Anthony Greenwald

"Using Large Data Sets to Improve Validity of the Implicit Association Test - A Latency Based Cognitive Measure"

Presented to the Center for Statistics and the Social Sciences, University of Washington, 23 October 2002.

The Implicit Association Test (IAT) provides a measure of strengths of associations among concepts, such as the association of positive or negative valence with stigmatized social categories (racial or ethnic groups, elderly, etc.) The measure of association strength test is computed from latencies of responses to a series of instances of the concepts. Virtues of the measure are that the respondent (a) need not be aware of the associations, (b) cannot readily control the manifestations of association strengths that the test elicits, and (c) is often surprised and either enlightened or disturbed (or both) by these manifestations. Four data sets each of about 10,000 respondents X 200 responses were used to explore alternative methods of combining the IAT's latency and error data into association strength measures, with the objective of maximizing construct validity of the measure.