Slowing the body slows down time perception

Acceptance summary:

This study investigates the effects of viscous movement environments on temporal processing. The results show that systematically impeding movement during interval timing leads to a subsequent compression of perceived duration in multiple tasks. Computational modeling shows that the effect of movement is directly on time perception and not on decision-making. This is an interesting and well performed study, providing novel insights into the interaction between time perception and reaching movements affected by viscous environments.

Decision letter after peer review:

Thank you for submitting your article "Slowing the Body slows down Time (Perception)" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Hugo Merchant as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Michael Frank as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Clare Press (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below extensive revisions are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

This study investigates the effects of viscous movement environments on temporal processing during a single interval categorization or a reproduction task using durations in the range of seconds. Healthy volunteers timed auditory intervals while moving a robotic arm that randomly applied four levels of viscosity. The results showed a consistent decrease in the bisection point (or point of subjective equality) in the categorization task and a decrease in duration of the produced intervals in the reproduction task as a function of viscosity. A drift-diffusion model and a Bayesian observer model suggested that the underestimation of intervals in the two tasks are due to changes in time perception instead of biases in decision making.

Essential revisions:

This is an interesting and well performed study, providing novel notion on the interaction between time perception and reaching movements affected by viscous environments. Nevertheless, we have a set of serious concerns that need to be addressed in the present version of the paper, as follows:

1. As the authors outline, previous studies have shown that a variety of movement features impact upon duration judgements. The novelty of their study is suggested to be that movement is restricted by environmental factors rather than 'self' modulations. E.g., the finding reflected in the title of "slowing the body slows down time" has already been demonstrated, by my understanding, in four experiments by Yon et al. (2017, JEPG). However, by my reading both in previous studies and the present studies in fact it is environmental modulations that result in self modulations. I.e., I did not comprehend the difference and this needs clarifying for the theoretical value of these studies to be understood. E.g., previous studies have generated the cells in their design by presenting targets at different locations (e.g., Yon et al., 2017 – external modulation) and participants then choose to move to those targets (volitional component). The present studies create the cells by imposing different forces upon the lever while keeping the targets in the same location. In all studies they are choosing to move to targets at different places and under specific environmental viscosities. The authors describe using sensory feedback due to these environmental features but they would use sensory feedback to monitor both visual target location and forcefields – both of which are constant during a specific movement in previous and present work. Once this line is clarified, what would be the assumed different mechanisms involved in the two cases?

2. The values on bisection point show enormous individual differences between subjects, with a range between 1.5 to 3 seconds, why? In contrast, the effect of viscosity on the BP is very subtle and a large proportion of subject showed negative slope of BP as a function of viscosity (please report). This means that in a subgroup of subjects, an increase in viscosity was accompanied with an increase instead of a decrease in the length of perceived time. This finding does not support one of the mayor conclusions of the paper.

3. In the present studies a number of features of the movement are confounded as viscosity is manipulated. As viscosity increases, the length traversed goes down in a given period of time, the force goes up, the speed goes down. As far as I could see the authors do not disentangle which movement feature was responsible for the influence on duration judgements. My guess is that the correlations are high between these features but perhaps there is enough natural variation in the dataset to disentangle. Could a regression analysis be performed to establish the unique contributions of the different movement parameters? As well as helping the reader to understand what is driving the effects it may resolve some of the inconsistencies with respect to previous literature. E.g., the authors observe that fast movements are associated with increased duration judgements, in contrast with some previous findings. They explain the difference as possibly driven by the fact the conflicting studies have unfilled intervals. However, they do not explain why this may alter the operation of the mechanism. Additionally, there is an experiment in Yon et al. (2017) where high speed is associated with compressed estimates and the intervals are filled here.

