Affective biases encoded by the central arousal systems dynamically modulate inequality aversion in human interpersonal negotiations
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
Negotiating with others about how finite resources should be distributed is an important aspect of human social life. However, little is known about mechanisms underlying human social-interactive decision-making. Here, we report results from a novel iterative Ultimatum Game (UG) task, in which the proposer’s facial emotions and offer amounts were sampled probabilistically based on the participant’s decisions, creating a gradually evolving social-interactive decision-making environment. Our model-free results confirm the prediction that both the proposer’s facial emotions and the offer amount influence human choice behaviour. These main effects demonstrate that biases in facial emotion recognition also contribute to violations of the Rational Actor model (i.e. all offers should be accepted). Model-based analyses extend these findings, indicating that participants’ decisions are guided by an aversion to inequality in the UG. We highlight that the proposer’s facial responses to participant decisions dynamically modulate how human decision-makers perceive self–other inequality, relaxing its otherwise negative influence on decision values. In iterative games, this cognitive model underlies how offers initially rejected can gradually become more acceptable under increasing affective load, and accurately predicts 86% of participant decisions. Activity of the central arousal systems, assessed by measuring pupil size, encode a key element of this model: proposer’s affective reactions in response to participant decisions. Taken together, our results demonstrate that, under affective load, participants’ aversion to inequality is a malleable cognitive process which is modulated by the activity of the pupil-linked central arousal systems.
Article activity feed
-
###Reviewer #2:
General assessment:
The paper studies how facial expressions of proposers in a repeated ultimatum game affect decisions by responders. The paper makes three main contributions. First, responder's decisions are affected by the facial expressions of proposers. Second, the paper statistically compares the fit of several decision functions (utility functions). In the preferred model, the degree of inequity aversion of the responder depends on the facial expression of the proposer. Third, facial expressions of proposers correlate with pupil dilation of responders. The second contribution is the main contribution of the paper, as the first point has been shown before in many different economic games. I think that the second point - the modeling exercise - is interesting, but should be improved. Moreover, I think the …
###Reviewer #2:
General assessment:
The paper studies how facial expressions of proposers in a repeated ultimatum game affect decisions by responders. The paper makes three main contributions. First, responder's decisions are affected by the facial expressions of proposers. Second, the paper statistically compares the fit of several decision functions (utility functions). In the preferred model, the degree of inequity aversion of the responder depends on the facial expression of the proposer. Third, facial expressions of proposers correlate with pupil dilation of responders. The second contribution is the main contribution of the paper, as the first point has been shown before in many different economic games. I think that the second point - the modeling exercise - is interesting, but should be improved. Moreover, I think the experimental design has some important issues, which seem hard to address without collecting new data.
Substantive concerns:
One of the main selling points of the paper is that it studies iterative/repeated games instead of one-shot interactions. The authors seem to ignore (rule out) repeated game strategies however. This is understandable, given that analyzing the repeated game (with signaling) is complex, and beyond the point of the paper. More importantly, the statistical analysis ignores the dynamic nature of the game. From what I understand, in the analysis all data are pooled, both across participants and trials. Given this, I think the authors overinterpret the model, as the interpretation in the text is often dynamic (for example, on page 10, lines 254-255, but also in several other instances), whereas the statistical analysis is not.
Given that facial expressions affect decision-making, it is no surprise that including facial expressions in the decision values improves the fit. The most interesting part (to me) of the modeling exercise is to determine how facial expressions are best incorporated in the model. The authors organized a kind of 'horse race' between several models to address this. But why select these models? The choice seems ad-hoc and could be better motivated. For example, the best performing model treats positive and negative deviations from neutral faces in the same way, whereas the emotion recognition task and the pupil dilation analysis suggest that participants treat positive and negative emotions differently. An arguably simpler model would be one where more positive emotions lead to a higher weight on the other's payoffs. In sum, it would be good to better motivate which models are included (or not), and perhaps include several other competing models.
Another interesting feature of the modeling exercise is that it can help to quantify the relative importance of facial expressions. The best performing model predicts 86% of the decisions correctly. To judge whether this is a lot or a little, it would be good to report the accuracy of competing models (e.g. self-interest or 'standard' inequity aversion without facial expressions). It would also be helpful to report the log-likelihood and BIC for each model. Reporting all this (for all models) would help to understand the added value of facial expressions.
In the experiment, participants are given explicit instructions on how to make decisions (page 23, lines 644-654). I think this is a poor design choice if you study how people make decisions.
