Share on

The Implicit Association Test

Jessica Massonie, Daniel C. Richardson, Hannah Spence, Jo Evershed

Daniel C. Richardson is a Professor of psychology at UCL who has been developing new ways to bring active research into undergraduate teaching. As part of the course, he has students collect data for an online IAT using their social media channels. The students typically get in the region of 1500 respondents! Subsequently, he has groups of 5 students create their own IAT to investigate their own hypotheses. You can read all about this remarkable way of teaching research methods here.

Below is a write-up of the results of one of these IATs. We present them here to show that Gorilla is a useful tool for doing behavioural research on the web.

The Implicit Association Test (IAT) is a widely used measure of attitudes, stereotypes, self-esteem and self-concept (Nosek, Hawkins & Frazier, 2011; Smith & Nosek, 2010). One version of the IAT has been implemented in Gorilla, focusing on racial attitudes.

The “Implicit” part of the IAT relies on the fact that participants are not asked to explicitly report their preferences (e.g: “Do you prefer Black people or White people?”). Instead, a specific methodology is used to measure, through reaction times, the association between each category (“White” or “Black” people), and positive or negative attributes.

Completing an IAT involves different steps:

First, they are asked to categorise faces according to the two groups of interest (“Black or “White” faces), by pressing two different buttons (e.g: “Q” for the left-hand side category, “P” for the right-hand side category).

Secondly, they categorise adjectives as “Good” or “Bad”.

Following these two separate “practice blocks” involving faces and adjectives, a mixed block is presented. One type of face, and one type of adjective are presented on each side. In the example below, participants have to press the same button for “Black” and “Bad”, and “White” and “Good”. The IAT assumes it is easier to make the same response a percieved related pair. Imagine a researcher is testing a White sample, and wants to assess whether subjects have negative implicit attitudes toward Black people, compared to positive attitudes toward White people. The example depicted here would correspond to a “Congruent Trial”: Black faces and negative attributes are on the same side and should be answered with the same button; White faces and positive attributes are associated in the same way.

Following this mixed block, the position of the “Good” and “Bad” adjectives are inverted. Several practice trials are presented so that participants get familiar with this new configuration.

Finally, a second set of mixed trials are displayed. This time, if we follow the previous hypothesis, “incongruents” trials are: “White” faces with “Bad” adjectives and “Black” faces with “Good” adjectives.

Implicit associations are measured by calculating the difference between the mean reaction time for Congruent and Incongruent trials. If “White faces” are implicitly associated with “Good” attributes, and/or “Black faces” to negative attributes, a participant will answer faster in the Congruent trials compared to Incongruent trials.

The IAT is one of the most used and cited implicit measure in social cognition research (Nosek, Hawkins & Frazier, 2011). Two influential studies, Greenwald, McGhee & Schwartz (1998) have reported an IAT effect of about 179 ms among White undergraduate American students. That is to say, participants answered, on average, 179 ms faster when response buttons were similar for typical White names and pleasant attributes, compared to typical Black faces and pleasant attributes. A broader web-based research study replicated these findings, using both names and faces stimuli and reporting an IAT effect of 158 and 187 ms, respectively (d = .71 and d = .88; Nosek, Banaji & Greenwald, 2002).

In partnership with researchers, Gorilla’s team aimed at replicating this effect on its platform and assess the validity of its measurement tools.

The IAT on Gorilla

Participants and Methods

143 students performed a version of the IAT hosted by Gorilla. Images of 6 White faces and of 6 Black faces have been used as stimuli, as well as 8 positive words (Joy, Love, Peace…) and 8 negative words (Agony, Terrible, Nasty…). The task is available here. 75 participants performed the Congruent block first, 68 started with the Incongruent block.

Results

Overall, students were 72.77 ms faster to respond to Congruent trials (M = 843.07 ms), compared to Incongruent ones (M = 915.83 ms; F(1, 141) = 19.51, p < .001; d = .36). Table 1 reports the descriptive statistics. The mean reaction times for Congruent and Incongruent trials are plotted in Figure 1.

Table 1. Mean reaction times and standard deviation for the congruent and incongruent trials.

	Total		Congruent First		Incongruent First
	Mean	SD	Mean	SD	Mean	SD
Congruent Trials	843.07	17.35	873.98	242.46	812.16	159.63
Incongruent Trials	915.83	17.33	911.25	177.62	920.42	235.21

Figure 1. Density plots representing the distribution of reaction time for congruent and incongruent trials

Moreover, as shown in Figure 2, the IAT effect interacted with the order of the task, where the difference between Congruent and Incongruent trials was more salient when participants started with the Incongruent condition (F(1, 141) = 4.64, p < .05).

Figure 2. Mean reaction time for the Congruent and Incongruent trials as a function of task order.

Conclusions

The main IAT effect was replicated, although reaction time differences and effect size were a bit smaller than in Greenwald, McGhee & Schwartz, 1998; Nosek, Banaji & Greenwald, 2002.

Between subject variability in equipment is not a concern for the IAT because it is a within subject design (Woods, Velasco, Levita, Wan & Spence, 2015). Congruent and incongruent conditions are presented to the same participants on the same devices.

Differences between the results in our sample and results from the studies we referred to can be due to several factors, such as the ethnicity of our participants (a mix of White and non-White students can decrease the overall IAT effect), their familiarity with our categories of interest (Black and White people), or these could be attributed to a difference in trial numbers.

Importantly, it must be understood that a person having been shown to associate “Black people” with “Bad attributes” at an IAT test might not show any sign of racist behaviour. According to dual process theories, implicit attitudes reflect automatic, spontaneous and/or impulsive processes, that might be overcame to take more deliberated and controlled decisions. However, the overlap between implicitness, automaticity, intentionality, awareness and control is complex and debated (De Houwer & Moors, 2007).

If you'd like to run an IAT on the web then sign up to gorilla here and look at these samples.

Alternatively, if you'd like to know more about using Gorilla as a teaching tool see this case study.

Want to learn more about Gorilla?

References

De Houwer, J., & Moors, A. (2007). How to define and examine the implicitness of implicit measures. Implicit measures of attitudes: Procedures and controversies, 179-194.

Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. (1998). Measuring individual differences in implicit cognition: the implicit association test. Journal of personality and social psychology, 74(6), 1464-1480.

Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvesting implicit group attitudes and beliefs from a demonstration web site. Group Dynamics: Theory, Research, and Practice, 6(1), 101-115.

Nosek, B. A., Hawkins, C. B., & Frazier, R. S. (2011). Implicit social cognition: From measures to mechanisms. Trends in cognitive sciences, 15(4), 152-159.

Woods, A. T., Velasco, C., Levitan, C. A., Wan, X., & Spence, C. (2015). Conducting perception research over the internet: a tutorial review. PeerJ, 3, e1058.

Smith, C. T., & Nosek, B. A. (2010). Implicit Association Test. In I. B. Weiner & W. E. Craighead (Eds.), Corsini's Encyclopedia of Psychology, 4th Edition (pp. 803-804). Wiley.