Covert attention dynamically shapes decision making

Reviewer #1 (Public review):

Summary:

In this paper, the authors combine a well-controlled decision-making task with a secondary probe task in order to understand how covert attention is shaped and how it influences the ongoing decision process. The authors report that the likelihood of overtly attending an alternative is shaped both by its decision relevance and its decision value. Further, covert attentional allocation comes at the expense of overt attentional allocation and attenuates the impact that covert gazes have on choice. Importantly, covert attentional allocation is dissociable from pre-saccadic activity. This paper sheds new light on the role of covert attention in decision-making. From an experimental aspect, the authors make excellent use of behavioural and process-tracing methods, presenting an exemplary paradigm that could be valuable to researchers working in related fields.

Strengths:

(1) The paper utilises a very rich and clever experimental design, which allows for probing covert attention at the level of a single trial. Extending the paradigm to ternary choices was an excellent addition. The decision task is also well-controlled. Overall, the results are very clearly presented, and the paper is very well written.

(2) The paper addresses important yet overlooked questions in decision neuroscience: what influences covert attention during decision-making and what is the functional relevance of covert attentional shifts. It is impressive that the authors are able to tackle these questions using behavioural and eye-tracking data alone.

(3) The authors thoroughly check that the secondary probe task can indeed be used as a proxy for covert attention. Checking that covert attentional allocation is dissociable from pre-saccadic activity was an excellent control.

Weaknesses:

(1) It is not clear why the decision task is described as "value-guided" instead of "perceptual". Participants receive momentary reward on the basis of their overall accuracy, but this is common practice in perceptual tasks. A value-based analogue of this task would provide trial-by-trial reward as a function of the perceptual magnitude (decision value) of the chosen stimulus. On a related note, the secondary task was not incentivised. The authors should further justify this choice.

(2) The authors do a great job in describing the determinants of covert attention. However, the relevant analyses are rather "phenomenological". It would be useful to try and dig further into the data in order to understand at a deeper, causal level these determinants. I suggest doing so by utilising in full the rich behavioural and overt attention data this paradigm offers. Specifically, the authors show that the higher the value of an alternative, the higher the likelihood this alternative is covertly attended. Decision values, however, could impact aspects of saccadic behaviour that directly influence covert attentional allocation (saccadic speed, dwell times, locus of gaze). At a more global level, decision value and/ or value difference could affect sampling behaviour (frequency of switching) or decision times, which can manifest in probe accuracy. Overall: a) overt sampling behaviour is at present reduced to the presence of saccades; but saccadic behaviour could be decomposed into richer metrics, b) although the task is "free-response", decision times are not analysed. Metrics based on a) and b) could be tested as mediating factors, enabling the authors to better understand the causal determinants of covert attention.

(3) The authors claim that covert attention attenuates: a) the impact of the last fixation on choice, b) the "time advantage" of the alternative that was overtly attended for longer. To further qualify these claims as functional/causal, further analyses are required. A few suggestions follow below. On trials where the "unattended" probe letter was successfully reported, the last fixation may exhibit different characteristics, which could explain away its reduced impact. For example, it may differ in duration, or it might correspond to a "switching" (rather than "staying" on the same alternative) saccade. These potential covariates need to be further examined. Additionally, correctly reporting unattended probes could come at the expense of decision accuracy due to dual task demands. If that is the case, then the reported attenuation of the impact of the last fixation could just be due to noisier responses. Similar points can be raised about the "time advantage", especially given that decision times do not feature in any of the analyses. More difficult decisions may lead to higher probe accuracy but also to prolonged decision times. The authors quantify the "time advantage" using cumulative fixation time, but perhaps relative metrics (e.g., relative fixation time or cumulative fixation time normalised by decision time) are better suited for this analysis.

Reviewer #2 (Public review):

Summary:

The authors combine a value-based choice task with a gaze-contingent covert attention paradigm to investigate how covert attention relates to value-guided decisions. This is an important and timely question, as many models of decision-making implicitly equate gaze with attention despite extensive evidence that covert attention can be dissociated from eye position. I particularly appreciated the relatively naturalistic task design, which allows participants to freely explore the options while covert attention is sampled during the decision process. Overall, the manuscript is well written, the experiments are carefully executed, and the analyses are generally appropriate and clearly presented.

My main reservation concerns the strength of some conceptual claims. The data convincingly demonstrate that covert attention can be decoupled from gaze and that probe report is modulated by option value. However, I found the evidence less directly supportive of the stronger conclusions that covert attention dynamically competes with overt attention throughout the decision process and that covert attention itself causally influences subsequent choice. Many of the reported analyses are consistent with these interpretations but, in my view, do not uniquely support them.

