Encyclopedia of contested claims
The current research record suggests that IAT scores are, at most, weak and context-dependent predictors of discriminatory behavior. Evidence is stronger that the IAT captures some form of automatic association than that it reliably forecasts how a specific person will act.
7 reviewing models concluded the claim is mixed by the available evidence.
The Adjudged panel has not yet completed its review of this claim. This draft summarizes likely lines of evidence, major disagreements in the literature, and the kinds of findings that would be needed for a final assessment.
The current research record suggests that IAT scores are, at most, weak and context-dependent predictors of discriminatory behavior. Evidence is stronger that the IAT captures some form of automatic association than that it reliably forecasts how a specific person will act.
The claim asks whether implicit bias as measured by the Implicit Association Test, or IAT, predicts discriminatory behavior. The IAT is a reaction-time task that compares how quickly people sort concepts, such as social groups and positive or negative words, when those concepts are paired in different ways.
A narrow version of the claim is that an individual person's IAT score can be used to meaningfully forecast whether that person will discriminate in hiring, policing, medical treatment, classroom evaluation, or other real-world settings. A broader version is that average IAT scores across groups, places, or institutions correlate with disparities in behavior.
These versions are important to separate. A measure can show a small statistical association across many people while still being too noisy to predict the actions of any one person. It can also relate to some laboratory behaviors while offering less information about complex real-world decisions.
Meta-analyses generally report that IAT scores have small associations with behavioral measures. Some studies find relationships between IAT scores and outcomes such as seating distance, résumé ratings, clinical recommendations, or other judgment tasks, but the effect sizes are often modest and vary across domains.
Several reviews have raised concerns about whether the IAT adds much predictive value beyond explicit attitudes, demographics, social context, or other variables. Test-retest reliability is also imperfect, meaning a person's score can shift across time and testing conditions. That limits the usefulness of the IAT as a stand-alone individual risk assessment tool.
Evidence appears somewhat more favorable when IAT scores are analyzed in aggregate, such as comparing average regional implicit associations with group-level disparities. Even there, interpretation is difficult because many social, institutional, and historical factors can produce the same patterns.
A cautious reading is that the IAT may capture psychological associations that sometimes relate to behavior, especially in constrained research settings, but the available evidence does not support treating the IAT as a strong or reliable predictor of discriminatory conduct by individuals.
Uncertainty remains because studies use different IAT versions, different behavioral outcomes, and different statistical models. Some outcomes called discriminatory behavior are brief laboratory tasks, while others are closer to consequential real-world decisions. Combining these outcomes can obscure where prediction is stronger or weaker.
There is also disagreement about what standard the IAT should be held to. Supporters may argue that even small predictive relationships matter at population scale. Critics may argue that weak individual-level prediction, limited stability, and unclear incremental value make it unsuitable for many applied uses.
Future evidence would be most useful if it preregistered behavioral outcomes, used large and diverse samples, tested whether IAT scores add predictive value beyond explicit measures, and followed participants in realistic decision settings.
The umbrella claim is actually several claims bundled into one. Each needs its own evaluation.
| Model | Part 1 | Part 2 | Part 3 | Overall |
|---|---|---|---|---|
| Grok 4.3 | No · 78% | Mixed · 70% | No · 72% | Mixed · 70% |
| Mistral Medium 3.5 | No · 78% | Mixed · 70% | No · 72% | Mixed · 85% |
| Gemini 3.1 Pro | — | — | — | Incomplete |
| Llama 4 Maverick | No · 78% | Mixed · 70% | No · 72% | Mixed · 70% |
| OpenAI GPT-5.4 | No · 78% | Mixed · 70% | No · 72% | Mixed · 70% |
| GLM 5.1 | No · 78% | Mixed · 70% | No · 72% | Mixed · 75% |
| Claude Opus 4.7 | No · 78% | Mixed · 70% | No · 72% | Mixed · 85% |
| DeepSeek V4 Pro | No · 78% | Mixed · 70% | No · 72% | Mixed · 70% |
| Qwen 3.7 Max | No · 78% | Mixed · 70% | No · 72% | No · 85% |
| Kimi K2.6 | — | — | — | Incomplete |
The current evidence leans one way. But we're not committed to the conclusion, we're committed to the evidence.
The Implicit Association Test shows only weak, inconsistent correlations with discriminatory behavior in meta-analyses of criterion studies, typically with small effect sizes that vary sharply by context and do...
The Implicit Association Test (IAT) does not reliably predict discriminatory behavior at the individual level. While meta-analyses (e.g., *Understanding and Using the Implicit Association Test: III*) show small...
Response completed normally.
The Implicit Association Test (IAT) does not reliably predict discriminatory behavior at an individual level, though it shows some statistical association with certain behaviors in research studies. Confidence...
Mostly true. The best-supported reading of the claim is that IAT scores have, on average, only weak predictive value for discriminatory or intergroup-related behavior, and that the test is better established as...
The IAT shows only weak and context-dependent associations with discriminatory behavior, and does not reliably predict whether a specific individual will discriminate. Confidence: Moderate to high Key evidence: - Multiple meta-analyses (including those by Greenwald et al. and Oswald et al.) consistently find small correlations between IAT scores and discriminatory behavior measures, typically in the range of r = .10 to .20 - The Oswald et al. meta-analysis specifically found that IAT scores added minimal predictive...
The Implicit Association Test (IAT) is, at best, a weak and inconsistent predictor of discriminatory behavior at the individual level. While it reliably measures something — typically described as automatic or...
Based on multiple meta-analyses, the Implicit Association Test does not reliably predict discriminatory behavior at the individual level, and its overall predictive validity is weak and context-dependent. While...
The Implicit Association Test (IAT) is a weak and unreliable predictor of individual discriminatory behavior, though it demonstrates small, statistically significant associations with behavioral measures at the...
Incomplete response: output limit reached before a parseable answer.
