The Implicit Association Test
Source: Julian Wills & Jay Van Bavel—New York University
One of the core constructs in social psychology is the notion of an attitude toward an object…
1. Participant Recruitment
- Conduct a power analysis and recruit enough participants to obtain sufficient statistical power to detect the observed effect size and then obtain informed consent.
2. Data Collection
- Assemble a list of 12 words with pleasant—good—associations (e.g., happy, lucky, gift) and 12 words with unpleasant—bad— associations (e.g., hatred, disaster, poison).
- Assemble 12 European American and 12 African American faces (half male, half female), cropped at the chin and forehead.
- Create a stimulus presentation script with five blocks (i.e., sequences).
- The script should only allow three keyboard responses: (1) ‘SPACEBAR’ for advancing instructions, (2) ‘E’ to select anchors on the left, and (3) ‘I’ to select anchors on the right.
- In addition to recording the participant’s key presses, the script should also record their response latency (i.e., time between each stimulus presentation and response provided).
- Block 1: Initial target-concept discrimination. Setup the race anchors so that Black is displayed on one side of the screen and White is displayed on the opposite side. The specific orientation should be counterbalanced across subjects (i.e., half of the subjects should see Black on the left and White on the right). This block consists of 40 trials total: The first 20 are practice trials, whereas data are analyzed in the latter 20 trials. Randomly sample (without replacement) the fifty faces so that each is displayed exactly twice (once for practice trials, once for real trials). Separate the display of each trial by an inter-trial delay of 100 ms, 400 ms, or 700 ms (randomly chosen each trial).
- Block 2: Associated attribute discrimination. Setup the valence anchors so that pleasant is displayed on one side of the screen and unpleasant is displayed on the opposite side. This block has all the same characteristics as Block 1 (i.e., counterbalancing, number of trials, inter-trial delay) except that the valence words (e.g., good, bad) are displayed instead of faces.
- Block 3: Initial combined task. Display both the race and valence anchors using the same orientation as the preceding blocks (i.e., if Black was displayed on the right for Block 1 then it should also appear on the right for this block). Present each of the 12 faces and 12 valence words at least twice, for a total of 40 trials. Use the same inter-trial delays from the preceding blocks.
- Block 4: Reversed target-concept discrimination. Remove the valence anchors and swap the race anchors so that Black is now displayed on the screen where White was originally displayed (and vice versa). Otherwise, this block retains all the characteristics of Block 1.
- Block 5: Reversed combined task. This block is identical to Block 3 except that the race anchors are now in the same position as they were in Block 4. Block 3 and 5 should be counterbalanced across participants.
- For all blocks, instruct participants to classify the faces/terms as quickly as possible, based on the category they are associated with.
- After completing the computer-administered IAT tasks, distribute several questionnaires that measure race-related attitudes and beliefs.
- To allow participants to know that they have privacy, have them complete these questionnaires in personal experimental rooms. Also inform them that they will be placing their completed questionnaires in an unmarked envelope before returning them to the experimenter.
- The surveys should include a feeling thermometer and semantic differential measures targeted at the racial concepts of Black and White—the Modern Racism Scale2—as well as the Diversity and Discrimination Scales.3
- Use 7-point scales for each of the five semantic dimensions with the following polar-opposite adjective pairs as anchors: beautiful–ugly, good–bad, pleasant–unpleasant, honest–dishonest, nice–awful.
- Instruct participants to rate items from all four object categories using these five semantic dimensions.
- Instruct participants to mark the middle of the scale if they considered both anchoring adjectives to be irrelevant to the category.
- Debriefing: To complete the study, tell the participant about the exact nature of the study.
3. Data Analysis
- Recode reaction times less than 300 ms to 300 ms, as well as reaction times longer than 3000 ms to 3000 ms, so that these extreme observations do not unduly influence the analysis.
- Because the reaction time data is positively skewed, log transform all reaction times data so that it is more normally distributed.
- Then compare the mean reaction times on Block 3 with Block 5.
- Subtract these scores to compute an index for the IAT effect.
- Positive scores reflect an implicit preference for Black versus White (i.e., pro-Black), negative scores reflect an implicit preference for White versus Black (i.e., pro-White), whereas a score of zero indicates equivalent implicit preference for Black and White.
- Compute a semantic differential score by averaging explicit ratings for the five dimensions for each concept, using a scale from -3 (negative) to 3 (positive).