Strengths:

The manuscript addresses an important and timely question at the intersection of visual attention and value-based decision making. The experimental paradigm is elegant and relatively naturalistic, allowing covert attention to be sampled during ongoing decision making while participants freely explore the choice display. The experiments are carefully executed, and the analyses clearly demonstrate that covert attention can be dissociated from gaze and is systematically modulated by option value. The control analyses addressing the contribution of presaccadic attention are an important addition that substantially strengthens the manuscript, even though I had questions regarding their interpretation.

Weaknesses:

(1) Interpreting the causal role of covert attention in guiding choice

The manuscript puts forward two central conclusions: first, that covert attention is shaped by decision-relevant factors such as option value; and second, that covert attention, in turn, influences subsequent choice behavior. I found the evidence for the first conclusion compelling. My reservations concern the second conclusion, namely whether the current data uniquely establish a causal influence of covert attention on choice. Throughout the manuscript, the authors conclude that covert attention exerts "downstream consequences for choice behavior" and that an estimate of covert attention "influences the final choice." However, I am not sure that the presented analyses fully disentangle a causal influence of covert attention from a common influence of option value on both covert attention and the eventual decision.

One aspect that illustrates this concern is the interpretation of the last-fixation bias. While the reported association between the final fixation and the chosen option is clear, it is less obvious that the final fixation itself biases the subsequent choice. An equally plausible interpretation is that participants have largely committed to a decision and subsequently direct one final gaze shift toward the option they have already selected before executing the button press. In other words, the relationship may reflect choice influencing gaze rather than gaze influencing choice. Unless the temporal dynamics allow these alternatives to be distinguished, I would encourage the authors to use more neutral language (e.g., an association between last fixation and subsequent choice) or explicitly discuss this alternative interpretation.

More generally, I found myself looking for a more direct demonstration that covert attention biases choice toward the attended option. Most analyses ask whether successful peripheral probe report attenuates established gaze-related choice biases, such as the last-fixation or time-advantage bias. While these are interesting findings, they remain relatively indirect. A particularly compelling demonstration would be to examine situations in which option value cannot explain the relationship between covert attention and choice.

(2) Dynamic competition between covert attention and overt attention

I think one of the key strengths of this manuscript is its explicit attempt to dissociate covert attention from gaze-a distinction that is often overlooked in both empirical studies and computational models of value-based decision making. My reservation concerns the stronger claim that covert attention is dynamically reallocated during decision making and competes with overt attention. The present results clearly show that covert attention can be dissociated from gaze, but I found less direct evidence for the proposed dynamic competition. Throughout the experiments, probe report accuracy remains highest at the currently fixated option, whereas peripheral benefits, although reliable, are comparatively modest (e.g., Figures 2 and 3). This raises the question of how often, and under which circumstances, covert attention is actually reallocated away from the current fixation.

Because the authors possess precise eye-movement timing, I think the manuscript would be substantially strengthened by an analysis of attentional dynamics within individual fixations, to provide more direct support for the proposed dynamic interplay between covert and overt attention.

More generally, I found several interpretations somewhat stronger than the presented evidence. For example, the statement that increased covert attention to higher-valued peripheral options occurs "at the expense of" overt attention at the fixated location seems to imply a direct trade-off that is not explicitly demonstrated. Unless such temporal dynamics can be shown, I would encourage the authors to distinguish more clearly between demonstrating that covert attention can be dissociated from gaze and demonstrating that it dynamically competes with overt attention throughout the decision process.

(3) Dissociating covert from presaccadic attention

The control analyses aimed at dissociating covert from presaccadic attention are an important strength of the manuscript. However, I found the rationale underlying these analyses difficult to follow. The key conclusion-that value modulates probe report even when covert attention is not presaccadic-rests on the distinction illustrated in Figure 6, yet it remained unclear to me exactly how this distinction isolates presaccadic from non-presaccadic covert attention. My understanding is that the authors compare probe performance at previously fixated locations depending on whether those locations are subsequently re-fixated. If this interpretation is correct, a more explicit explanation of why this operationalization isolates presaccadic attention would help readers evaluate this conclusion. Conceptually, I found it more intuitive to think about a dissociation based on probe performance at locations that are subsequently fixated versus not subsequently fixated following the first fixation, when the currently fixated option, the upcoming saccade target, and a third unfixated option are distinct. Such situations would seem to provide the clearest opportunity to separate attentional enhancement associated with the upcoming saccade target from covert attention directed elsewhere.

