A Bigger Surprise During Exposure Doesn't Buy a Better Outcome
- A longitudinal study tested the core mechanistic claim of the inhibitory-learning model of exposure therapy: that the size of the "threat prediction error" during exposure drives clinical improvement.
- Forty-three adults in exposure-based CBT for social anxiety disorder logged 523 in- and between-session exposures via smartphone (ecological momentary assessment).
- The manipulation worked as expected – unexpected outcomes felt more surprising and more instructive than expected ones – but the size of the threat prediction error did not predict anxiety at the next session or symptom change at 3-month follow-up.
- The null result on the primary hypothesis cuts against a popular clinical heuristic: that engineering a maximally dramatic disconfirmation of the feared outcome makes exposure work better.
Exposure therapy has a leading explanation for why it works: each time a feared outcome fails to materialize, the brain registers a "prediction error" – the gap between what was expected and what happened – and updates its threat model accordingly. The bigger the gap, the theory goes, the more updating. A team at Swinburne University of Technology and the University of Melbourne put that claim to a direct test, outside the lab, in real exposures done by real patients.
What the Numbers Show
After each of 523 exposures, participants reported their threat expectation, the actual outcome, anxiety, surprise, and a sense of having learned something. Multilevel models linked the resulting prediction-error scores to next-session anxiety and to 3-month outcomes. The manipulation check held: unexpected outcomes did register as more surprising and more instructive. The prediction itself did not. Prediction-error magnitude was unrelated to anxiety reduction the following session and unrelated to overall symptom change three months later.
Improvement Without Its Assumed Cause
None of this says exposure failed – most patients still improved. It says the mechanism proposed to explain the improvement, at least measured this way, didn't show up.
A Practical Correction
For a clinician building an exposure hierarchy around "make the disconfirmation as dramatic as possible," that's a specific piece of guidance losing its evidentiary floor. Chasing surprise for its own sake is not obviously the lever worth pulling. The size of the gap between expectation and reality never carried the weight the theory assigned it.
The exposure worked. The theory about why did not.
Observational and correlational, not a randomized trial – the design can't rule out other drivers of outcome. Single disorder (social anxiety) and single clinical sample, so generalization to other exposure-based protocols is unproven. The paper itself is behind a paywall; the figures above come from the published abstract only.