4. Based on DDM fits authors claim that the effect of viscosity was perceptual (and not through decision-biases). This claim is based on the observation that the viscosity level did not affect the starting point or threshold setting but it exclusively modulated the drift rate. Although on the face value, this appears as a valid conclusion based on the model fit results, the methodological details actually put this conclusion into question by invalidating the model fits (unless participants could predict the next viscosity level – see alternative explanations below). Let me elaborate more on this. The standard assumption in the conventional DDM that authors also used in their analysis is that the threshold is set adaptively in response to the context the participant is in (e.g., taking into account the response-to-stimulus interval, reward/penalty ratio, whether instructions emphasize speed or accuracy, prior probability of different alternatives). In other words, the decision threshold and starting point of the decision process are modulated in response to some average representation of the task statistics or in response to a discriminative stimulus that allows participants to predict the characteristics of the next trial (e.g., a trial with vs. without penalty for errors or trial in which the viscosity will be higher vs. lower). But in the current study, the viscosity levels were manipulated randomly from one trial to the next (without any discriminative signaling) and thus participants could not have adjusted the decision thresholds or starting points (unless authors are willing to step out of the conventional DDM domain in which case they have to assume within-trial dynamic modulation of these parameters, which is rather debatable – see Karsilar et al. (2014). Speed Accuracy Trade-off Under Response Deadlines. Frontiers in Neuroscience, 8:248. doi3: 10.3389/fnins.2014.00248). Thus, for DDM to inform us regarding the source of the observed effects, one needs a blocked design where the viscosity levels are kept constant within a given block. What the current results tell me is the following; Yes, the effect of movement on temporal judgments can be mediated by the perceptual effects. What they do not tell me is the following; Movement does not affect the other components of the decision process (for which authors need a different experimental design). I recommend authors to look at Bogacz et al. (2006). The physics of optimal decision making: a formal analysis of models of performance in two-alternative forced-choice tasks. Psychological Review, 113(4), 700-765. https://doi.org/10.1037/0033-295X.113.4.700 for relevant discussions (look at the sections on the optimal performance).

But there is one way out based on the sequential nature of the diffusion process in the temporal bisection task. At least as it is formulated in Balcı and Simen (2014), which authors cite in the paper, the decision process that would be captured by conventional DDM fits is the second stage diffusion process the parameters of which are contingent on the first stage diffusion (opponent Poisson DDM) process. Within this framework, authors can argue that the viscosity experienced during the movement during the timing of the test stimulus (prior to the offset of the stimulus) can actually modulate the relevant parameters of the second stage diffusion before its launching. If authors are willing to take this route, then they should be mechanistically savvier about such contingencies; otherwise, the results and assumptions are not tied to a concrete process and thus lack the necessary theoretic grounding (as they currently are in the paper). I recommend authors to let the model tie things together, which will bolster the theoretical cohesion of the work.

I have concerns regarding the use of latent parameter estimates from the HDDM (or slopes estimates gathered from their regression on a factor) in the correlation analysis. As the authors state in the Methods section, HDDM uses the mean overall parameters to constrain the estimates for individual subjects; this is a great analytical tool to increase the statistical power for between-group comparisons but at the same time it makes the estimates for individual subjects unsuitable for correlation analysis. If possible given the number of data points, authors should fit the individual data for estimates to be used in the correlation analyses.

I have another concern regarding the theoretical handling of the application of DDM to the data. Using DDM in the temporal bisection task actually requires assuming a sequential diffusion process (see Comment 1). This has been previously formulated in Balci and Simen, (2014). Fitting conventional DDM is just an approximate method to capture the behavioral output of the sequential process. This sequential process requires yoking the drift-rate and starting point during model fits or testing the relationship between them to validate the model in the context of the dataset at hand. Based on this model, if one assumes that viscosity of the movement affects the perceptual/timing process (as hypothesized in this work) then one would not only expect a relationship between viscosity and the drift rate but also a relation between viscosity and the starting point. This issue should be addressed in the paper.

5. The quality of DDM fits are not shown in the paper. The quality of the fits should be certainly demonstrated for both models. Please provide an example of the actual and predicted DDM psychometric and chronometric functions.

6. The fact that the model parameters (DDM and Bayesian) show a non-linear relation with viscosity does not imply that they could also have a role in explaining the behavior. A downstream non-linear reader could produce the observed behavioral effects of viscosity. Since the changes in DDM threshold and bias as a function of viscosity are nonlinear, it would be interesting to correlate the individual slope of BP x viscosity with the negative coefficient on an exponential or power function fitted to these parameters vs viscosity.

