A study published in Nature Human Behaviour by Kyle J. LaFollette, Doroteja Rubez, Heath A. Demaree, and Amit Goldenberg ran racing diffusion models on 115,601 IAT sessions across 39 topics and found that IAT scores mostly reflect conscious response caution — a strategic decision to slow down and prioritize accuracy over speed — not automatic mental association at all. That’s a direct hit against the assumption the Implicit Association Test was built on: that a hidden bias lives beneath your conscious opinions, a truer, automatic self that a few thousand sorting trials can catch before it has time to lie to you. Jean-Paul Sartre spent Being and Nothingness (1943) arguing there was never a hidden, fixed self dictating your behavior from below consciousness in the first place. He wasn’t being contrarian. He was describing exactly the kind of thing this study just measured.

The Quick Verdict

Question	What we found
What was studied?	Whether IAT scores (D-scores) reflect automatic associative memory, as long assumed, or something more deliberate
Who ran it?	Kyle J. LaFollette, Doroteja Rubez, Heath A. Demaree, and Amit Goldenberg — Case Western Reserve University and Harvard University
Where and when?	Nature Human Behaviour, covered by PsyPost on September 15, 2026 and Chicago Booth Review
Sample size	115,601 IAT sessions across 39 topics (race, age, sexuality, politics, economic systems, and more)
The method	Racing diffusion models, which split reaction time into decision ease, response caution, and non-decision time
The twist	Response caution — a conscious tradeoff of speed for accuracy — explained significantly more variance in D-scores than decision ease, the automatic component the test was designed to isolate
What mattered most	Response caution also best predicted people’s own explicitly reported biases — better than the “automatic” component did
What Sartre already argued	In Being and Nothingness (1943), there’s no fixed unconscious self dictating behavior beneath conscious control. Consciousness is radically free, and even reactions that look automatic are choices you’re responsible for

The Quick Version

A 2026 study of 115,601 Implicit Association Test sessions across 39 topics found that scores are driven mainly by response caution — a conscious decision to slow down and be careful — not by the automatic, unconscious associations the test claims to reveal. Sartre argued in 1943 that the whole idea of a fixed hidden self pulling the strings beneath your awareness doesn’t hold up. What looks automatic is still something you’re doing, and doing means you’re responsible for it.

What the Study Actually Found

The Implicit Association Test asks you to sort words and images into categories as fast as you can — pleasant words with one group, unpleasant words with another, over and over, with the pairings swapped partway through. The idea, going back to the test’s origins in the 1990s, is that you’ll be faster when the pairing matches an association already wired into your head, and slower when it doesn’t. That speed gap becomes your D-score. Faster on the “compatible” block than the “incompatible” one, and the test infers a hidden preference — one you might not endorse, or even know about.

LaFollette’s team didn’t run a new experiment so much as re-examine the reaction-time data that’s been piling up for decades, using a tool built to pull that data apart more carefully. A racing diffusion model doesn’t just clock how long a response took. It decomposes the process into separate pieces: how much evidence you needed before committing to an answer (response caution), how easily that evidence pointed toward one option over the other (decision ease, the piece closest to what “automatic association” is supposed to mean), and time that’s just mechanical — reading the word, moving your hand.

Run across 115,601 sessions and 39 different topics, the pattern held with unusual consistency: response caution explained more of the variance in D-scores than decision ease did. People weren’t mostly struggling to sort words because of a conflict buried in memory. They were mostly choosing, session by session and block by block, how much accuracy to demand of themselves before answering — and that choice, not the automatic pull of association, is what the score was largely tracking.

The detail that should worry anyone who treats the IAT as a diagnostic instrument: response caution also did a better job predicting what people said their own attitudes were — their explicit, self-reported biases — than decision ease did. The piece of the score that’s supposed to be irrelevant noise turned out to be the piece most connected to what people actually, consciously believe about themselves. As LaFollette put it in PsyPost’s coverage, “IAT scores aren’t as simple as they’re often made out to be. In short, an IAT score reflects a blend of factors, and part of that blend is more deliberate than the ‘implicit’ label suggests.”

Two people can post the identical D-score for entirely different reasons. One might be experiencing something closer to the genuine associative conflict the test was designed to catch. The other might just be a careful test-taker — someone who, on this topic, decided accuracy mattered enough to slow down for. Same number. Different story. The test, as currently scored, can’t tell you which one you’re looking at.

