The aversive value of pain in human decision‐making

  1. Hocine Slimani
  2. Pierre Rainville
  3. Mathieu Roy

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This behavioural study in healthy participants examines how people trade-off a brief phasic pain stimulus with a monetary reward, reporting a quadratic effect of pain on decision making. It supports and adds to previous findings of a context-dependency deriving from the distribution of rewards, which is a deviation from conventional rational choice theory (which proposes that a particular level of pain should carry the same price, regardless of small context changes). Broadly, the reviewers found the work well executed and the data compelling, but there were some suggestions for alternative explanations that are not ruled out given the current data.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Background

In order to decide between avoiding pain or pursuing competing rewards, pain must be assigned an abstract value that can be traded against that of competing goods. To assess the relationship between subjectively perceived pain and its value, we conducted an experiment where participants had to accept or decline different intensities of painful electric shocks in exchange of monetary rewards.

Methods

Participants (n = 90) were divided into three groups that were exposed to different distributions of monetary rewards. Monetary offers ranged linearly from $0 to $5 or $10 in groups 1 and 2, respectively, and exponentially from $0 to $5 in group 3. Pain offers ranged from pain detection to pain tolerance thresholds. The value of pain was assessed by identifying the indifference points corresponding to a 50% chance of accepting a certain level of pain for a given monetary reward.

Results

The value of pain increased quadratically as a function of the anticipated pain intensity and was found to be relative to the mean and standard deviation of monetary offers. Moreover, decision times increased as a function of the intensity of accepted painful stimulations. Finally, inter‐individual differences in psychological traits related to harm avoidance and persistence influenced the value of pain.

Conclusions

This is the first demonstration that the value of pain follows a curvilinear function and is relative to the mean and standard deviation of competing monetary rewards. These new observations significantly contribute to our understanding of how pain is assigned value when making decisions between avoiding pain and obtaining rewards.

Significance

This work provides a description of the pain value function indicating how much people are willing to pay to avoid different intensities of pain. We found that the function was curvilinear, suggesting that the same unit of subjective pain has more value in the high vs. low pain range. Moreover, the pain value was influenced by the experimental manipulation of the rewards distribution and of the inter‐individual differences in harm avoidance and persistence. Altogether, the present study provides a detailed account of how subjectively experienced pain is assigned value.

Article activity feed

  1. eLife

    Evaluation Summary:

    This behavioural study in healthy participants examines how people trade-off a brief phasic pain stimulus with a monetary reward, reporting a quadratic effect of pain on decision making. It supports and adds to previous findings of a context-dependency deriving from the distribution of rewards, which is a deviation from conventional rational choice theory (which proposes that a particular level of pain should carry the same price, regardless of small context changes). Broadly, the reviewers found the work well executed and the data compelling, but there were some suggestions for alternative explanations that are not ruled out given the current data.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

  2. eLife

    Reviewer #1 (Public Review):

    The manuscript describes a series of behavioural economics experiments aiming to determine a value function for pain by giving participants the explicit choice to endure a series of electric shocks of varying intensity, for which they may receive a specified remuneration. The ambition is to establish a behavioural method for determining the value of pain and pain relief across the spectrum from low to high pain intensity. Theoretically, the research is informed by and aims to inform in turn, the Fear Avoidance Model of chronic pain. In its present form, the manuscript lacks the detailed methodological information needed for accessibility beyond a specialist audience.

    The authors conclude that the value of pain is curvilinear, however pain was not measured during the experiment, and pain ratings are not presented for the whole sample (N=90), leaving open the question of whether the pain intensity of the eight levels of electric shock, may also be curvilinear. If so, it appears possible that the relationship between pain intensity and value is strictly linear.

  3. eLife

    Reviewer #2 (Public Review):

    This is a behavioural study (healthy participants) looking at how people trade-off a brief phasic pain stimulus with a monetary reward. People make pairwise choices between a painful electrical stimulus and an amount of money, with different groups receiving offers from different ranges (0-5$; 0-10$) and distributions (skew or not). This allows the authors to estimate indifference points.

    There were several findings:

    i) a curvilinear intensity x value function, regardless of context.
    ii) a context effect: people require more money to accept pain if the range of offers is higher
    iii) decisions slowed when accepting pain, especially for high pain
    iv) higher trait harm avoidance was associated with high pain avoidance in the task
    v) with a skewed (exponential) distribution of trials, subjects end up accepting less pain, but tend to accept higher offers when made.

