Figures and data
Example of a single trial.
Trials began with an intertrial interval followed by a pre-trial fixation cross. Three circular patches were presented onscreen; their contents vary by experiment and peripheral option visibility condition. Choice-relevant options each contained 100 dots (cyan or orange; one of which was designated the target colour). In Exp. 1, there were two choice-relevant options such that the third option was choice-irrelevant (i.e., blank), whereas in Exp 2, all three options were choicerelevant. On trials in the visible condition, peripheral information was visible during sampling regardless of gaze location; on trials in the hidden condition, patch contents were gaze-contingent, such that peripheral information was hidden. Participants saccaded freely between patches and at any time could select an option via keypress. Selecting the option with the most dots of the designated target colour earned a reward. A secondary attentional probe task was integrated into this primary choice task. During the trial, participants were briefly presented with probe letters for 150 ms at each of the three patch locations, at a pseudorandom timepoint during the decision; in Exp. 1, this was 220 ms after either the 1st- or 2nd-visited option, and in Exp. 2, this was 220 ms after either the 1st, 2nd, or 3rd-visited option. The trial then continued after letter-probe presentation until a selection was made via keypress. After selection, participants reported probe-letters they detected during the trial using a probe-response screen. Selecting the patch with the most dots of the designated target colour (e.g., cyan) earned a reward.
Relative value modulated the accuracy of both peripheral and fixated location probe letter reports.
Fixated and unfixated options were defined at the time of probe. (A) We asked whether probe response accuracy for the choice-relevant unfixated option was modulated by its value relative to the fixated option in Experiment 1, and (B) whether probe response accuracy for each of the two unfixated options were modulated by their respective values relative to the fixated option, in Experiment 2. (C-D) Across both experiments, probe report for an unfixated option became more accurate as the option became more valuable than the fixated option. (E-F) On the other hand, probe report for the fixated option became less accurate as the relative value of peripheral option/s increased (in Experiment 2, this was defined as the average value of the two unfixated options).
Choice-relevance modulated the correctness of probe reports at peripheral locations.
Fixated and unfixated options were defined at the time of probe. (A) We asked whether probe responses were more accurate for choice-relevant unfixated options compared to a completely choice-irrelevant unfixated location in Experiment 1. (B) In Experiment 2, we asked whether probe responses were more accurate for the unfixated option that was more choice-relevant (unfixated option with higher value; unfixatedhigh) compared to a less choice-relevant, lower-valued unfixated option, unfixatedlow . (C) In Experiment 1, probe report accuracy was higher for probes that appeared at the choice-relevant unfixated option relative to the unfixated choice-irrelevant option (p < .001), and generally higher when the peripheral option was visible compared to when it was hidden (p = .029). (D) In Experiment 2, probe report accuracy was higher for the unfixatedhigh relative to the unfixatedlow option (p < .001), and also generally higher when the peripheral option was visible compared to when it was hidden (p < .001).
Correct peripheral probe reports reduced fixation-related choice biases associated with the last-fixated option.
(A) We obtained an estimate of covert attention earlier in the decision process by using an attention probe, then investigated whether this measure affected the final choice. (B) We compared the scenario when the probe at the last-unfixated option was incorrectly reported, compared to when it was correctly reported. (C-D) A choice bias linked to the last-fixated option was observed across both experiments, where participants were more likely to choose the last-fixated option compared to the last-unfixated option, controlling for relative value (dashed line vs. solid line). This fixated-related choice bias was attenuated when the probe at the last-unfixated option was correctly detected from earlier on in the trial, as shown in the reduction in difference between the dashed line vs. solid line in the left vs. right subpanels. (E-F) This plot re-visualizes the last-fixation effect (choice bias) - defined as the difference between the dashed line vs. solid lines in Panels C-D, collapsed across all value bins – as a function of probe report correctness for last-unfixated option. The interaction between effects of last-fixated option and probe report correctness for last-unfixated option was significant across both experiments (ps < .05), suggesting that the choice bias was attenuated when the probe at the last-unfixated option was correctly reported.
Correct peripheral probe reports reduced fixation-related choice biases associated with excess looking time.
