Dear Vishal,

Those are all good questions, although the relevance (safety?) is not apparent. In any case, neither my nor anyone else's speculation is worth more than yours without additional experiments. One thing that I will note (as you have), is that correlation does NOT imply causation. So, experiments would need to be devise that more specifically isolate and modify the suspected cause. As a controlled reflex system, I do not believe that you will find that eye pupil dilation is a cause of anything. Rather, it is a response to other inputs, which may be external (light intensity, field of vision, focus, etc.) or internal (stress, drugs, and the like).

As a former professional driver and career risk management expert I have an interest in the effect of many environmental factors on alertness, awareness, identification and response times. As such, I would be interested in the expected implications of pupil dilation for any of those things.

Best regards,

Steven

The accommodation reflex input to pupil diameter usually requires an accommodation/vergence response to REAL changes in viewing distance. I presume that your near/far targets in the simulation were presented on the same flat screen and, hence, the same working distance for accommodation/vergence. Thus, unlikely to introduce changes in accommodation. However, I do seem to recall that accommodation-in-the-dark tends to increase in a smaller room compared to a larger room (i.e., top-down influences on accommodation are possible).

I suspect that your unusual finding of decreasing pupil diameter with increasing workload demand is related to the constricted range of eye movement behavior with increased task demands (Which we have also documented in real-world driving studies). Perhaps the central part of the screen had higher luminance stimuli than the near-periphery. Thus: workload increases...tunnel vision ensues...luminance adaptation state increases...tonic pupil diameter decreases.

Of course, this is mere speculation.

Frank Schieber

University of South Dakota

What I am writing partly repeats Frank's comment, but perhaps different formulations of the same idea can be helpful as well.

The first question is whether the construction of of your driving simulator is such that it enforces accommodation changes. The simulator might include a model dashbord at arm's length and a projection screen for the driving environment that is farther away, so that the subjects' gaze is alternating between near and far targets and perhaps is more often directed to near targets during the more demanding tasks. In this case accommodation could explain the seemingly paradox behaviour of the pupil.

Otherwise, if all vision targets are at the same distance, it might be that the tasks with the higher mental workload require the gaze to linger on brighter areas of the display for a longer time, which could again explain a decrease in pupil size. All this has in principle been said above.

But there is still some other possibility. It might be the case that your pupil measuring equipment has a gaze dependent bias. I just saw an article which mentions that "pupil size is underestimated for leftward and downward gaze, and overestimated for rightward and upward gaze when using an EyeLink© 1000 eye-tracker; the opposite result is obtained when using Tobii© (T120 and X120) eye-trackers. (https://www.sciencedirect.com/science/article/pii/S0960982214001961#bib7)

Hope you can find the correct explanation

Detlef

Thank you all for the answers, very useful insights. Please find a few clarifications:

> The projectors screens of simulator environment, the instrument cluster, and second-task tablet were not on the same view plane.

> Most lead vehicle colors were kept consistent throughout all scenarios (white, light blue, light brown, occasional red/green/blue) . Pupillary changes due to color bias can be assumed to be not that significant. Also, the same participants drove all levels of Task demand and this should further eliminate the bias.

> I have attached a few figures to show the overall trends in a more descriptive manner. (also you might notice that all scales are from 0-1 as I normalized individual participants values using task difficulty). Out of the 6 task demand levels, which ever an individual found to be the easiest was deemed 0 and the most difficult as 1. I'm assuming we all agree on the definitions of task demand (set by experimenter: usually changes in known increments) vs task difficulty (experienced by the driver).

> I know that pupils are to dilate as tasks get more difficult. However, these are also for short periods of time (events) and not driving for prolonged periods. typical literature involves cognitve/visual load for short bursts.

> I also understand that correlation does not indicate causation. But, the really high correlation does suggest something if < 0.005. A bivariate Pearson Correlation test between average right pupil diameter and SART resulted in a value of 0.953 (p = 0.003 ).

> Keep in mind that my values were averages for the entire 90 participants but on an individual level, I can see a few trends in the time series plot (I can post them).

> I find the trend boiling down to the amount of compensation happening. At high task demands, participants compensate fairly quick by slowing down and increasing gaps, and this in turn causes moments of large pupils and then small pupils due to less WL as a result of increased gaps and decreased speeds. I can identify locations of compensation but it is interesting to see that the overall trend in decreasing.

> I am still working on finding surrogate measures through blinks and gaze patterns.

Thank you again for your guidance!

Regards,

Vishal

Dear Joost,

Thank you! I will definitely take the luminance into account and try other surrogate measures for now. I will treat the pupil diameter results more carefully during analysis. Also, i used a chest strap for the heart rate rather than a more accurate technique for variability (ECG/EKG), so i'm a bit skeptical about the values.

Kind regards