I also found the interpretation of the time-to-saccade control stronger than the evidence directly supports. The authors argue that a purely presaccadic account predicts a monotonic decline in probe accuracy as a function of time-to-saccade. This assumption seems stronger than supported by the presaccadic attention literature, which primarily characterizes attentional enhancement during approximately the final 100-200 ms before saccade onset. Within this temporal window, the present data actually appear consistent with such an increase (Figure S6). By contrast, I am not aware of theoretical or empirical work predicting a monotonic presaccadic attentional shift extending hundreds of milliseconds, or even a second, before saccade onset.

The observed U-shaped relationship is nevertheless inconsistent with a simple presaccadic-only account, as probe performance remains elevated even when the upcoming saccade is relatively distant. I therefore agree that the findings suggest additional attentional processes. However, I do not think they uniquely establish that probe performance indexes covert attention independently of presaccadic planning.

Author response:

We thank the reviewers for their thoughtful and constructive feedback on our work.

We have attempted to synthesize the main suggestions across both reviewers into four main areas, which we plan to address in a revised version of the manuscript.

(1) The relationship between covert attention, overt attention, and choice

While the current analyses address elements of relationship between covert attention, overt attention, and choice, the reviewers highlighted an opportunity to make full use of our rich gaze data to examine more explicitly how these processes are related to one another over the course of a decision. [Reviewer #1 Point 2, Point 3 i; Reviewer #2 Point 2 para 2]

(2) The role of decision difficulty and decision time

The reviewers point out that decision difficulty may i) account for the relationship between covert attention (probe response correctness) and choice; and ii) influence decision times, which are currently not analyzed; and that these considerations may affect how we interpret the relationship between covert attention and choice. [Reviewer #1 Points 2, 3 ii and iii; Reviewer #2 Point 1 paras 1 and 3]

(3) Evidence for competition between covert and overt attention

The reviewers requested more direct evidence for our claim that covert attention to higher-valued peripheral options occurs "at the expense of" overt attention [Reviewer #2 Point 2 paras 1 and 3, Point 3 para 3]

(4) Figure 6: Presaccadic vs. non-presaccadic attention

Reviewer 2 found the rationale behind the analysis distinguishing presaccadic from non-presaccadic covert attention difficult to follow and suggested and alternative to make this comparison more explicit [Reviewer #2 Point 3].

In response to these suggestions, we will include additional analyses in the revised manuscript which more directly examine the relationship between covert attention, overt attention, and choice. We will also introduce new analyses which more simply and directly depict: why decision difficulty cannot wholly account for the link between covert attention and choice; a competition/trade-off between overt and covert attention; and that value modulation existed at non-presaccadic options. In addition to these four main areas, we will clarify several elements of the task design and analysis which may not have been sufficiently clear in the manuscript.

Here, we offer a provisional response to each of the points above.

(2) The relationship between covert attention, overt attention, and choice

We agree that the relationship between covert attention, overt attention, and choice could be examined in more detail using the available eye-tracking data. In the revised manuscript, we plan to include further analysis looking at fixation and sampling measures before the decision is made.

For example, to investigate the relationship between covert attention and choice (as reviewer 1 suggested), we plan to include an analysis which asks about how decision accuracy changes as a function of decision difficulty (relative option value) when the unfixated probe is correctly reported, versus when it is not. This will straightforwardly test for the influence of probe report on choice, controlling for the influence of decision difficulty.

More generally, we will revise the manuscript to distinguish more carefully between associations demonstrated by the present data and stronger causal interpretations. To address this, we will either attempt more direct statistical tests of causality, or use more neutral language where the causal directionality between fixation, covert attention, and choice cannot be uniquely established by the present design.

An important clarification on the dynamics of the task is relevant to several of the reviewers’ comments concerning the last-fixation analyses. Probe performance was generally measured earlier than the final fixation and choice – as it was always at the first to third option visit by the participant (or only first to second option visit in Experiment 1). Only 10.4% of probes occurred during the final option visit, and on average, participants made 2.8 further option visits after probe presentation. We can see that our original presentation of these results may have been difficult to follow. We will make this clarification in the revised manuscript, particularly where we refer to “last-fixated” or “last-unfixated” options. These terms refer retrospectively to which option was ultimately visited or not visited last before choice, rather than to the fixation occurring at the time the probe was presented.

(2) The role of decision difficulty and decision time

We agree with the reviewers that decision difficulty and in turn decision time should be considered more carefully.

We will make explicit that relative option value was already controlled for in the existing analyses relating to covert attention and choice. Specifically, for the last-fixation analysis, relative value difference was included as a control variable in the regression. For the time-advantage analysis, we use a corrected dependent variable which is intended to remove possible influences of relative value (i.e., decision difficulty) on the fixation durations, following what has been done in previous work (e.g., Eum et al., 2023; Krajbich et al., 2010; Krajbich & Rangel, 2011).