The sample size is rather small (n=44). Moreover, almost half (21 out of 44) of the participants are told to be playing against a computerized strategy, although the authors note that this did not affect decisions. I do not understand the reasons why it was not possible to match people with a confederate (page 22). Given that the study uses deception, it seems easy enough to always tell people that they are playing with a real person, but perhaps I miss something. Additionally, it is unclear what 'playing against a computerized strategy' means here. Are participants told that their decisions affect someone else's earnings? This seems crucial for social preferences to have a bite.
In the experiment, the proposers' expressions and offers are a function of the history of the game (responders do not know this). This makes it hard to identify if responders really respond to the expressions on the pictures, or if they respond to other factors in the history of the game, such as previous earnings or previous offers. For example, Figure 4 shows that responders' decisions are affected by the offer in the preceding trial (n-1). However, as the offer in trial (n) is a function of the offer in trial (n-1), this could simply pick up the effect of the current offer (n).
-
###Reviewer #1:
The authors use an iterative ultimatum game to show that the proposer's facial expression, as well as the offer amount, influence human choice behavior. In particular, it is suggested that a proposer's facial responses to a participant's decisions specifically modulate the negative influence of perceived inequality on decision values. The combination of a game theoretic behavioral choice paradigm with computational cognitive modeling and a physiological arousal measure is appealing. I do, however, have some major concerns with novelty and interpretability, listed below in order of importance.
It is not particularly surprising that participants are more willing to accept an advantageous inequality if the proposer signals, with a smile, that it pleases them (or, conversely, less willing to accept if the proposer signals …
###Reviewer #1:
The authors use an iterative ultimatum game to show that the proposer's facial expression, as well as the offer amount, influence human choice behavior. In particular, it is suggested that a proposer's facial responses to a participant's decisions specifically modulate the negative influence of perceived inequality on decision values. The combination of a game theoretic behavioral choice paradigm with computational cognitive modeling and a physiological arousal measure is appealing. I do, however, have some major concerns with novelty and interpretability, listed below in order of importance.
It is not particularly surprising that participants are more willing to accept an advantageous inequality if the proposer signals, with a smile, that it pleases them (or, conversely, less willing to accept if the proposer signals discontent), particularly in light of previous work having already shown that both advantageous and disadvantageous inequalities are more frequently accepted if the proposer is smiling than if the proposer looks angry (e.g., Mussel et al., 2013). The addition of pupillary data could have added a fundamentally different dimension to such findings; however, since pupil size could not be significantly related directly to model-based decision values (please make this null effect more salient to the reader, unless I have misunderstood it), the choice data and physiological measure seem disconnected, which weakens the impact of each.
The authors argue that the ecological validity of previous work assessing the influence of facial expressions on UG decisions (e.g., Mussel et al., 2013) was limited by the use of non-contingent affective stimuli in independent, one-shot, games. It could be argued, however, that the response-contingent affective and monetary feedback used in the current study threatens construct validity, by conflating game theoretic strategizing with basic reward learning. This is particularly problematic since the computational models lack a representation of learning, or any incorporation of feedback over trials, in spite of such information being shown to profoundly influence acceptance decisions in model-free analyses. Given the overall emphasis on changes in participants' behavior across trials, it is important to formally characterize those learning curves, using reinforcement learning or some other relevant computational framework.
It appears that a parabolic modulation was considered for the inequality term, but not for the self-reward term. Given the dramatic improvement in model-fits across exponential and parabolic modulations of the inequality term, it would be interesting to see the performance of a model that includes parabolic modulation of both self-reward and inequality.
Given the apparent difference in affective modulation of advantageous vs. disadvantageous inequality, the exclusive focus on advantageous inequality in the discussion of model-based analyses makes it difficult to map modeling results to potential underlying psychological constructs (also, it is unclear how results from separately modeled advantageous and disadvantageous inequalities were integrated during model selection).
Another difficulty with data interpretation is the absence of a comparison across different total amounts (e.g., 200 out of 1000 vs. 200 out of 300). It seems to me that the constant total (of 1000) may have unduly focused participants on the inequality, over self reward.
"This indicated that participants' affective biases were more prominent for negative emotions, causing them to under-estimate the severity of negative affective displays". It is unclear from the methods whether asymmetries in the rated valence of facial expressions reflect a bias on the part of participants, or a limit on the confederates' abilities to simulate a range of negative expressions.
"After excluding six extreme outliers [...]" Please account for the methods and effects of outlier exclusions.
-
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 3 of the manuscript.
###Summary:
There was consensus among the reviewers that this paper addresses an interesting and important question of how social, affective and economic variables are formally integrated in strategic decision-making. However, the absence of a model-based account of how repeated game strategies and learning processes were shaped by the transition probabilities was a major concern, as was the lack of coherence between decision-making and pupillary effects.
-
