Bayesian multilevel modeling of group and participant level effect sizes: Stronger inter-individual differences in pupil dilation compared to EEG alpha power during aversive conditioning

Abstract

Aversive conditioning prompts reliable changes in the power of EEG alpha-band oscillations and pupil dilation. Both variables have been used to test hypotheses on the acquisition, generalization, and extinction of conditioned threat. Existing studies have largely relied on trial averages and group-level analyses. Thus, the variability of these physiological markers to aversive learning at the subject level is currently unknown. Comparisons of group-level analyses in prior studies suggest that pupil dilation and EEG-alpha activity capture complementary information. However, to date, no study has directly compared these two markers in terms of their effect sizes at the level of individual participants. The present study employed Bayesian multilevel modeling to quantify the variability of conditioning effect estimates for alpha-band power and pupillometry. Estimates were examined at the group level and at the participant level, across two conditioning paradigms, involving visual and auditory cues. Although the two metrics shared similar effect sizes at the group level, participant-level variability in these effect sizes was substantially higher for pupil-dilation compared to alpha-power, and this finding was replicated across both paradigms. These findings have important implications for clinical and inter-individual difference research which requires both the quantification of effects at the participant-level as well as meaningful variability between-participants that can be linked to relevant differences such as anxiety.