I’m happy to announce that Human Brain Mapping published our paper on the neural correlates of the inverse base-rate effect on Friday; it’s available as open access here. The lead author was Angus Inkster.
The inverse base-rate effect is an example of irrational decision making. Participants learn that two symptoms (let’s call them A and B) predict a common disease (“Jominy Fever”). The also learn that two overlapping symptoms (A and C) predict a rare disease (“Phipps syndrome”). When later asked to decide whether symptom A alone is more likely to predict Jominy or Phipps, people say “Jominy”. This is rational. Symptom A has been associated with both diseases, but Jominy is the more common disease, so that’s more likely to be the correct answer. This is called following the base rates. However, when asked about the novel combination of symptoms B and C, people predict the rare disease. This is in opposition to the base rates, and it is difficult to come up with any rational account of this behaviour.
It can, however, be explained by an account that assumes attention is driven by the avoidance of error. The symptom compound AB is more common than AC, and so people learn AB -> Jominy first. Then, when they see AC, they tend to also predict Jominy fever, which is an error. In order to avoid making that error in future, their attention shifts away from A and towards C. This is effective in avoiding error, but the learned attention to C persists, meaning that when they see BC, they’re looking mostly at C rather than B.
This account of the inverse base-rate effect is most fully worked out in Kruschke’s EXIT model, and has previously been supported by behavioural, eye-tracking, and EEG evidence. In the paper we just had published, we add fMRI data to the mix. The picture above shows that regions of the brain that have previously be associated with the computation of prediction error are more active for symptom C than symptom B (relative to frequency-matched control stimuli). This suggests, as previously discussed by, for example, Elsa Fouragnan, that brain areas that seem to be involved in the calculation of prediction error may in fact be representing persistent changes in attentional processing.
Interestingly, our results are somewhat different to those of the only other fMRI study to look at the inverse base-rate procedure (O’Bryan et al., 2018). We put these differences in brain activation down to the fact that participants in that previous study did not show an inverse base-rate effect. Nonetheless, further research may be merited.