    Some comments on each finding:

    i) interpreting the shape (linear, non-linear) of the value function is always a bit tricky - it reminds me of the old data on the power law for stimulus response functions. But especially the issue here is that ratings or %tolerance is bounded, but intensity isn't, so inherently one is likely to get a non-linear function when doing any mapping onto a bounded scale. Of course this isn't really the main point of the study, but is worth noting

    ii) the context effects are interesting. Similar effects were shown by Vlaev for the context effect of range. The effect of exponential distribution I think is consistent with Chater's model of relative value effects.

    iii) Slower decisions with more pain are interesting. Fields's motivation-decision model deals with inhibiting pain when accepting 'greater' rewards, and shows slower innate pain responses (e.g. tail-flick). Did the authors gather intensity ratings? The lack of a choice difficulty effect is also interesting - is a drift diffusion model applicable (I don't think they placed time-pressure on the response)?

    iv) The harm-avoidance finding is not that surprising. How did the authors correct for multiple comparisons across the 5 principal components?

    v) for the greater profitability index - is this confounded by the fact that these subjects overall received less pain. This would an issue, for instance, if there was a cumulative effect of overall pain ('I can only take so many pains in one experiment')?

    So overall, the study supports and adds to previous findings of a context-dependency deriving from the distribution of rewards, which is a deviation from conventional rational choice theory. It's a nice experiment, appropriately powered and carefully executed, and develops the ideas behind the behavioural economics of pain.

  4. eLife

    Reviewer #3 (Public Review):

    Slimani and colleagues provide different groups of participants with offers to accept different levels of pain - in the form of electric shocks with different levels of intensity - for different amounts of money. They use their responses to map out the rate at which participants trade money for pain, explore the form of this function, examine how this form varies with the range and distribution of rewards available and examine how differences relate to questionnaire measures. They find evidence that the value of pain is curvilinear, exhibits range normalisation (i.e., adapts to context) and interacts with a factor relating to harm avoidance and persistence.

    The manuscript does a good job of articulating why understanding how individuals value pain could be important. But I felt it lacked a set of clearly defined hypotheses for the research questions they are interested in. As one example, why would the distribution of rewards potentially influence the value of pain and what prediction could we make about how it might do this? As another example, why use factor analysis to try to relate individual differences to pain evaluation? Without this, it currently reads as a mixture of different research questions without a clear understanding of how they are connected exactly or what predictions could be made.

    How humans put a price on pain is interesting and as the authors rightly point out, there is a lack of knowledge in the field about valuation in aversive domains. The design used to investigate this is simple and ideal to start to address questions that relate to this. However, the manuscript needs to be much tighter in terms of reporting the methods and statistics. I felt that there were some key gaps in the statistical reporting such as not making clear what the statistical tests used are, what the exact value of N is in each group and confidence intervals. I try to put some specific examples of this below.

    • To create the pain value function, a logistic regression predicting choices (accept/reject) using different levels of shock and reward is used. But key details about how this model is specified are missing (e.g., if this is a multilevel analysis, what were taken as random effects? In the quadratic model, did they include the linear effect as well?). When analysing questionnaire responses, did the authors include all 5 PCA components in their model to predict acceptance rates? Or did they run 5 separate models with a different component in each? And are the results for the questionnaire analysis (e.g., Figure 3a) pooled over all 4 groups of participants (if so did they control for group in the analysis, do the effects interact with group, etc.)?

    • Sample Size: The authors do not report whether an appropriate sample size was computed when the study was being designed - how did they determine the sample sizes that were used (e.g., power analysis)? The authors state in the transparent reporting form that this information is reported in the Statistical Analysis section but this is not the case from what I could see. They do report how they calculate the effect size (and they report the effect sizes throughout the paper); but this doesn't say anything about how the sample size was arrived at. The authors also don't state how many participants are assigned to each group, it should be stated clearly within the manuscript, figure legends etc.

    • Model Comparison: The authors compare models using AIC scores and report that the AIC was lowest for the quadratic model (Group 1 = 58, Group 2 = 60, Group 3 = 57), compared to the linear (Group 1 = 60, Group 2 = 60, Group 3 = 59) and the cubic (Group 1 = 60, Group 2 = 62, Group 3 = 58) models. But there is no metric provided whether the improvement of one model over and above another is meaningful or not - can anything really conclusive be taken from a difference in AIC scores of 2 for instance? Maybe it can - but the authors really need to make the case. It also seems that the scores are actually the same for Group 2 between quadratic and linear?

    • The authors report a t statistic showing the significance of the curvilinear effect in Figure 1a (t = 5.04, p < 0.001). But what is this testing exactly and why are there no degrees of freedom or confidence intervals reported?

    • Similarly, when comparing differences in the profitability index metric between groups, the authors report significant differences between group 3 vs all other groups (in the text) and between group 1 vs all other groups (in table 1). But there is no indication as to what the tests are used to make these comparisons and if any correction has been made for multiple comparisons.

    • There are some decision steps taken in various analysis that are not justified. For example, they use a smoothing kernel in the design matrix - is that necessary, what's the motivation? Why control for trial number with not just 1 but 3 different regressors? How did the authors decide on a criteria for deciding how many factors the optimal solution has?

    • Figure 3A and 3B: Is it the case that pain ratings are correlated with the factor ratings for harm avoidance / persistence. To put it another way, do the results in 3A hold if they control for pain ratings?

    • What software and packages were used to run the models?

  5. Version published to 10.1002/ejp.1895
  6. Version published to 10.1101/2021.02.19.431988v1 on bioRxiv