7. The authors suggest that the DDM results support the notion that the shift in BP produced by larger viscosities is due to a decrease in drift rate, which in turn produced changes in the perceived duration rather than decision parameters. However, the authors also suggest that time perception and decision making are intermingled in the bisection task and that once the categorical boundary has been crossed, subjects stop accumulating temporal information (Wiener and Thompson, 2015). How do you reconcile the two results?

8. Although I appreciate the Bayesian Observer Model (as I indicated at the very beginning of my review), I think this is not empirically well-justified given the lack of an effect on the level of CV. Do authors assume that the uncertainty characteristics of the representation itself and reproduction (motor component) are independent enough (I see that they are not) but even in that case, the problem continues since the motor output would inevitably inherits noise that results from the representational uncertainty. Consequently, I think that the motivation behind this cool analysis, with contemporary relevance to the interval timing literature, should be buttressed with stronger empirical or theoretical justifications (of course, while making sure of not being over-speculative).

9. It would be interesting if the authors could comment upon the precise mechanisms via which each of these features may shape duration perception. Presumably an influence of movement duration on duration perception is easier to explain mechanistically than an influence of distance per se? Why would distance be thought to provide information concerning auditory duration, independently from the distance-duration correlation? E.g., by my understanding size-duration relationships often observed are thought to be mediated by statistical relationships in natural environments, but such that independent information of magnitude would not add to that provided by duration (e.g., Vincent Walsh, TICS, 2003).

10. Authors talk about the self-regulation of movement vs. induced constraints on the movement and at times they allude to the embodied cognition. This brings a relevant issue to my mind; within the semi-theoretical stance taken in the paper what would the predictions be for the effect of observed biological vs. non-biological movement on time perception and how do such predictions compare to the results reported in the literature? This could be one way in which the paper can set a stronger theoretical foundation and consolidate some of the other findings in the literature under an overarching theoretical framework in light of the current findings reported in the paper. I see this as a great opportunity for the authors/paper.

Reviewer #1:

This study investigates the effects of viscous movement environments on temporal processing during a single interval categorization or a reproduction task using durations in the range of seconds. Healthy volunteers timed auditory intervals while moving a robotic arm that randomly applied four levels of viscosity. The results showed a consistent decrease in the bisection point (or point of subjective equality) in the categorization task and a decrease in duration of the produced intervals in the reproduction task as a function of viscosity. A drift-diffusion model and a Bayesian observer model suggested that the underestimation of intervals in the two tasks are due to changes in time perception instead of biases in decision making.

This is an interesting and well performed study, providing novel notion on the interaction between time perception and reaching movements affected by viscous environments. Nevertheless, I have a set of serious concerns that need to be addressed in the present version of the paper, as follows:

1. The values on bisection point show enormous individual differences between subjects, with a range between 1.5 to 3 seconds, why? In contrast, the effect of viscosity on the BP is very subtle and a large proportion of subject showed negative slope of BP as a function of viscosity (please report). This means that in a subgroup of subjects, an increase in viscosity was accompanied with an increase instead of a decrease in the length of perceived time. This finding does not support one of the mayor conclusions of the paper.

2. The fact that the model parameters (DDM and Bayesian) show a non-linear relation with viscosity does not imply that they could also have a role in explaining the behavior. A downstream non-linear reader could produce the observed behavioral effects of viscosity. Since the changes in DDM threshold and bias as a function of viscosity are nonlinear, it would be interesting to correlate the individual slope of BP x viscosity with the negative coefficient on an exponential or power function fitted to these parameters vs viscosity.

3. The authors suggest that the DDM results support the notion that the shift in BP produced by larger viscosities is due to a decrease in drift rate, which in turn produced changes in the perceived duration rather than decision parameters. However, the authors also suggest that time perception and decision making are intermingled in the bisection task and that once the categorical boundary has been crossed, subjects stop accumulating temporal information (Wiener and Thompson, 2015). How do you reconcile the two results?

4. It is peculiar that the effect of viscosity was on the produced rather than the measured noise term of the Bayesian model, since the viscosity changes were only present during the interval encoding. Please discuss.

