3.1. Behavioral Measures
Of the 64 possible test trials of face morphs used to determine perceptual biases in judging emotion, participants completed an average of 57.49 (SD = 7.01) baseline test trials and 60.21 (SD = 5.43) post-adaptation test trials.
In order to assess the goodness-of-fit of the psychometric function to our data, we considered measures of deviance, quantified using psignifit, for PSE measures at baseline and post-adaptation. On average, the deviance at baseline was 5.50 (SD = 2.82) and post-adaptation was 5.33 (SD = 2.89).
An ANOVA, Bonferroni corrected to account for multiple comparisons (alpha = 0.0083), was run to test the hypothesis that the strength of changes in perception varied across adaptation conditions. We expected the largest perceptual change, shift in PSE, for congruent adaptation, an intermediate effect for only visual adaptation, relatively weaker changes for incongruent adaptation, and the weakest change for only auditory adaptation.
We found a significant main effect of adaptation condition on PSE shift after adapting to negative emotions (F
(3,77) = 9.080, p
< 0.001, partial η2
= 0.261; see Figure 3
, with a significantly more positive PSE shift for Ac, Ai, and Av compared to Aa. This indicated that the mean neutral face appeared happier post-adaptation for Ac, Ai, and Av relative to Aa (Ac: t
(40) = 3.584, p
< 0.001; Av: t
(35) = 4.716, p
< 0.001; Ai: t
(38) = 3.657, p
= 0.001). No significant differences were found between other conditions (Ac versus Av: t
(39) = −0.964, p
= 1; Av versus Ai: t(37): 1.183, p
= 1; Ac versus Ai: t(42) = 0.100, p
= 1). One sample t
-test indicated all conditions except Aa showed significant adaptation effects; PSE shifts were significantly different from baseline (Ac: t
(22) = 5.185, p
< 0.001; Av: t
(17) = 8.833, p
< 0.001; Ai: t
(20) = 5.674, p
< 0.001; Aa: t
(18) = −0.384, p
We also examined perceptual shifts with a Bayesian ANOVA in JASP (JASP Team, 2018), using a Cauchy prior distribution with r = 1/sqrt(2). The Bayes factor (BF10) of 654.274 suggests that the data were approximately 650 times more likely to occur under the alternative hypothesis than under the null (suggesting extreme evidence), with an error percentage of 0.011%. This indicates that PSE shift differs as a function of adaptation condition. Post-hoc tests corrected for multiple testing by fixing the prior to 0.5 suggest a similar pattern of results to the ANOVA above: more positive PSE shifts for Ac, Ai, and Av compared to Aa (Ac: posterior odds = 13.946, Ai: posterior odds = 16.031, Av: posterior odds = 206.343) and no differences between other conditions (Ac versus Av: posterior odds = 0.184; Av versus Ai: posterior odds = 0.224; Ac versus Ai: posterior odds = 0.124). Bayesian one sample t-tests indicated all conditions, except Aa showed an adaptation effect (Ac: BF10 = 724.4, error < 0.001%; Av: BF10 = 157,852, error < 0.001%; Ai: BF10 = 1531, error < 0.001%; Aa: BF10 = 0.254, error < 0.05%).
Although PSE shifts did not differ across congruent, incongruent, and visual only conditions, another quantification of our data, slope, might differ, with steeper slopes indicative of less variance in perceptual data [44
]. However, we found no significant differences in slope changes across adaptation conditions (p
= 0.919; data not shown).
3.2. Physiological Measures
An ANOVA, Bonferroni corrected to account for multiple comparisons (alpha = 0.0083), was run to test the hypothesis that the strength of changes in cortisol varied across adaptation conditions. We expected the largest cortisol change for congruent adaptation, an intermediate change for only visual adaptation, relatively weaker changes for incongruent adaptation, and the weakest change for only auditory adaptation. Furthermore, we expected relatively weak increases in cortisol and mostly effects on relative differences in decreases in cortisol.
We found no significant main effect of adaptation condition on cortisol shift after exposure to negative emotions (F
(3,58) = 1.618, p
= 0.195, partial η2
= 0.077; see Figure 4
. One sample t
-test indicated cortisol changes were significantly different from baseline, after exposure to negative emotions, except for condition Aa (Ac: t
(12) = −2.562, p
= 0.025; Av: t
(16) = −3.184, p
= 0.006; Ai: t
(13) = −5.224, p
< 0.001; Aa: t
(17) = −2.027, p
We also examined physiological shifts with a Bayesian ANOVA in JASP (JASP Team, 2018), using a Cauchy prior distribution with r = 1/sqrt(2). The Bayes factor (BF01) of 2.314 suggests that the data were approximately 2.3 times more likely to occur under the null hypothesis than under the alternative, with an error percentage of < 0.001%. This indicates that the strength of cortisol shifts did not vary under different adaptation conditions. Bayesian one sample t-tests indicated cortisol changes were more likely to occur under the alternative relative to the null hypothesis for all conditions, with varying degrees of evidence (Ac: BF10 = 2.779, error < 0.005%; Av: BF10 = 8.505, error < 0.001%; Ai: BF10 = 186.5, error < 0.001%; Aa: BF10 = 1.276, error < 0.01%).
Changes in cortisol levels are shown for each adaptation condition. The x-axis depicts the adaptation condition and the y-axis depicts the mean change in cortisol, normalized by baseline (+/− SEM across participants), with data from individual participants shown via open circles. A mean shift in the negative direction indicates that cortisol decreased post-adapt relative to pre-adapt. Conversely, a mean shift in the positive direction indicates that cortisol increased post-adapt relative to pre-adapt. There were no significant differences in cortisol shift across conditions and mean cortisol shifts tended to be negative.
3.3. Correlations between Behavioral and Physiological Measures
Given the high variability in our data, considering only differences in means across participants might obscure important relationships arising from individual differences. Thus, we tested whether shifts in the perception of emotion (PSE shift) correlated with shifts in cortisol, using a Pearson correlation between PSE and cortisol shifts (see Figure 5
). We found a significant negative correlation between PSE shifts and cortisol shifts across adapt conditions (r
= −0.303, p
= 0.017). This negative correlation suggests that, following exposure to negative emotions, as the neutral face was judged more positive, post-adapt cortisol levels were more negative relative to baseline cortisol levels, possibly due to lower stress, lower arousal, or less attention. The same correlation run with Bayesian statistics yielded a BF10
of 2.596, suggestive of anecdotal evidence for a relationship between shifts in cortisol and shifts in perception.
3.4. Underlying Biases in Behavioral and Physiological Measures
All our measures quantifying perceptual and cortisol changes were normalized to starting baseline values. We examined perceptual and cortisol measures at baseline, to determine if baseline biases could influence the effects of interest. We found no significant main effect of adaptation condition on perception at baseline nor on cortisol at baseline (perception: F(3,77) = 0.803, p = 0.496, partial η2 = 0.030; cortisol: F(3,58) = 0.506, p = 0.680, partial η2 = 0.026; data not shown) and no significant correlations between baseline cortisol and baseline PSE (r = 0.0180; p = 0.161; data not shown), or baseline PSE and baseline state affect (PA: (r = 0.196; p = 0.192); NA: (r = 0.055; p = 0.647; data not shown)).
Of note, while not a main measure, as state affect was only assessed before but not after adaptation, we found a significant negative correlation between cortisol and positive affect at baseline (r = −0.237, p = 0.037; data not shown), such that more elevated cortisol at baseline was associated with less positive affect at baseline. No significant correlation was found between cortisol and negative affect at baseline (r = −0.117, p = 0.369; data not shown).