Asking someone to say what?s on their mind can be disconnected from the beliefs that they are willing to discuss or are even knowingly aware of.
While traditional methods often ask individuals to report on their own attitudes?say to evaluate personal feelings about members of a stigmatized group?their opinion is explicit, involving deliberate thought.
Because the topic is sensitive, people are more likely to state unprejudiced, egalitarian views and portray themselves positively even though they may truly harbor negativity.
To circumvent this social-desirability bias, implicit attitudes?evaluations that occur outside of conscious awareness and control?must be examined.
This video demonstrates how to conduct the Implicit Association Test?an influential measure for investigating the strength of associations between a concept like race and automatic evaluations based on the original work of Greenwald and colleagues.
In this experiment, European American participants are recruited to establish out-group homogeneity. They are shown pictures and words?associated with either a particular race or attribute?and asked to quickly and accurately sort them across five different blocks.
The catch is that the stimuli appear in rapid succession?without time for participants to explicitly process them?hence, the name Implicit Association Test.
In the first block of trials, concept discrimination, the initial targets?faces of White European or Black African origin?are randomly presented, without replacement.
If the face shown is White, participants should press the key corresponding to "White". The first half are considered practice trials, and not saved, as errors are expected while participants get used to responding quickly.
Likewise, for Block 2?attribute discrimination?participants are exposed to only "good" and "bad" words. That is, if "horrible" appears, the correct reply would be a keypress corresponding to "bad". Thus, the first two blocks serve as baseline latencies associated with correct responses.
During Block 3?the combined portion?either images or words are shown with a race and attribute paired to one response key. Participants must now decide whether the presented face or word like "fabulous" corresponds with "Black or Good" or "White or Bad".
Block 4?the reversed concept discrimination?is a repeat of Block 1 except that the computer keys for "Black" and "White" are flipped. With this block, participants adjust to the keypresses for the new "Black/White" correspondences.
Finally, Block 5?reversed combined?is similar to Block 3 except that Race is flipped across attributes, such that "Black or Bad" and "White or Good" are now paired on the same response key. Implicit preferences should be observed in latency differences compared to Block 3.
The dependent variable then is the latency to react across block types. Participants are expected to sort faster when good words and White faces go with the same key compared to the opposite, good and Black. Thus, reaction times reveal the strength of each participant?s implicit preference, which agrees with a stereotypical bias.
In addition, the responses can be log-transformed and compared to self-reported attitudes provided on a race-related questionnaire given after the task has been completed.
In this case, if participants? biases are indeed unspoken, it?s hypothesized that the self-reported beliefs will not correlate to the scores identified during the Implicit Association Test, thus revealing a form of social-desirability.
Before starting the experiment, conduct a power analysis to determine the appropriate number of participants?specifically of European-American descent?that are required.
To begin, greet each one in the lab, explain they will be sorting images and words over several blocks of trials, and have them sign a consent form to participate.
Seat the participant in front of a computer. Further explain that they are to classify, as quickly and accurately as possible, the word or image that appears on the screen by pressing "E" if it fits the category on the left or "I" for the right side. Answer any questions and leave the room.
Start Block 1 by pressing the spacebar. During this initial phase, notice that participants are merely responding to classify faces based on the race anchors Black and White over the course of 100 trials.
Proceeding to Block 2, associated attribute discrimination, observe that now only the list of words and the valence anchors good and bad serve as the classification choices for another 100 trials.
Upon progressing into Block 3, the initial combined task, either images or words appear?but are now paired onto one response key?for a total of 200 trials.
Block 4, reversed target-concept discrimination, is related to Block 1, except that the race anchors now appear on the opposite side for 100 trials.
Finally in Block 5, the reversed combined segment, notice that participants again classify faces and words with combined anchors but the attributes are flipped compared to Block 3.
Following the Implicit Association Test, explain that there are some additional questionnaires to complete on the computer. Emphasize that they will have complete privacy, and then leave the room.
Allow participants enough time to complete the surveys. To conclude, return for debriefing and thank them for taking part in the study.
To visualize the data, plot the mean trial latencies across block type. See the manuscript for details on recoding reaction times.
Notice that responses were slower for the Black/good trials as compared to the White/good ones. Slower responses reflect more difficult associations, suggesting European-American participants found it challenging to associate Black faces with pleasant nouns. In other words, they exhibited an implicit attitudinal preference for the White anchor over the Black anchor.