Sartre: There Was No Hidden Bias to Begin With

Jean-Paul Sartre wrote Being and Nothingness in 1943, and a good chunk of its early sections are a direct argument against Freud’s model of the mind — the Conscious, the Preconscious, the Unconscious, or later, the Id, Ego, and Superego. Sartre’s objection wasn’t squeamishness about the unconscious as a concept. It was structural. Freud’s model, he argued, tries to explain self-deception — how a person can lie to themselves — by splitting the mind into a part that knows the truth and hides it (the unconscious) and a part that gets lied to (the conscious ego). But that just relocates the problem. Something in you still has to know the truth well enough to hide it from the rest of you, and that something is doing exactly the deceiving that the unconscious was supposed to explain away. Sartre thought this was less an explanation than a shell game — pushing the mystery one level down and calling it solved.

His alternative, laid out through the concept of bad faith (mauvaise foi), was that consciousness isn’t split into a knowing part and a deceived part at all. It’s one thing, radically free, and what looks like an automatic reaction beneath your control is still, structurally, something you are doing. Not a fixed trait being revealed. An ongoing choice, renewed moment to moment, that you can fail to notice you’re making — but a choice all the same. This site has covered bad faith before: the waiter who performs “waiter” so completely he treats it as his essential nature, rather than a role he’s continuously choosing to inhabit.

The IAT was built on something close to the Freudian model Sartre spent his career arguing against — a fixed unconscious attitude, sitting beneath awareness, that the test’s timing mechanism catches red-handed before the conscious mind can intervene. LaFollette’s study didn’t set out to test Sartre. It set out to test a psychometric assumption using diffusion modeling. But what it found — that the score everyone read as “your hidden self, unfiltered” turns out to be driven mainly by a conscious strategic choice about how carefully to respond — is close to a live demonstration of Sartre’s argument. There wasn’t a hidden thing being revealed underneath the response time. There was a person deciding, in real time, how much caution this particular sorting task deserved.

What Did Sartre Mean by Rejecting the Freudian Unconscious?

1. Splitting the mind doesn’t explain self-deception — it just relocates it. If an unconscious “knows” something and hides it from the conscious mind, you still need to explain how a part of you can know and not-know simultaneously. Freud’s model names the puzzle. It doesn’t solve it.
2. Consciousness is unified, not layered. For Sartre, there’s no separate compartment running the show from below. There’s one consciousness, capable of directing attention away from things it would rather not confront — which looks like hiding something from itself, without needing a second, hidden mind to do the hiding.
3. What looks automatic is still chosen. A habit, a reflex, a “gut reaction” — these aren’t proof of a fixed inner essence. They’re patterns you’ve built and keep rebuilding through repeated choices, even when the choosing has become fast enough to feel involuntary.
4. Because it’s chosen, you’re responsible for it. This is the part that makes bad faith an ethical concept and not just a psychological one. If your reaction were the output of a hidden mechanism, you’d be off the hook for it. Sartre’s whole point is that you’re not.

Why the Study and the Philosophy Rhyme

The connection here isn’t that Sartre predicted diffusion modeling — he didn’t, and it would be a stretch to pretend a 1943 phenomenology book anticipated a specific psychometric finding eighty-three years later. The connection is narrower and, I think, more interesting: both the study and the philosophy are pushing back on the same assumption from different directions. The assumption is that there’s a truer, more automatic version of you sitting underneath your conscious responses, and that if you can just get fast enough or clever enough in your measurement, you’ll catch it before it composes itself into something presentable.

LaFollette’s team found that the measurement everyone trusted to catch that hidden self mostly caught something else: a conscious decision about how much evidence to gather before answering. Sartre argued that the hidden self was never going to be there to catch, because consciousness doesn’t work by hiding a truer self in a locked room and sending out a manicured spokesperson. It works by choosing, continuously, including choosing not to look closely at what it’s choosing. That’s not the same claim as “nobody has biases” or “stereotypes don’t shape behavior” — the site’s own coverage of implicit bias research elsewhere doesn’t make that argument and neither does this study. It’s a narrower, sharper claim: the specific instrument built to catch bias operating below conscious control turns out to be substantially measuring something operating quite close to conscious control. Caution is a strategy. Strategies are chosen.

The piece on akrasia — acting against your own better judgment — draws a related distinction worth bringing in here. Akrasia is knowing what you should do and failing to do it. Bad faith is refusing to acknowledge that you’re doing anything at all, recasting a choice as a fixed fact about who you are. The old reading of IAT scores leaned toward treating bias as something closer to a diagnosis than a choice — a number revealing what you secretly are, independent of anything you’re doing on purpose. That framing has its own kind of bad faith built into it, just distributed at the level of an entire testing paradigm rather than one person’s self-story: it lets both the test-taker and the institutions using the test treat the number as a revealed fact about an unconscious self, rather than as a measurement partly constructed by a conscious act the test wasn’t designed to isolate.

