Back to Instructor's Page

Introduction
Method
Analysis
References

An Implicit Association Test (IAT) measures the relative ease with which people are able to make associations between certain groups of people, such as older adults, and the concepts of "good" and "bad." Ease of association, measured by judgment speed, is taken as evidence for an implicitly-held attitude toward that social group. For example, the finding that people are quicker to associate "good" with "young" compared to "old" is interpreted as evidence of a generally-held bias in favor of youth.  The version of an IAT given here measures implicit associations regarding race, which was one of the first topics investigated by the developers of the IAT--Anthony Greenwald of the University of Washington and Mahzarin Banaji of Yale University. Their Yale website http://buster.cs.yale.edu/implicit/info1.html offers more IAT examples and links to articles that deal with the test and its interpretations.

Method

Task

In the IAT used here, research participants complete a series of five tasks.  In each, they categorize sets of faces and words. The faces are of Blacks  and Whites and the words are ones that have either a good or bad connotation.  Figure 1 shows a screen shot of a trial from a task in which a face is to be categorized as "European-American"  or "African-American." Categorization is made using the 1-key to indicate the face belongs to the category on the left and the 2-key to indicate that it belongs to the category on the right. The words that are to be categorized (e.g., Agony, Friend, War, Laughter)  have connotations that are either good or bad.  The categorization task is to again use the 1-key and 2-key to indicate which category--Good or Bad--the word belongs to.  An example is given in Figure 2..

Figure 1: Phase 1 Trial

Figure 2: Phase 2 Trial

Three of the five tasks that make up the IAT are like the ones above.  The other two are tests which combine the face and word categorization tasks.  In these, a single key is used to make face categorizations and word categorizations.  Thus the 1-key might be used for European-American faces and Good words, African-American faces and Good words, or any of the other combinations of Face and Word categories.  The data from these two tests can be combined to determine how the category combinations affect speed of performance.  The screen shot below is from a test trial in which European-American is paired with Good and African-American with Bad.  This type of test trial is called a "stereotype congruent trial" because the racial stereotypes that have been held historically by the majority culture in the USA have assigned negative attributes to Blacks and positive attributes to Whites. The other type of test trial is a "stereotype non-congruent trial", which is one on which European-American is paired with Bad and African-American with Good.

Figure 3: Phase 3 Trial

In Phase 1 of the experiment, participants complete a 16-face categorization task (8 Whites and 8 Blacks). In Phase 2 they complete a 16-word categorization task (8 Good words and 8 Bad).  In these first two tasks, the 1- and 2-keys are assigned randomly as the keys to use for the two Face and Word categories.  In Phase 3, a test is given using the same 16 faces and 16 words as in Phases 1 and 2.  The 32 items on the test (16 words plus 16 faces) are presented in random order and participants use the same key assignments for the categories that they used in Phases 1 and 2.  This is a stereotype congruent test when a common key is used for European-American and Good and it is stereotype non-congruent when a common key is used for European-American and Bad.  Given the random assignment of keys to categories in Phases 1 and 2, approximately half of the participants receive a congruent test in Phase 3 and half receive a non-congruent test.  Phase 4 consists of a second face categorization task, but this time the key assignments are reversed from what they were in Phase 1.  If the 2-key was assigned to the African-American category, then in Phase 3 the 1-key is used for this category.  The final phase, Phase 5, consists of the second test using common keys for paired categories. This time, however, the key assignments are consistent with the assignment from Phase 2 and Phase 4 (the reversed faces phase) and, therefore, if the first test was congruent the second is non-congruent and vice versa.
« Return to Top»

The dependent variable of primary interest is called the IAT Score, which is a measure of the time difference between time taken to complete the stereotype non-congruent test and the stereotype congruent test.  It is computed by subtracting the congruent test time from the non-congruent test time. If this value is positive, it means that completing the non-congruent test took longer.  Objectively, this implies that it was harder to use a common response key to categorize stereotype non-congruent categories (e.g., European-American and Bad) than to use a common key for congruent categories.  Past research has found that white Americans on average have positive IAT scores on the IAT-Race task, even when they are totally unprejudiced in their overt attitudes and behavior.  To know whether the data collected at PsychExperiments replicate this finding, students need to compute the average congruent and non-congruent times and then compare them using a related samples t-test.  Alternatively, they could do a one-sample t-test using the IAT scores to test the hypothesis that the population mean IAT score is zero. The two tests are equivalent and will indicate whether the observed difference is considered statistically significant. 

Students wishing to do additional data analyses might want to consider the effect of test order on IAT score.  The developers of the IAT have found that order is relevant; therefore, they take test order into account when reporting results at their website.  The question to answer is whether test times differ as a function of test order. 

Additional variables of theoretical interest are the measures labeled Preference and Warmth.  These items are somewhat redundant and provide participants self-appraisal of their racial attitudes.  The Preference score is a number 1 to 5 (1=I strongly prefer Whites to Blacks, 3= I like Whites and Blacks equally, 5=I strongly prefer Blacks to Whites) that indicates the choice that was made to the 5-choice item "Which choice best describes you?"  The Warmth number is the difference between the two ratings made on the Black and White scales used to provide a response to the item "Please rate how warm or cold you feel toward the following groups (0 = coldest feelings, 5 = neutral, 10 = warmest feelings)". If the difference score is positive it means the Warmth score was greater for Whites than Blacks.  Negative scores imply more warmth toward Blacks and scores of 0 indicate equal warmth toward both races.  On both items, participants have the option of declining to respond, so some values will be recorded as "Decline."

The data obtained via the Download data link on the PsychExperiments homepage are comma delimited.  The first entry (up to the first comma) is the affiliation of the research participant.  Typically this is the class to which the participant belongs.  Non-affiliated participants use the default entry of Interested Person. Next is the randomly assigned ID code for the research participant. The third entry is the date on the server when the data were received. The fourth is a measure of "experiment time" in seconds.  This is how long it took the participant to complete the experiment, beginning with Trial 1 and ending with the last trial.  (Time spent reading instructions and so forth are not included).  The measure is useful as a screen for "bad" data.  The time measure should be appropriate to the task--not too fast and not too slow.

The task data consist of 32 stereotype congruent trials and 32 stereotype non-congruent trials.  The first data entry for each trial is a trial number (1-32).  This is followed by a numerical entry for the response (1-5) to the preference item, a numerical value (+10 to -10) which indicates Warmth toward Whites relative to Blacks, a numerical value for test order (1=Congruent test first, 2=Non-congruent test first), a text entry for the TestType (C=Congruent, NC = non-congruent), then a text entry for Stimulus type ("white," or "black" when the stimuli are faces and "good" or "bad"when the stimuli are words). The last entry is the time taken to complete the test trial.  This time is measured using Authorware's TimeInInteraction system variable.  It reflects the time from the appearance of a test item on the screen until a correct keypress is made.  

The table below gives the data format (excluding user information) The text line below that shows a complete data set for one research participant.

ATI Workshop--CSUF,BEH683,8/2/2002,13:16,1,3 ,0 ,1,C ,Good ,1.377,1,3 ,0 ,1,NC ,Good ,1.324,2,3 ,0 ,1,NC ,white ,1.255,2,3 ,0 ,1,C ,Good ,1.009,3,3 ,0

Last revised:November 16, 2004 11:29:37 AM
Copyright© 1995 The University of Mississippi. All rights reserved.
Questions about this page?PsychExps