(A) We examined the effect of cumulative fixation/dwell time (“time advantage”) on participants’ likelihood of choosing a given option. (B, D) A choice bias linked to time advantage was found in both experiments. The plots show the time advantage towards an arbitrarily defined Option 1 on the x-axis, and the corrected choice probability on the y-axis (corrected such that this probability should be 0 independent of time advantage, if a relationship between this variable and time advantage did not exist; i.e., the gray line). The plotted data has been subsetted to trials where the comparator option was also unfixated at probe. Increasing time advantage towards an option increased its choice probability (p < .001). (B) In Experiment 1, the comparator option was the other option in the binary choice set (Option 2). (C) We found evidence that the time-advantage-linked choice bias was attenuated when the probe letter at a peripheral comparator option was correctly reported. (D) In Experiment 2, the comparator option was the lower-dwelled alternative option out of the two possible alternatives. (E) We again found evidence that the time-advantage-linked choice bias was attenuated when the probe letter at a peripheral comparator option was correctly reported.
In multialternative choice (Experiment 2), covert attention at a peripheral option was influenced by value for both non-presaccadic and presaccadic attention.
(A) Panel illustrates how options would be defined as fixated, previous-visited unfixated, or next-visited unfixated on a hypothetical trial. On this example trial, the top option is fixated at the time of probe, while the option the participant fixated before this was the right option, and the option the participant fixates after this is the left option. Plots illustrate the probe report accuracy for the previous-visited unfixated option (B) when that option was different from the next-visited option (i.e., indexing non-presaccadic attention; an example of this scenario is depicted in (A)), versus (C) when that option was the same as the next-visited option (i.e., indexing presaccadic attention).
Main text Figure 2 disaggregated by peripheral option visibility and probe timing conditions.
(A) Data from Experiment 1 (corresponds to main text Figures 2C and E. (B) Data from Experiment 2 (corresponds to main text Figures 2D and F).
Results from full model assessing the effect of relative value on probe correctness at the choice-relevant unfixated option, Experiment 1.
Results from full model assessing the effect of relative value on probe correctness at the unfixated options, Experiment 2.
Results from full model assessing the effect of relative value on probe correctness at the fixate option, Experiment 1.
Results from full model assessing the effect of relative value on probe correctness at the fixated option, Experiment 2.
Results from full model assessing the effect of value on probe correctness at the unfixated option (separate regressor terms for options’ values), Experiment 1.
Results from full model assessing the effect of value on probe correctness at the fixated option (separate regressor terms for options’ values), Experiment 1.
Results from full model assessing the effect of value on probe correctness at the unfixated option (separate regressor terms for options’ values), Experiment 2.
Results from full model assessing the effect of value on probe correctness at the fixated option (separate regressor terms for options’ values), Experiment 2.
Results from full model assessing the influence of the lower-valued unfixated option on probe report correctness at the higher-valued unfixated option, Experiment 2.
Results from full model assessing the influence of the higher-valued unfixated option on probe report correctness at the lower-valued unfixated option, Experiment 2.
Main text Figure 2 disaggregated by probe timing condition.
Experiment 1, full generalized linear mixed model assessing effects on probe report correctness.
Control analysis: Main model reported in Table S1, controlling for distance of probe letter to gaze.
Experiment 2, full generalized linear mixed model assessing effects on probe report correctness.
Results from full model assessing the last-fixation choice bias, Experiment 1.
Results from full model assessing the last-fixation choice bias, Experiment 2.
Main text Figure 4C (last-fixation choice bias results for Experiment 1), disaggregated by peripheral option visibility condition.
Main text Figure 4D (last-fixation choice bias results for Experiment 2), disaggregated by peripheral option visibility condition.
Results from full model assessing the last-fixation choice bias, subset to instances where probe report correctness indexed covert attention only, Experiment 1.
Results from full model assessing the last-fixation choice bias, subset to instances where probe report correctness indexed covert attention only, Experiment 2.
Time advantage choice bias results, disaggregated by peripheral option visibility corresponding to (A) main text Figure 5B (Experiment 1) and (B) main text Figure 5D (Experiment 2).
Results from full model assessing the fixation time advantage choice bias, Experiments 1 and 2.
Full model assessing the effect of upcoming visit on accuracy of probe report at a peripheral location, Experiment 2.
Full model assessing value modulation for peripheral locations that are not upcoming gaze target, Experiment 2.
U-shaped relationship between the time to an upcoming saccade to an unfixated option, and the option’s associated probe report accuracy.
(A) Figure schematic illustrating key regressors: time to an upcoming saccade to an unfixated option, and the probe report accuracy to that unfixated option. There was a non-monotonic U-shaped relationship between the two variables in both (B) Experiment 1 (p < .001) and (C) Experiment 2, suggesting that the covert attention indexed by probe reports was not purely presaccadic.
Relationship between time to an upcoming saccade and probe correctness at option towards which the saccade was directed: Likelihood ratio tests comparing linear versus polynomial models.
