'Investigating pupil-linked arousal to uncertainties in auditory patterns' (AsPredicted #94511)
Author(s) Hamit Basgol (Department of Computer Science, University of Tübingen) - hamit.basgoel@uni-tuebingen.de Peter Dayan (Max Planck Institute for Biological Cybernetics) - dayan@tue.mpg.de Volker Franz (University of Tübingen) - volker.franz@uni-tuebingen.de
Pre-registered on 2022/04/18 - 11:46 PM (PT)
1) Have any data been collected for this study already? No, no data have been collected for this study yet.
2) What's the main question being asked or hypothesis being tested in this study? Humans track the statistical structure of their environments and build mental models to cope with the uncertainties in the world. An observer can encounter two types of uncertainties, namely expected and unexpected uncertainties. Expected uncertainties occur because of the known unreliability of or ignorance about predictive relationships in a model, and unexpected uncertainties are above and beyond the capabilities of the current model, requiring a model change (Dayan & Yu, 2003; Dayan & Yu, 2006; Payzan-LeNestour et al., 2013; Yu & Dayan, 2005; Zhao et al., 2019). While expected uncertainty correlates with Acetylcholine (ACh), unexpected uncertainty correlates with norepinephrine (NE) (Dayan & Yu, 2003; Marshall et al., 2016; Yu & Dayan, 2005). NE is produced by Locus Coeruleus (LC) in the brainstem and has a mechanistic correlation with pupil dilation response (PDR) (Gilzenrat et al., 2010; Kalwani et al., 2014; Murphy et al., 2011; Reimer et al., 2016; Zhao et al., 2019). Because of this correlation, recent research investigated the response of LC-NE to expected and unexpected uncertainties (Zhao et al., 2019). In this study, we plan to replicate the results of Zhao et al. (2019) by measuring pupil responses (PDRs) to RAND-REG and REG-RAND auditory transitions that simulate expected and unexpected uncertainties, respectively. We will use REG and RAND auditory signals that do not involve transitions for comparing pupil measures received from RAND-REG and REG-RAND. As a novel contribution, we plan to pinpoint model violation, construction, and refinement by REGa-REGb auditory signals involving a transition between regularity patterns. We expect a PDR response to transitions in REG-RAND and REGa-REGb (Hypothesis 1), marking LC-NE activation in model violation, and a pupil constriction response in REGa-REGb (Hypothesis 2), suggesting model construction and refinement. We do not expect a substantial PDR response to RAND-REG, REG, and RAND auditory signals (Hypothesis 3).
3) Describe the key dependent variable(s) specifying how they will be measured. 1- Throughout the experiment, participants will be instructed to find locations of silent gaps in auditory sequences as accurately as possible. This task is orthogonal to transitions between different auditory patterns and is for keeping attentions of participants on ongoing auditory sequences. Silent gaps will involve three missing tones for RAND and two missing tones for REG signals to equate task difficulty, as in Zhao et al. (2019). We will receive button responses and their RTs.
2- Throughout the experiment, we will measure the pupil diameters of participants and their gaze locations by the Eyelink 1000.
4) How many and which conditions will participants be assigned to? There will be five conditions in the experiment. Each condition will be generated by combining short tone pips lasting 50 ms according to a regular or a random pattern.
RAND10: 10 tones selected from a tone pool will be repeated and shuffled.
REG10: 10 tones selected from the tone pool will be repeated to generate a regular signal
RAND10-REG10: 10 tones selected from the tone pool will be repeated. They will be shuffled for the first part of the trial, and they will be repeated to generate a regular signal for the second part of the trial.
REG10-RAND10: It is the same as RAND10-REG10 but in inverse order.
REG10a-REG10b: 10 tones will be randomly selected from the tone pool. Selected tones will be repeated for generating the first regularity. Then, the selected ten tones will be shuffled and repeated for generating the second regularity.
The number of trials determined for each condition is given below. In this way, the number of trials that have a transition and do not have a transition, and the number of types of transitions (i.e., transitions leading to expected and unexpected uncertainties) are matched.
RAND10 (Baseline): 48 trials (EXP1B in Zhao et al., 2019: 24 trials)
REG10 (Baseline): 48 trials (EXP1B in Zhao et al., 2019: 24 trials)
RAND10-REG10 (Expected Uncertainty): 48 trials (EXP1B in Zhao et al., 2019: 24 trials)
REG10-RAND10 (Unexpected Uncertainty): 24 trials (EXP1B in Zhao et al., 2019: 24 trials)
REG10a-REG10b (Unexpected Uncertainty): 24 trials (Zhao et al., 2019: condition was not measured)
Each participant will see 192 trials in total. The probability of a gap for each trial will be 20%, and each condition will involve the same number of gaps. As in the previous study, gaps can occur at any time between 250 ms post-onset and 750 ms pre-offset. The length of auditory sequences will be varied randomly between 6 and 8s, and transitions will be jittered between 3 and 4s after the sequence onset.
