By Anke Braun & Tobias Donner.
Our decisions are often influenced by the history of our preceding choices: We tend to repeat (or alternate) decisions more often than expected by chance. Computational models postulate that these choice history biases result from the across-trial accumulation of a quantity called ‘decision variable’. It is commonly assumed that the decision variable is encoded in associative regions of the cerebral cortex (e.g. posterior parietal or prefrontal cortex). The sign of the decision variable dictates the choice and its magnitude the confidence about the correctness of that choice.
We here provided psychophysical evidence for the across-trial accumulation of the decision variable as a basis for choice history bias, and we also established the adaptive utility of this mechanism. In two experiments, we asked participants to judge whether a cloud of coherently moving dots embedded in randomly moving dots moved upwards or downwards and to report their choice by a button press. In the first experiment we decoupled their perceptual choices (up or down) from the motor responses used to report the choice (press a button with the left or right hand) on each trial. This showed that participant’s choice history biases were due to their previous perceptual choices, but not their motor responses.
In a second experiment we manipulated the statistical structure of the environments in which participants performed the perceptual choice task – specifically, the correlations among stimulus identities across trials (Figure 1A). In a so-called Neutral environment, motion directions were selected from trial to trial at random, as in standard psychophysical experiments. In a Repetitive environment, directions were more likely to be repeated. In an Alternating environment, directions were more likely to be alternated. We found that participants consistently adjusted their choice history biases to these different environments (Figure 1B and C). This was striking because participants performed the task under overall high uncertainty (around psychophysical threshold) and did not receive any trial-by-trial feedback about the correctness of their choices. Further, this adjustment improved their overall performance to a similar extent as their perceptual sensitivity. Finally, we found that the strength of the bias adjustment scaled with two proxies of the confidence about the correctness of the previous choice, with a stronger bias after correct (Figure 1D) or fast choices (Figure 1E).
Taken together, our results are in line with the idea that choice history biases originate from an action-independent, context-dependent accumulation of previous choices modulated by previous decision confidence. The next step of our research is to uncover the dynamical signatures of this adaptive process in the brain.