Study supports validity of test that indicates widespread unconscious bias

In the decade since the Implicit Association Test was introduced, its most surprising and controversial finding is its indication that about 70 percent of those who took a version of the test that measures racial attitudes have an unconscious, or implicit, preference for white people compared to blacks. This contrasts with figures generally under 20 percent for self report, or survey, measures of race bias.

A new study published this week validates those findings, showing that the Implicit Association Test, a psychological tool, has validity in predicting behavior and, in particular, that it has significantly greater validity than self-reports in the socially sensitive topics of race, gender, ethnicity, sexual orientation and age.

Anthony G. Greenwald

The research, published in the Journal of Personality and Social Psychology, is an overview and analysis of 122 published and unpublished reports of 184 different research studies. In this analysis, 85 percent of the studies also included self-reporting measures of the type generally used in surveys. This allowed the researchers, headed by University of Washington psychology Professor Anthony Greenwald, to compare the test's success in predicting social behavior and judgment with the success of self-reports.

"In socially sensitive areas, especially black-white interracial behavior, the test had significantly greater predictive value than self-reports. This finding establishes the Implicit Association Test's value in research to understand the roots of race and other discrimination," said Greenwald. "What was especially surprising was how ineffective standard self-report measurers were in the areas in which the test measures have been of greatest interest – predicting interracial behavior."

Greenwald created the Implicit Association Test in 1998 and he and Mahzarin Banaji, a Harvard psychology professor, and Brian Nosek, a University of Virginia associate professor of psychology, further developed it. Since then the test has been used in more than 1,000 research studies around the world. More than 10 million versions of the test have been completed at an Internet site where they are available as a self-administer demonstration.

The research looked at studies covering nine different areas – consumer preference, black-white interracial behavior, personality differences, clinical phenomena, alcohol and drug use, non-racial intergroup behavior, gender and sexual orientation, close relationships and political preferences.

Findings also showed that:

* Across all nine of these areas, measures of the test were useful in predicting social behavior.
* Both the test, which is implicit, and self-reports, which are explicit, had predictive validity independent of each other. This suggests the desirability of using both types of measure in surveys and applied research studies.
* In consumer and political preferences both measures effectively predicted behavior, but self-reports had significantly greater predictive validity.

Studies in the research came from a number of countries including Germany, the Netherlands, Italy, the United Kingdom, Australia, Canada, Poland and the United States. They looked at such topics as attitudes of undecided voters one-month prior to an Italian election; treatment recommendations by physicians for black and white heart attack victims; and reactions to spiders before and after treatment for arachnophobia, or spider phobia.

"The Implicit Association Test is controversial because many people believe that racial bias is largely a thing of the past. The test's finding of a widespread, automatic form of race preference violates people's image of tolerance and is hard for them to accept. When you are unaware of attitudes or stereotypes, they can unintentionally affect your behavior. Awareness can help to overcome this unwanted influence," said Greenwald. ###

Co-authors of the new study are Banaji, T. Andrew Poehlman of Southern Methodist University and Eric Uhlmann of Northwestern University. The National Science Foundation, National Institute of Mental Health, the Third Millennium Foundation and the Rockefeller Foundation funded the research.

For more information, contact Greenwald at 206-543-7227 or agg@u.washington.edu; Banaji at banaji@fas.harvard.edu Nosek at 434-924-0666 or nosek@virginia.edu.

More information about the Implicit Association Test and examples of the test can be found at Project Implicit's Web site: implicit.harvard.edu/implicit/

Contact: Joel Schwarz joels@u.washington.edu 206-543-2580 University of Washington