Context changes retrieval of prospective outcomes during decision deliberation

Abstract

Foreseeing the future outcomes is the art of decision-making. Substantial evidence shows that, during choice deliberation, the brain can retrieve prospective decision outcomes. However, decisions are seldom made in a vacuum. Context carries information that can radically affect the outcomes of a choice. Nevertheless, most investigations of retrieval processes examined decisions in isolation, disregarding the context in which they occur. Here, we studied how context shapes prospective outcome retrieval during deliberation. We designed a decision-making task where participants were presented with object-context pairs and made decisions which led to a certain outcome. We show during deliberation, likely outcomes were retrieved in transient patterns of neural activity, as early as 3 s before participants decided. The strength of prospective outcome retrieval explains participants' behavioral efficiency, but only when context affects the decision outcome. Our results suggest context imparts strong constraints on retrieval processes and how neural representations are shaped during decision-making.

Keywords: EEG; context; decision-making; multivariate decoding; outcome retrieval.

Fig. 1

Task design and behavioral outcomes. A) Design decision-making task. Each trial started with an object (gardening tool) which was presented for 1.5 s. then the context (fictional season) was presented together with the object for 1.5 s. This period of the first 3 s was considered as decision deliberation. Next, participant’s decisions on which market to sell the gardening outcome given the object and context were asked. After the participant’s response, the outcome was presented together with a sign (tick or cross) for 1 s and followed by a 1.5 to 2.5 s long fixation cross. The thought balloon indicates the deliberation period before a decision was made. B) The types of decisions in the task. A decision was context-dependent if object and context together determined the outcome. It was a context-free decision when only object was enough to determine the decision outcome. C) Exemplar object-context-outcome association table. Three gardening tools (the most left column) were used as objects and 2 geometric shapes (top row) indicated the fictional seasons. Every object-context pair was associated with a gardening outcome. For 2 objects, context determined the likely outcome of a given object (first 2 rows of the table) whereas for the control condition outcome was independent of context (last row of the table). d) Behavioral accuracy in context-free was significantly higher than in context-dependent decisions [Wilcoxon signed-rank test, z(25) = 3.14, P < 0.002]. e) RT for context-dependent decisions was significantly longer than for context-free decisions [Wilcoxon signed-rank test, z(25) = −3.42, P < 0.001]. In B and C, each dot corresponds to a participant’s average behavioral accuracy in context-free or context-dependent decisions.

Fig. 2

Schematic representation of our data analysis. Top: patterns for the three outcome images were decoded by training classifiers on EEG responses during the functional localizer, as the first step of the analysis. Bottom: the decoded and tested patterns were then used to detect outcome representation during deliberation, before making a decision. We first reconstructed the probability of outcome representations at every time point on the single-trial EEG responses during deliberation based on the classifiers trained to discriminate the three outcomes. As an outcome can be retrieved at any time during deliberation, we analyzed the strength and temporal structure in retrieval of outcome representations via PSD and autocorrelation of reconstructed probabilities on single-trial level.

Fig. 3

EEG responses to outcomes and decoding of outcome identity from EEG responses. A) Average EEG responses to outcome images presented during the functional localizer across the group of participants. B) Time point-by-time point decoding of the identity of the three outcomes. Black solid line shows the group average of decoding accuracy whereas black dashed line shows the chance level. Gray-shaded areas show standard error. Horizontal bar with asterisks (*) indicates time periods in which decoding accuracy was above chance. Gray dashed vertical lines at 0.09 and 0.23 s show the time period used for training final decoders which were used to detect outcome representations during deliberation.

Fig. 4

Outcome retrieval and its relation with behavioral performance. A) Strength of outcome retrieval (solid black line) vs. chance (dashed line), quantified via power spectral analysis on reconstructed outcome probabilities. Chance level was evaluated based on the randomly permuted classifiers of outcome images. Horizontal bar with asterisks (*) indicate frequencies which were different from chance. B and C) Neuro–behavioral coupling between the behavioral performance and strength of outcome retrieval. Regression of behavioral performance on strength of outcome retrieval in B) context-free and V) context-dependent conditions. In context-free decisions, there was no significant regression of behavioral performance on strength of outcome retrieval [panel B, linear regression, F(1, 24) = 0.08, P = 0.78]. However, the regression was significant for context-dependent decisions [panel C, linear regression, F(1, 24) = 6.98, P = 0.014]. For context-dependent decisions (panel C), single-subject results are provided for 2 exemplar participants (1 with high behavioral performance, and another 1 with low behavioral performance). D and E) Neuro–behavioral coupling between the RTs and strength of outcome retrieval for D) context-free and E) context-dependent decisions. In context-free decisions, there was no significant regression of RT on the strength of outcome retrieval [panel D, linear regression, F(1,24) = 0.022, P = 0.88], which was the case for context-dependent ones [panel E, linear regression, F(1,24) = 5.39, P = 0.03].

Fig. 5

Neuro–behavioral coupling between the change in behavioral performance and in strength of outcome retrieval in A) context-free and B) context-dependent decisions. In context-free decisions, there was no significant regression of change in behavioral performance on change in retrieval strength [panel a, linear regression, F(1, 44.91) = 1.7, P = 0.2]. However, the regression was significant for context-dependent decisions [panel B, linear regression, F(1, 101.86) = 10.52, P = 0.0016]. Each data point corresponds to a pair of object and context for a participant. For visualization purposes the data points from each participant were not grouped together however, analysis was done with a linear mixed effect model, taking into account the participants as a random factor.

Fig. 6

Effect of context on outcome retrieval. A) Time course of reconstructed outcome probabilities during decision deliberation, for context-free (green) vs. context-dependent (pink) decisions. Time zero corresponds to the time when context is revealed. Outcome retrieval was significantly stronger for context-dependent than context-free decisions, between 0.16 and 0.23 s post-context onset (Wilcoxon signed-rank test, P_uncorr < 0.05). B) Time course of reconstructed outcome probabilities, time-locked to the period when an object was presented alone without context (blue line) vs. the time period when the context was presented together with the object (green line), for context-free decisions. The time courses of reconstructed probabilities were identical, which suggests that for the case of context-free decisions, they were not affected by context. C and D) Neuro–behavioral coupling in context-dependent decisions after context presentation was significant for the c) relevant outcome [left panel, solid line, linear regression, F(1, 24) = 5.39, P = 0.03], but it was nonsignificant for the D) irrelevant outcome [right panel, dashed line, linear regression, F(1, 24) = 1.79, P = 0.19].