The post on Rawls and self-respect covered a different 2026 study where a psychological state — depression — changed what people could consciously perceive as fair, without their awareness that perception had shifted. That’s a genuinely different mechanism from what’s happening here. But both studies share a caution worth stating plainly: measuring a number and confidently narrating what’s happening beneath it are two different acts, and it’s easy to let the second one run ahead of what the first one actually supports.

Where This Doesn’t Hold Up

This study doesn’t show implicit bias is fake, that stereotypes don’t shape behavior, or that the IAT measures nothing real. It shows that the specific mechanism long assumed to explain D-scores — automatic associative memory — explains less of the variance than a conscious process the test wasn’t designed to isolate. Those are different claims, and collapsing them would be its own kind of overreach in the other direction.

It’s also worth being precise about what “response caution” is and isn’t. It’s not the same as consciously endorsing a bias, or deliberately manipulating your score. It’s a general strategic setting — how much accuracy you demand of yourself before answering — that can shift based on the topic, your mood, how tired you are, or how much the topic feels like it matters to get right. The study found this factor predicts explicit attitudes better than decision ease does. It didn’t find that people are secretly aware of manufacturing their IAT results in real time.

And Sartre’s argument, useful as the resonance is here, was not built as a psychometrics critique. Being and Nothingness is a work of phenomenological ontology, concerned with the structure of consciousness and freedom generally, not with the internal validity of a specific reaction-time test invented five decades after his death. The honest version of this piece says the two are describing something structurally similar — a rejection of the idea that a fixed, hidden mental layer explains what looks automatic — without claiming Sartre foresaw diffusion modeling of IAT data. He didn’t need to. He was arguing against the shape of that assumption wherever it showed up, in whatever century it showed up in.

If you’re working through your own biases — implicit, explicit, or somewhere in between — a single test score, old model or new, was never going to settle the question on its own. That work is closer to what Sartre would call an ongoing project than a diagnosis you receive once and carry around as fact.

A Practice Worth Trying

Neither LaFollette’s team nor Sartre wrote a self-help program, but the shared shape of what they’re pointing at suggests something specific to try.

1. Notice when you explain a reaction with “that’s just automatic.” The study suggests plenty of what feels automatic is closer to a conscious setting — how carefully you’re choosing to respond — than a hidden force acting on you. Ask what you’d find if you slowed down and looked at the choice itself.
2. Treat a bias test result as a data point, not a verdict. A D-score reflecting response caution as much as association means it’s telling you something real about how you approached that specific task, on that specific day — not handing you a fixed reading of your unconscious mind.
3. Ask what you’re choosing when a reaction feels fast. Sartre’s diagnostic question from bad faith applies here directly: what am I choosing, and why, when I say “I just reacted that way”? Fast doesn’t mean uninvolved. It often just means well-practiced.
4. Hold both possibilities at once. You can take stereotypes and their real effects on behavior seriously and be skeptical of any single instrument’s claim to have measured your hidden mind with precision. Those aren’t in tension. Sartre would say the second skepticism is exactly what taking freedom seriously requires.

The Self Was Never Waiting to Be Caught

The IAT’s appeal was always partly about relief — the promise that a number could finally settle an anxious, recursive question (am I secretly biased?) that ordinary introspection can’t seem to close. Sartre would have found that appeal familiar, and worth suspecting for exactly that reason. He thought the wish for a fixed, discoverable self — good or bad, biased or clean — was itself a form of bad faith. It’s the wish to be a thing, an inkwell, something with a settled nature you could look up rather than a project you’re continuously choosing and answerable for.

What LaFollette’s team found doesn’t hand you the diagnostic certainty the test seemed to promise. It hands you something closer to what Sartre was arguing for all along: even the reactions that feel fastest and most automatic are still, structurally, yours. Not a hidden verdict revealed. An ongoing choice, most of it unexamined, waiting for exactly the kind of attention you just gave it by reading this far.

---

This post discusses a 2026 psychology study and a classical work of existentialist philosophy. It isn’t a diagnosis of anyone’s biases or a substitute for the harder, ongoing work of examining them. If questions about bias or discrimination are affecting your relationships, workplace, or wellbeing, a therapist or diversity-focused professional can help in ways a philosophy essay can’t.