We additionally plan to include new analyses which more directly examine the role of decision difficulty and decision time. In particular, we will test whether the relationship between peripheral probe performance and choice remains after accounting for decision difficulty, and whether decision time differs as a function of probe performance.

We would also like to take the opportunity to clarify that “time advantage” was operationalised as a relative measure: the difference in cumulative dwell time between the two relevant options, rather than the absolute cumulative dwell time on a single option. This definition will be stated more clearly in the main text and figure caption.

Finally, to address the possibility that peripheral probe report errors reflect a general dual-task cost or increased response noise, we will examine decision accuracy as a function of decision difficulty separately for trials in which the peripheral probe was reported correctly versus incorrectly. This will help determine whether successful peripheral probe report is associated with poorer decision performance.

However, we should note that a closer examination of Figure 4 offers clues to this question; it suggests that correctly reporting the probe at the last-unfixated item makes people more accurate, rather than happening at the expense of decision accuracy. To explain this:

When objectively Option 1 is worse than Option 2 (i.e., to the left of x-axis center), choice for a last-fixated Option 1 (solid line) is unfairly boosted; but if the last-unfixated probe was correctly reported, then the choice proportion is correctly closer to 0%.

When objectively Option 1 is better than Option 2 (i.e., to the right of x-axis center), choice for a last-unfixated Option 1 (dotted line) is unfairly discounted; but if the last-unfixated probe was correctly reported, then the choice proportion is correctly closer to 100%.

That is, the bias-attenuation effect of correctly reporting the probe at the last-unfixated item seems to operate on 1) reducing choice boosting of the objectively-worse fixated option (left of x-axis center, solid line is further down on right panels than left panels); and 2) increasing choice boosting of the objectively-better unfixated option (right of x-axis center, dotted line is further up on right panels than left panels).

(3) Evidence for competition between covert and overt attention

To address more directly the question of whether covert attention occurs at the expense of overt attention, we will add an analysis testing whether successful report of a probe at an unfixated option negatively predicts report of the simultaneous probe at the fixated option. A negative relationship would provide a more direct test of the predicted trade-off between attentional allocation to the fixated and peripheral locations. We will revise the strength of the corresponding language in the manuscript according to what this analysis supports.

(4) Figure 6: Presaccadic vs. non-presaccadic attention

We agree that the rationale underlying the current Figure 6 analysis is too implicit in the current manuscript.

To clarify the underlying rationale: The key logic of the analysis was to ask whether value modulation of probe performance is present at an unfixated option even when that option is not the target of the upcoming saccade. In the original analysis, we only compared attention at peripheral options that had previously been viewed: we compared attention at peripheral options that had previously been viewed and were subsequently fixated next, with those that had previously been viewed but were not subsequently fixated next. The requirement that the option had previously been viewed arose because, in the gaze-contingent “hidden” condition, the value information of an unvisited peripheral option would not yet have been available to the participant, and therefore there would be no possibility of any value modulation effects existing at that location.

Although this analysis was intended to distinguish value modulation at likely presaccadic targets from value modulation at other peripheral locations, we now see that its rationale is not immediately transparent. In the revised manuscript, we therefore plan to replace or supplement this analysis with a simpler visualisation and analysis that addresses the same question more directly.

We will also revise text associated with the time-to-saccade analysis. Our use of the term “monotonic” was imprecise. The intended point was not that a purely presaccadic account predicts a strictly monotonic change across the full time-to-saccade interval, but rather that the observed U-shaped relationship is difficult to reconcile with a simple presaccadic-only account, because probe performance remains elevated even when the upcoming saccade is relatively distant. We will clarify this point and more carefully distinguish evidence inconsistent with a simple presaccadic-only explanation from the stronger claim (which we did not intend to make) that presaccadic contributions have been completely excluded.

(5) Additional clarifications

We agree with Reviewer 1 that the task can reasonably be described as perceptual rather than strictly value-based, but we believe that our central concerns regarding how covert and overt attention interact with multi-alternative decision-making still stand.

We will clarify why the secondary probe task was not incentivized. The probe task was intended to provide an incidental measure of covert attention while participants performed the primary decision task. Incentivizing probe performance could have encouraged participants to explicitly trade off performance on the two tasks or to alter their allocation of attention in response to the secondary task.

In summary, the revisions will combine clarification of several existing analyses with the addition of new analyses that more directly test the reviewers’ central concerns. In particular, we will more explicitly address whether the relationship between covert attention and choice remains after accounting for option value/decision difficulty; whether peripheral and fixated probe performance show evidence of an attentional trade-off between overt and covert attention; how decision time and richer measures of overt sampling relate to covert attention; and whether value modulation is present at peripheral locations that are not the target of the upcoming saccade. We will also moderate causal and mechanistic language where the current data do not uniquely establish directionality.