Over a decade ago, I devised a test for detecting attitudes and biases operating below the level of a person’s awareness.
Known as the Implicit Association Test, or IAT, it is presently the most widely used of the measures of implicit attitudes that have been developed by social psychologists over the past 25 years. It has been self-administered online by millions, many of whom have been surprised—sometimes unpleasantly—by evidence of their own unconscious attitudes and stereotypes regarding race, age, gender, ethnicity, religion, or sexual orientation.
Now it is my turn to be surprised—pleasantly. The test has been used for a purpose that I long imagined as possible, but never dared attempt, knowing that it needed the attention of psychologists who focus on romantic relationships. The results suggest that the IAT is effective in predicting which romantic relationships will last.
The report, just published in the journal Psychological Science, is provocatively titled “Assessing the Seeds of Relationship Decay.” In it, three psychologists at the University of Rochester -- Soonhee Lee, Ronald Rogge, and Harry Reis—describe their research predicting relationship breakup. They recruited participants by many means, including referrals by psychology faculty and various Internet sources. The mostly female participants were married, engaged, or otherwise in exclusive, committed relationships.
The research started with the collection of several measures—not only the IAT, but also some established questionnaire measures of relationship quality—all of which might be useful predictors of breakup. Of the 222 participants who started, 116 were successfully re-contacted to obtain reports on the status of their relationships at various times up to 12 months later.
Nineteen (16%) of the re-contacted participants reported that a breakup had occurred. Remarkably, the IAT measure of a subject's attitude toward her partner did a better job of predicting the breakup than did several questionnaire measures of relationship quality.
The authors concluded that the questionnaire measures might have been ineffective either because participants were unaware of negative attitudes toward their partners or perhaps because they knew about them but were unwilling to report them. If that’s correct, the IAT worked because it depends on neither awareness of the attitude nor willingness to report it.
What exactly is the IAT, and how does it tap into mental processes that can operate outside of awareness?
First, an invitation: Come try it for yourself here. Select the “Demonstration” option to choose among more than a dozen IATs. Or you can select the “Research” option to be assigned at random to one of a varying set of academic research studies hosted at Project Implicit’s site.
Most IATs are computer administered, and take less than ten minutes to complete. The measure produced by the IAT is called “implicit” because it does not depend on awareness of what is being measured. The IAT asks no questions. It generally involves two tasks, which I’ll call A and B, that call for rapid keyboard responses to words or images presented on the computer’s display screen.
The version of the IAT that is most widely known is one that measures implicit racial attitudes, so let’s use that one as our example.
Task A requires a left-key response to pictures of Black faces and pleasant-meaning words—such as peace and laughter—and a right-key response to pictures of White faces and unpleasant-meaning words—such as evil and failure.
Task B is similar, except for switching sides for the Black and White faces: the left key is now for White faces and pleasant words and the right key for Black faces and unpleasant words.
Respondents are randomly assigned to do A or B first. Most respondents (American or other) find B to be much the easier. That is, when White faces and pleasant words get the same response they respond much more rapidly than when White faces and unpleasant words get the same response.
The IAT is interpreted using the assumption that faster performance in one task means that the categories assigned to the same key in that task are more strongly associated. For the race IAT, faster performance in Task B thus implies stronger association of White than Black with pleasant. That result has been described as showing an “automatic preference” for White relative to Black.
There has been debate about interpretation, but many accept this IAT procedure as a measure of a form of race attitude that differs from what is picked up by questionnaire measures. A recent meta-analysis has established that the race attitude IAT does a better job of predicting racially discriminatory judgments and behavior than do standard questionnaire measures of race bias.
The implicit measure used in the Rochester study—known as the Go/No-go Association Test (GNAT)—was developed by Brian Nosek and Mahzarin Banaji, two leading IAT researchers. Unlike the four-category structure of the standard IAT (the Race Attitude IAT described before is an example) this measure has just three categories: pleasant words, unpleasant words, and a set of partner-specific items, including name and pet name. Task A pairs the partner items with pleasant words; Task B pairs the partner items with the unpleasant words.
In case you'd like to put your own relationship to this test, alas, that's premature. The test needs more research to confirm that it is superior to picking petals off a daisy, which it very likely is. There's also the problem that it must be programmed individually for every use, to incorporate the partner-specific items. Eventually, someone may make a pretty penny marketing the test's concept.
An interesting alternative to the 3-category task used by Lee, Rogge, and Reis is a standard 4-category IAT that adds a set of "self" items. An indicator of a durable relationship might be the pattern of responding as rapidly to the task that pairs your partner with pleasant words as to the one that pairs yourself with pleasant words.
I haven’t had time yet to create and try this IAT—one in which my wife and I, in effect, square off in a contest for my IAT-measured affection. I will learn something interesting when I do try this one. Among the many IATs I have taken, this will be the first one for which I have some trepidation about learning the result (and do I dare ask my wife to try it?). That anxiety may be the most telling indication of how much I have come to trust the IAT as a valid measure.