Review Analyzes the Weaknesses of the Implicit Association Test on Predicting Behavior

Published: March 29, 2021 | Last Updated: April 26, 2024by Noah Shaw

Summary of:

Meissner, F., Grigutsch, L.A., Koranyi, N., Müller, F., & Rothermund, K. (2019). Predicting behavior with implicit measures: Disillusioning findings, reasonable explanations, and sophisticated solutions. Frontiers in Psychology, 10. https://www.frontiersin.org/articles/10.3389/fpsyg.2019.02483/full

Background & Theory:

The Implicit Association Test (IAT) is commonly utilized by educators and researchers worldwide to make salient the presence of subconscious implicit associations. Over the past 20 years, researchers have been excited by the idea that the IAT provides an outlet to bridge the gap between implicit attitudes and behavior. However, research has shown that the IAT’s ability to predict behavior is quite weak. This review presents explanations for the weak predictive validity of the IAT and similar measures, along with suggesting promising avenues researchers can take moving forward.

Question(s):

Research was analyzed by Franziska Meissner et al. to answer the following questions:

    1. Why does the Implicit Association Test have weak behavior predictability?
    2. What can researchers do differently to improve the predictive validity of the IAT and its counterparts?

Methods:

The authors analyzed a total of 150 research studies and other resources to create this review. This resulted in four problematic features of the IAT that lead to weakened predictive validity as well as potential solutions to these four features.

Results:

The four potentially problematic features of utilizing the IAT as a predictor of behavior that the authors discovered, include the following:

1. Extraneous Influences on Implicit Measures:

Past studies have confirmed that implicit measures like the IAT which measure attitudes are not void of additional, non-attitudinal influences. The fact that extraneous factors play a role in these implicit measures bring down predictive validity of behavior. However, the systematic error variance across measures has a common core: recoding.

Recoding is a process by which people assimilate attributes into superordinate categories. Through recoding, people are able to complete the compatible tasks of the IAT by associating images to words with a feature other than identity, such as familiarity, salience, or valence. Recoding cannot happen with the IAT’s incompatible tasks because they involve double categorization. This gap between assessments that allow recoding and assessments that do not, is an irrelevant influence on the outcome of the test that can lead to weakened predictive validity of behavior. Therefore, IAT scores must be understood as a mix of both relevant influences like associations and irrelevant influences like recoding.

Suggested Solutions: The authors suggested two possible solutions to the issue of extraneous influences on implicit measures. First, the results of studies on recoding suggest that it is perhaps the biggest threat to the IAT’s validity. Therefore, a simple solution that would prevent recoding is dropping the block structure of the IAT in favor of other versions that operate out of one test block, such as the Single-Block IAT and the Recoding-Free IAT. A second solution is to adopt a modeling approach via the ReAL Model, which effectively untwines the effects of evaluative associations from the influence of task recoding.

2. Distinguishing Between Liking and Wanting:

Neuropsychological research over the past few decades has formulated the incentive salience hypothesis, which suggests that liking something and wanting it are separate by nature. Each is mediated by different processes in the brain and is affected by different factors. While liking and wanting are often highly correlated, they can diverge from one another, with ‘wanting’ often more strongly guiding behavior. An example of this divergence can be seen in substance addiction: People suffering from addiction can want the substance they are addicted to without liking it, even when recognizing the substance’s harmful effects. This being said, when trying to predict behavior, as with the IAT, researchers should incorporate measures of ‘wanting’ due to its ability to guide behavior.

Suggested Solutions: The authors suggest utilizing the element of ‘wanting’ as opposed to ‘liking’ in trying to close the gap between attitude-behavior. One such method of doing this is through the Wanting-Implicit Association Test.

3. Focus on Associations Versus Beliefs:

Associations, like those made by participants in the IAT, do not hold qualitative relational information and therefore can have ambiguous meaning. For example, if a person associates the words “I” and “good” together, the reason why this association occurs could be answered with a variety of explanations. Maybe this person believes they are good—or maybe they believe they are not good—or maybe they desperately would like to be good. The reason behind why they associate “I” and “good” to each other is unclear. Therefore, the ambiguity surrounding associations may influence the weak predictive validity of the IAT.

Suggested Solutions: The authors suggest that measuring implicit propositional beliefs may be an effective way of bypassing the limitations of associations. They propose utilizing the Implicit Relational Assessment Procedure, the Relational Responding Task, and/or the Propositional Evaluation Paradigm. These implicit measures of beliefs are promising due to their ability to measure complex, relevant relationships, which implicit measures of association cannot do.

4. Lack of Fit Between Predictor and Criterion:

This problematic feature of utilizing the IAT as a predictor of behavior has less to do with the IAT itself and more to do with the measurement of the respective criterion. If researchers do not properly assess the criterion of a particular study, then results showing weak predictive validity of the IAT may not be the full responsibility of the IAT. The authors argue that “the predictive validity of implicit measures suffers from the fact that (1) studies often do not assess behavior proper but rather employ self-report measures as a criterion, and (2) implicit measures typically do not provide contextual information; details that are crucial for real-life behavior” (Meissner et al., 2019).

First, studies have shown that the accuracy of self-reported behavioral intentions is not high when attempting to predict actual behavior. This is especially true when testing the relationship between implicit measures like the IAT and behavioral outcomes. Second, behavior itself is context specific. This being said, implicit measures that do not account for context like the IAT often operate in a vacuum which is insubstantial for predicting real-life behavioral outcomes.

Suggested Solutions: The authors suggest researchers replace self-report measures with specific hypotheses on the type of behavior to be predicted and/or clarifying under which specific conditions a relationship should occur. Additionally, they suggest introducing context to implicit measures through context-specificity. By introducing context-specificity on the level of implicit measures of attitudes, researchers will likely be able to better predict and explain behavioral outcomes.

What We Can Learn:

Looking over this research, we can take away the following key insight:

  • The authors determined four major reasons why the Implicit Association Test has weak predictive validity in regards to behavior, including the presence of extraneous influences on implicit measures, the lack of distinguishing between ‘liking’ and ‘wanting,’ the focus on associations rather than belief, and inadequate measurement of the criterion in regards to use of self-reporting and lack of context.

Final Takeaways

For Consultants: The Implicit Association Test can be useful for personal recognition of implicit associations and biases but should be taken with a grain of salt. There are other influencing factors that can impact results, so it may not be the most reliable measure for predicting behavior without any of the provided solutions above.

For Everyone: Measures like the IAT can be helpful for individually recognizing the presence of subconscious implicit associations—consider trying it or other similar assessments out!

Avatar for Noah Shaw

Noah Shaw

Noah Shaw is a double alumnus from Pepperdine University, holding a Masters in Dispute Resolution (MDR) from the Caruso School of Law’s Straus Institute for Dispute Resolution and a Bachelor of Arts in Integrated Marketing Communication from Seaver College. He additionally received a Certificate in Conflict Management from the Straus Institute in 2019. In his role as a Research Writer with Pollack Peacebuilding Systems, Noah examines the latest workplace conflict resolution research and applying it to both content distribution and PPS’ best practices. Learn more about Noah here!