5. The drift rate can be associated with the likelihood function of the measured time of the Bayesian model, whereas the starting point (z) can be linked to the offset shift (b). However, the results showed that viscosity affects the drift rate in the bisection task, whereas affects the offset in the reproduction task. Please discuss why.

Reviewer #2:

The current paper by De Kock et al. investigates the effect of movement on time perception. The novelty of the work is that it considered the effect of externally modulated movements (through manipulating viscosity) instead of self-modulated movements (as done in earlier research) as well as the use of generative and descriptive models to elucidate the mechanisms that underlie such effects. Participants were tested in the temporal bisection (Experiment 1) and temporal reproduction tasks (Experiment 2 – presumably to eliminate the decision component). The consistent result was that higher viscosity primarily resulted in the compression of perceived time (in terms of the rightward shifts in the PSE in Experiment 1 and weaker over reproduction in Experiment 2). The results of DDM fits in Experiment 1 showed that these effects were due to the modulation of the perceptual rather than the decision components. This conclusion was further supported by the effect of viscosity on the perceptual bias parameter (i.e., reduction) of the Bayesian Observer Model fits in Experiment 2. The experiments conducted in this study are certainly interesting and offer a number of novelties; I think that the paper is potentially impactful. That being said, I also have a number of concerns about the work, which I specify below.

1. Based on DDM fits authors claim that the effect of viscosity was perceptual (and not through decision-biases). This claim is based on the observation that the viscosity level did not affect the starting point or threshold setting but it exclusively modulated the drift rate. Although on the face value, this appears as a valid conclusion based on the model fit results, the methodological details actually put this conclusion into question by invalidating the model fits (unless participants could predict the next viscosity level – see alternative explanations below). Let me elaborate more on this. The standard assumption in the conventional DDM that authors also used in their analysis is that the threshold is set adaptively in response to the context the participant is in (e.g., taking into account the response-to-stimulus interval, reward/penalty ratio, whether instructions emphasize speed or accuracy, prior probability of different alternatives). In other words, the decision threshold and starting point of the decision process are modulated in response to some average representation of the task statistics or in response to a discriminative stimulus that allows participants to predict the characteristics of the next trial (e.g., a trial with vs. without penalty for errors or trial in which the viscosity will be higher vs. lower). But in the current study, the viscosity levels were manipulated randomly from one trial to the next (without any discriminative signaling) and thus participants could not have adjusted the decision thresholds or starting points (unless authors are willing to step out of the conventional DDM domain in which case they have to assume within-trial dynamic modulation of these parameters, which is rather debatable – see Karsilar et al. (2014). Speed Accuracy Trade-off Under Response Deadlines. Frontiers in Neuroscience, 8:248. doi3: 10.3389/fnins.2014.00248). Thus, for DDM to inform us regarding the source of the observed effects, one needs a blocked design where the viscosity levels are kept constant within a given block. What the current results tell me is the following; Yes, the effect of movement on temporal judgments can be mediated by the perceptual effects. What they do not tell me is the following; Movement does not affect the other components of the decision process (for which authors need a different experimental design). I recommend authors to look at Bogacz et al. (2006). The physics of optimal decision making: a formal analysis of models of performance in two-alternative forced-choice tasks. Psychological Review, 113(4), 700-765. https://doi.org/10.1037/0033-295X.113.4.700 for relevant discussions (look at the sections on the optimal performance).

But there is one way out based on the sequential nature of the diffusion process in the temporal bisection task. At least as it is formulated in Balcı and Simen (2014), which authors cite in the paper, the decision process that would be captured by conventional DDM fits is the second stage diffusion process the parameters of which are contingent on the first stage diffusion (opponent Poisson DDM) process. Within this framework, authors can argue that the viscosity experienced during the movement during the timing of the test stimulus (prior to the offset of the stimulus) can actually modulate the relevant parameters of the second stage diffusion before its launching. If authors are willing to take this route, then they should be mechanistically savvier about such contingencies; otherwise, the results and assumptions are not tied to a concrete process and thus lack the necessary theoretic grounding (as they currently are in the paper). I recommend authors to let the model tie things together, which will bolster the theoretical cohesion of the work.

