mix up the categories to be "Male or Family" and "Female or Career", if we have a strong association between male and career, we will end up sorting more slowly (on the order of a few hundred milliseconds).

The claim is that no matter how many times you take the test or how you prep yourself, you cannot change the results, i.e. it is consistently reproducible.

You can take a number of tests here. Go to the Demonstration page and select the test you want. I have taken two and my results are as follows:

You have completed the Asian American-European American IAT.

The line immediately below summarizes the results of your task performance.

Your data suggest little or no association between Asian American and European American with American and Foreign.


You have completed the Gender-Science IAT.

The line immediately below summarizes the results of your task performance.

Your data suggest a moderate association of Male with Science and Female with Liberal Arts compared to Female with Science and Male with Liberal Arts.


I am planning to take the Gender-Science one again, because I think the result is wrong - I was expecting little or no association. Granted, in 40 years I have met only one woman who had a love of science similar to my own and only two men. However, it tends to be easier to find males who enjoy science and technology rather than females.

I think the test is breakable and open to influence by the test subject. I will let you know how it goes.

[Update: 27-October-2006 @ 08:50]

I took three more tests last night, repeating the Gender-Science one.

You have completed the Gender-Science IAT.

The line immediately below summarizes the results of your task performance.

Your data suggest a moderate association of Female with Science and Male with Liberal Arts compared to Male with Science and Female with Liberal Arts.


This was interesting. As far as I understand, the results are not supposed to change, since the bias is supposed to be constant. It is difficult for me to analyze since I don't have a quantitative measure of my score. Perhaps it indicates that I am truly neutral, or perhaps it indicates something else.

To rule out that timing was not my problem (you are supposed to complete the test as quickly as possible), I noted the estimated time for the subsequent tests and timed myself

You have completed the GAY PEOPLE-STRAIGHT PEOPLE IAT.

The line immediately below summarizes the results of your task performance.

Your data suggest a slight automatic preference for Straight People compared to Gay People.


Estimated time 5 minutes. Actual time 3:49.

You have completed the Judaism-Other Religions IAT.

The line immediately below summarizes the results of your task performance.

Your data suggest a moderate automatic preference for Judaism compared to Other Religions.


Estimated time: 5 minutes. Actual time 3:41

I may take some more follow up tests and repeat the ones I have already taken to get more data to analyze. I will also reveal in a few days what my strategy in taking the tests is (and why I think this may bias the results).

[Update 31-October-2006 @ 09:28]

What do I do that I believe skews the test? It is pretty simple. Given 4 categories, split into 2 column, I realized that I can't really keep them all in my head. So, I only focus on one column. Given the categories male, female, science and Liberal Arts, if they are grouped female-science and male-liberal arts, I only focus on only one column - in my case, the left column.

This means I don't care what the other words and images are. If it is not a female or science term, I simply place it in the other column. I reduce the problem from 4 categories and two columns to 2 categories and 1 column – if it doesn't fit, it goes in the other column..

I suspect this skews the results because I remove from my mind the categories male and Liberal Arts.

A more accurate test, I think, would be to have 3 columns and 6 categories because I would be unable to perform this simple binary partitioning. (Actually, you could. You would first partition one column against the other two. If it doesn't go in the first column, then you would partition the next two columns in the same way.)

Comments

KayMac said…
interested in where you are going w/ this
Richard said…
I am not sure I am going anywhere with it. I thought it was an interesting way to try and quantitatively measure subconscious bias. However, like all psychological measures, I suspect we can skew the outcome. It is the typical problem in quantum physics that the act of observing may change the result. It is the same in psychology, if subjects know they are being observed, then they may skew the results. Psychologists go to a lot of trouble to design tests that eliminate possible influence by the test subject on the results (whether intentional or inadvertent).

I am pretty certain I am not going to reveal some momentous secret.

Popular Posts