5) Specify exactly which analyses you will conduct to examine the main question/hypothesis. We use the same analyses as Zhao et al. (2019). Our analyses can be separated into two groups: analyses for the PDRs analysis and analyses for the gap detection performance.
1. For the PDRs analysis, the following steps will be applied in order:
Epoching: Pupil size data measured during trials will be epoched (i.e., segmented) and further analyses will be conducted on the epoched data. The data measured for RAND10 and REG10 conditions will be epoched by considering 1 second before sequence onset and 7 seconds after sequence offset. On the other hand, the data measured for RAND10-REG10, REG10-RAND10, and REG10a-REG10b conditions will be epoched according to the timing of transitions during trials, namely by considering 1 second before and 3 seconds after transitions.
Blink construction: Complete or partial blinks detected by the Eyelink 1000 will be interpolated by cubic interpolation.
Smoothing: The data will be smoothed by a 150 ms Hanning window.
Normalization: After smoothing, each participant's data will be z normalized by the mean and the standard deviation computed for the data of each block (for each participant, the mean and the standard deviation will be calculated separately).
Baseline correction: A linear regression-based baseline correction will be applied to regress the effect of transitions on pupil size. For each participant and condition, 1 second before transition intervals will be a baseline for each trial. Then, the impact of baselines on pupil responses will be regressed out by linear regression for each time point by accepting residuals. For conditions that do not have a transition (REG10, RAND10), dummy transition points will be determined by transition points in other conditions (REG10-RAND10, RAND10-REG10, REG10a-REG10b).
Data aggregation: After the baseline correction, all data will be aggregated based on individual conditions, and therefore, time-series data will be computed for each condition.
Statistical analyses: A series of two-tailed t-tests will be conducted in a pair-wise manner over the entire epoch after down sampling to 20 Hz to compare conditions. The family-wise error rate will be controlled by a non-parametric permutation procedure.
In addition to frequentist statistical tests, we will include Bayes factors to our PDRs analysis schema.
2. Gap detection performance analyses:
Hit and false alarm rates will be calculated based on conditions (RAND10, REG10, RAND10-REG10, REG10-RAND10, REG10a-REG10b) and auditory types (RAND10, REG10) after aggregating performances in different conditions.
6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations. Practice block: The main author of Zhao et al. (2019), S. Zhao, mentioned the involvement of a practice block and gave some details about this phase (personal communication, March 25, 2022). Before each experiment, a practice block, which is the same as the one used in the main experiment, will be conducted, and the performances of participants in the gap detection task will be determined. If participants cannot reach 80% gap detection accuracy in both auditory types (RAND10, REG10), they will have a second chance to do practice. If they cannot pass 80% accuracy on the second occasion, they will not have an opportunity to attend the experiment again.
Behavioral performance in gap detection: Participants whose hit rate for gap detection is three standard deviations below the group mean will be excluded from the analysis.
Blinks: Trials having more than 50% missing data due to blinks or involving missing data after interpolation will be excluded from the analyses.
Gaze locations: Participants whose gazes exceed three standard deviations from the group mean will be excluded from the analyses.
7) How many observations will be collected or what will determine sample size? No need to justify decision, but be precise about exactly how the number will be determined. Zhao et al. (2019) conducted a power analysis based on the results of their initial experiments, namely EXP1A and EXP1B. They estimated the effect size (Cohen's d) to be 0.8 for a within subjects design study with a one-sided t-test and found that N = 12 is enough for an adequately powered study with β = 0.2 and α = 0.05. By using G*Power (Faul et al., 2009), we estimated the required sample size in the same settings with a d of 0.8 and found that N = 20 would increase the power of the study to 1- β = 0.964. In addition, we will have at least as many or more trials than Zhao et al. (2019) (see section 4 for the number of trials in their EXP1B), which increases power in our study even further. Therefore, we expect that N = 20 (after exclusion of participants), will constitute an adequately powered replication.
8) Anything else you would like to pre-register? (e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?) We will explore and control the effect of overlap between regularity patterns in REG10a-REG10b on PDRs.
We may explore individual differences in pupil responses to auditory pattern changes and model construction and refinement processes.