2. I have concerns regarding the use of latent parameter estimates from the HDDM (or slopes estimates gathered from their regression on a factor) in the correlation analysis. As the authors state in the Methods section, HDDM uses the mean overall parameters to constrain the estimates for individual subjects; this is a great analytical tool to increase the statistical power for between-group comparisons but at the same time it makes the estimates for individual subjects unsuitable for correlation analysis. If possible given the number of data points, authors should fit the individual data for estimates to be used in the correlation analyses.

3. I have another concern regarding the theoretical handling of the application of DDM to the data. Using DDM in the temporal bisection task actually requires assuming a sequential diffusion process (see Comment 1). This has been previously formulated in Balci and Simen, (2014). Fitting conventional DDM is just an approximate method to capture the behavioral output of the sequential process. This sequential process requires yoking the drift-rate and starting point during model fits or testing the relationship between them to validate the model in the context of the dataset at hand. Based on this model, if one assumes that viscosity of the movement affects the perceptual/timing process (as hypothesized in this work) then one would not only expect a relationship between viscosity and the drift rate but also a relation between viscosity and the starting point. This issue should be addressed in the paper.

4. The quality of DDM fits are not shown in the paper. The quality of the fits should be certainly demonstrated.

5. Authors talk about the self-regulation of movement vs. induced constraints on the movement and at times they allude to the embodied cognition. This brings a relevant issue to my mind; within the semi-theoretical stance taken in the paper what would the predictions be for the effect of observed biological vs. non-biological movement on time perception and how do such predictions compare to the results reported in the literature? This could be one way in which the paper can set a stronger theoretical foundation and consolidate some of the other findings in the literature under an overarching theoretical framework in light of the current findings reported in the paper. I see this as a great opportunity for the authors/paper.

6. I found that the paper could be a bit more inclusive in terms of the previous work on the relationship between movement and time perception. As it stands, the paper does not fully "exploit" the relevant literature to set the stage for the reader.

7. "Entry into the response location prior to the tone offset was penalized by restarting the trial, and so the optimal strategy was to move the cursor closer to the "short" location, and then gradually move to the "long" location as the tone elapses (Wiener et al., 2019)." This happens to be an emergent action pattern in these tasks; do authors think that differential reinforcement was necessary for the emergence of this strategy? It would also be interesting (but not necessary) to look at the parametrization this action pattern in relation to the optimization of temporal decision making under uncertainty.

8. There are several tests in the paper, where the null effects are theoretically important. Those frequentist tests should be coupled with Bayesian counterparts.

9. Figure 1b (left side): One thing I don't understand here is how the hand can reach to the same location for all test durations prior to the termination of test duration. Am I misinterpreting the figure? E.g., Maybe different sigmoidal curves do not correspond to different test durations.

10. Figure 4b bottom panel should be better explained (e.g., explain what slope refers to in the figure caption).

11. Authors should better justify the inclusion of certain parameters in the model (e.g., T_naught). On a related note, the proper model comparisons should be exercised; for instance, models with different levels of complexity (theoretically guided) should be compared and the proper model comparison statistics should be interpreted and reported (e.g., DIC).

12. What would empirically complement Experiment 2 is exposing participants to different levels of viscosities during the reproduction, in which case the predictions regarding the direction of the effect on temporal reproduction would be the opposite. I understand that the effect of viscosity on the speed of movement could be a concern here but when the task is reproducing an interval, I do not think this is a big issue.

13. Although I appreciate the Bayesian Observer Model (as I indicated at the very beginning of my review), I think this is not empirically well-justified given the lack of an effect on the level of CV. Do authors assume that the uncertainty characteristics of the representation itself and reproduction (motor component) are independent enough (I see that they are not) but even in that case, the problem continues since the motor output would inevitably inherits noise that results from the representational uncertainty. Consequently, I think that the motivation behind this cool analysis, with contemporary relevance to the interval timing literature, should be buttressed with stronger empirical or theoretical justifications (of course, while making sure of not being over-speculative).