Also for each participant, compute an index of the Implicit Association Effect by first log transforming the reaction times and then subtracting the means in Block 3 from Block 5. A positive score reflects an automatic preference for Black, whereas a negative one reveals an inclination for White.
Compare these indexes with the semantic differential score calculated by averaging the explicit ratings on the final questionnaire. Here, a value of zero indicates a self-reported egalitarian preference?no race bias.
Results show that most participants did self-report egalitarian preferences despite their IAT scores revealing a moderate to strong implicit preference for White over Black. These results suggest that a social-desirability bias may have distorted their responses to the questionnaire.
Now that you are familiar with how the Implicit Association Test can examine automatic bias and prejudice for socially sensitive topics, let?s look at other real-life situations where the task can be applied.
Researchers have published the Implicit Association Test on the internet allowing anyone to participate, on a number of topics, including social attitudes and mental health. Participants receive immediate feedback on their implicit preferences, providing a straight-forward approach for examining their own beliefs and biases in the comfort of their own home or office.
Other researchers have used the task to measure self-esteem by pairing self/other with pleasant/unpleasant words. These tests can see through any fa?ade or deception leading to better success with interventions and treatment.
Lastly, children are notoriously difficult to get reliable information from, especially when asked to self-report measures. For this reason, researchers use Implicit Association Tests to assess young attitudes and beliefs, ranging from versions that investigate preferences towards race, gender, and even healthy-eating behaviors.
You?ve just watched JoVE?s video on the Implicit Association Test. Now you should have a good understanding of how to design and execute an experiment using Implicit Association tests, how to analyze and assess the results, as well as how to apply the principles to a number of real-world situations.
Thanks for watching!
Q1: Why is the Implicit Association Test better than asking people directly about their attitudes?
Direct questioning relies on explicit attitudes requiring deliberate thought, but people often report unprejudiced views due to social-desirability bias—wanting to appear egalitarian even if harboring negative associations. The IAT bypasses this bias by measuring implicit attitudes that occur outside conscious awareness and control, revealing automatic evaluations people may not knowingly acknowledge or be willing to discuss.
Q2: How does the Implicit Association Test measure automatic preferences?
The IAT presents stimuli in rapid succession without time for explicit processing, forcing participants to rely on automatic associations. By comparing reaction times across blocks—such as when good words pair with White faces versus Black faces—the test reveals the strength of implicit preferences. Faster responses indicate easier associations, while slower responses suggest more difficult or conflicting associations, similar to evaluating the accuracy of snap judgments.
Q3: What do the five blocks in an Implicit Association Test measure?
Block 1 establishes baseline responses to race concepts (White/Black faces). Block 2 measures responses to attribute words (good/bad). Block 3 pairs race with attributes in one configuration. Block 4 reverses the race key positions to adjust participants to new mappings. Block 5 reverses the attribute pairings, allowing comparison of implicit preferences between the two configurations.
Q4: What does a positive Implicit Association Effect score indicate?
The Implicit Association Effect is calculated by log-transforming reaction times and subtracting Block 3 means from Block 5 means. A positive score reflects an automatic preference for the concept paired with the reversed configuration, while a negative score reveals preference for the original pairing. In race-based IATs, negative scores typically indicate implicit preference for White over Black.
Q5: How can the Implicit Association Test reveal social-desirability bias?
Researchers compare IAT scores with self-reported attitudes from questionnaires completed after the task. When participants show implicit preferences on the IAT but report egalitarian beliefs on surveys, the discrepancy suggests social-desirability bias—they consciously endorse unprejudiced views while harboring automatic negative associations they may not consciously acknowledge.
Q6: What are real-world applications of the Implicit Association Test beyond measuring racial bias?
The IAT has been adapted to measure self-esteem by pairing self/other with pleasant/unpleasant words, assess mental health attitudes, and evaluate children's implicit preferences toward race, gender, and healthy-eating behaviors. Online versions allow public participation with immediate feedback, providing accessible tools for examining automatic beliefs and biases in various contexts where self-report measures may be unreliable or subject to distortion.
Q7: Why is the Implicit Association Test particularly useful for studying children's attitudes?
Children are notoriously difficult to assess through self-report measures, as they may not accurately report their true beliefs or may lack the metacognitive awareness to do so. The IAT bypasses these limitations by measuring automatic associations without requiring conscious introspection, making it an effective tool for reliably assessing young people's implicit attitudes toward various social categories and behaviors.