14. Authors should specify whether the between-trial variability in the core parameters of the DDM was included in the model (as in extended DDM) or not (as in pure DDM).

15. "Analysis of RT values demonstrated faster RTs with longer perceived duration, consistent with previous reports." Please contextualize this finding better in relation to earlier findings and explain the presumed reasons behind it. Authors also do not really elaborate on the finding that the RTs were shortest for the middle viscosity. Is it possible that this finding is also a result of a range effect/Bayesian optimization, where the expectancy was highest for the middle viscosity level?

16. "For the CV, a significant interaction between viscosity and movement length was observed [F(1,27)=7.694, p=0.01, 2p=0.222], in which the CV was significantly lower when subjects moved more, but only when the viscosity was zero [t(27)=-2.237, p=0.034, D=1.2] (Figure 3b), and not for any other viscosity (all p>0.05)." Instead of using a median-split method for the movement length here, I recommend the authors to use a linear mixed-effects model to analyze their data.

I tried to keep my comments as concise as possible not to overwhelm the authors of this fine work but I think that the issues I raised are important on many fronts. I tried to guide authors as much as possible on how they can tackle some of my comments but admittedly they are not necessarily the best ways to address them (e.g. using a blocked design vs. assuming that the thresholds can be set during the timing stimulus in my first point).

Reviewer #3:

The authors required participants to move a robotic arm in environments of varying viscosity. They found that when moving the lever in the presence of high viscosity their estimates of durations of auditory events were compressed. They conducted drift diffusion modelling and two different procedure types to indicate that this was likely due to an influence of action on duration perception, rather than decisional processes. I found the studies interesting but I was unsure how this shaped our understanding of underlying mechanisms, which I think needs clarifying.

1. As the authors outline, previous studies have shown that a variety of movement features impact upon duration judgements. The novelty of their study is suggested to be that movement is restricted by environmental factors rather than 'self' modulations. E.g., the finding reflected in the title of "slowing the body slows down time" has already been demonstrated, by my understanding, in four experiments by Yon et al. (2017, JEPG). However, by my reading both in previous studies and the present studies in fact it is environmental modulations that result in self modulations. I.e., I did not comprehend the difference and this needs clarifying for the theoretical value of these studies to be understood. E.g., previous studies have generated the cells in their design by presenting targets at different locations (e.g., Yon et al., 2017 – external modulation) and participants then choose to move to those targets (volitional component). The present studies create the cells by imposing different forces upon the lever while keeping the targets in the same location. In all studies they are choosing to move to targets at different places and under specific environmental viscosities. The authors describe using sensory feedback due to these environmental features but they would use sensory feedback to monitor both visual target location and forcefields – both of which are constant during a specific movement in previous and present work. Once this line is clarified, what would be the assumed different mechanisms involved in the two cases?

2. In the present studies a number of features of the movement are confounded as viscosity is manipulated. As viscosity increases, the length traversed goes down in a given period of time, the force goes up, the speed goes down. As far as I could see the authors do not disentangle which movement feature was responsible for the influence on duration judgements. My guess is that the correlations are high between these features but perhaps there is enough natural variation in the dataset to disentangle. Could a regression analysis be performed to establish the unique contributions of the different movement parameters? As well as helping the reader to understand what is driving the effects it may resolve some of the inconsistencies with respect to previous literature. E.g., the authors observe that fast movements are associated with increased duration judgements, in contrast with some previous findings. They explain the difference as possibly driven by the fact the conflicting studies have unfilled intervals. However, they do not explain why this may alter the operation of the mechanism. Additionally, there is an experiment in Yon et al. (2017) where high speed is associated with compressed estimates and the intervals are filled here.

3. It would be interesting if the authors could comment upon the precise mechanisms via which each of these features may shape duration perception. Presumably an influence of movement duration on duration perception is easier to explain mechanistically than an influence of distance per se? Why would distance be thought to provide information concerning auditory duration, independently from the distance-duration correlation? E.g., by my understanding size-duration relationships often observed are thought to be mediated by statistical relationships in natural environments, but such that independent information of magnitude would not add to that provided by duration (e.g., Vincent Walsh, TICS, 2003).