We tested behavior and brain activity during a social decision task in which individuals faced the same underlying tradeoffs between options, but learned about these tradeoffs through past experience or concurrent description. In each condition, individuals made tradeoff decisions between either maximization of self-profits or societal-benefits (FIGURE 1). We define self-profit (society-benefit) maximizing decisions as those in which the individual receives more (less) points than society. In addition to the decision task, we also measured social preferences using the Social Value Orientation scale (SVO), which gives the degree of a person’s prosociality angle with lower values indicating more selfish preferences46.
Prosocial choices in description and experience trials
We first tested the outcomes of social decisions as a function of prosocial preferences. We computed a hierarchical Bayesian beta regression model that sought to explain the proportion of prosocial choices each participant made as a function of the condition (either description (DE) or experience (EX) trials), controlling for their prosociality, as determined by the SVO scale. Prosocial choices in the description and experience trials did not differ (Bayes Factor10 = 0.06; regression coefficient = -0.02 95% credible interval (CI) = [-0.13, 0.08]; see Equation 1). According to scores on the independent SVO scale, most participants had moderately selfish preferences, but there was considerable variability in prosocial preferences across participants (FIGURE 2B). As expected prosocial choices in the decision task increased significantly as a function of SVO across both conditions (coef = 0.57, 95% CI = [0.46, 0.68], PPmcmc > 0.999; FIGURE 2C). However, the relationship between SVO and prosocial choices did not differ between description and experience choices (Bayes Factor in favor of a regression model without an SVO*condition = 18). In summary, the regression results indicate that decision outcomes did not differ between description and experience trials in the specific sets of social tradeoffs we tested here.
Response times in experience versus description decisions
We also compared response times across conditions and found that they were faster in experience relative to description trials. A hierarchical Bayesian linear regression explaining the natural logarithm of response times as a function of condition and SVO showed that the mean response times were faster in experience than description trials (EX mean = 0.98 ± 0.54 ms; DE mean = 1.53 ± 0.66 ms; coef = -0.5, 95% CI = [-0.51, -0.48], PPmcmc > 0.999; see FIGURE 2A and Equation 2). In addition, there was a main effect of SVO; response times increased as a function of SVO in both decision conditions (coef = 0.08, 95% CI = [0.04, 0.12], PPmcmc > 0.999). In other words, more prosocial individuals made their choices more slowly. However, there was also an interaction between SVO and condition such that response times were less strongly related to SVO in experience compared to description trials (coef = -0.05, 95% CI = [-0.07, -0.03], PPmcmc > 0.999). These results indicate that highly prosocial participants required more time than less prosocial participants to resolve the tradeoffs between self-profit and society benefits they faced in this task, especially if the tradeoffs were presented as explicit descriptions rather than learned through experience.
Information sampling in description and experience decisions
We fit hierarchical diffusion decision models (HDDM)47–49 to test how the chosen and foregone payoffs to self and society influenced the decision process in description relative to experience trials. We fit and compared two types of models that differed in how the chosen and unchosen payoffs to self and society influenced the evidence accumulation or drift rate (see Methods section on HDDM). The best-fitting HDDM parameters were able to explain the distributions of response times for prosocial and selfish choices in both the description and experience trials and (FIGURE 3).
Behavior during the description trials was explained by HDDM1, in which the mean drift rate was proportional to the chosen payoffs for self and society as well as the differences between chosen and unchosen payoffs for self and society (HDDM1 Description, DIC = 27419; TABLE 1). This is consistent with a large body of past literature reporting that evidence accumulation rates are proportional to the difference in (subjective) values between options during description decisions42,50–54. In contrast, including the differences between self and society payoffs did not improve the model fits to experience decisions. Instead, behavior on experience trials was slightly better fit by the simpler model, HDDM2, in which the mean drift rate was proportional to the chosen payoffs for self and society only (HDDM1 Experience, DIC = 11397; HDDM2 Experience, DIC = 11392; TABLE 1). Although, the model comparison results did not provide strong evidence in favor of either HDDM for experience trials, the posterior estimates of the drift weighting parameters from both HDDM1 and HDDM2 show that the evidence accumulation rates during experience trials are more strongly related to chosen outcomes than differences between chosen and unchosen outcomes.
TABLE 1. Posterior estimates and model fit for HDDM Models.
|
HDDM1 Description
|
HDDM1 Experience
|
HDDM2 Experience
|
Parameter
|
Mean
|
SD
|
Mean
|
SD
|
Mean
|
SD
|
Drift (b0), Intercept
|
0.065
|
0.026
|
0.260
|
0.044
|
0.214
|
0.037
|
Drift (b1), Payoff Outcome Society
|
0.120
|
0.016
|
0.108
|
0.023
|
0.089
|
0.018
|
Drift (b2), Payoff Outcome Self
|
0.285
|
0.018
|
0.414
|
0.026
|
0.428
|
0.020
|
Drift (b3), Payoff Difference Society
|
0.053
|
0.026
|
-0.105
|
0.044
|
-
|
-
|
Drift (b4), Payoff Difference Self
|
0.197
|
0.032
|
-0.041
|
0.045
|
-
|
-
|
Non-decision time (theta)
|
0.413
|
0.016
|
0.336
|
0.009
|
0.337
|
0.009
|
Boundary separation (alpha)
|
4.173
|
0.094
|
2.759
|
0.052
|
2.735
|
0.051
|
Starting Point Bias (bias)
|
0.480
|
0.007
|
0.458
|
0.009
|
0.457
|
0.009
|
DIC
|
27419.43
|
11397.05
|
11392.22
|
We used hierarchical drift diffusion models (HDDM) fit to the response time data from description and experience decisions. In HDDM 1 (see Equation 3 in the Methods section), we modeled the evidence accumulation rate as a function of both payoff outcomes (b1, b2) and payoff differences (b3, b4). In HDDM2 (see Equation 4 in the Methods section), we modeled the evidence accumulation rate only as a function of payoff outcomes (b1, b2). Additionally, we included the following free parameters at the group and subject levels: boundary separation (i.e., the evidence threshold for making a response), starting point bias for evidence accumulation and non-decision times. We list the mean and standard deviation (SD) for all parameter estimates. A lower DIC indicates better model fit.
Thus, the information upon which description and experience choices are based appears to differ, with experience choices being less sensitive to tradeoffs between chosen and unchosen outcomes for self and society. This decrease in sensitivity to unchosen outcomes during experience relative to description trials may explain why response times show a weaker relationship to social preferences (i.e., SVO) in experience compared to description decisions.
BOLD activity during description and experience social choices
To investigate how brain activity differed in social decisions made from description and experience, we analyzed the BOLD signal during each type of choice. We fit a generalized linear model that included regressors for the mean level of activity during description and experience trials as well as parametric regressors for the mean HDDM evidence accumulation rate on each trial. We report results that survive correction for multiple comparisons at the whole-brain level using non-parametric permutation tests with 5000 permutations for each contrast.
We found significant differences in mean BOLD activity during social decisions in description compared to experience trials. There was greater average activity during description than experience trials in several brain regions including the caudate, middle frontal gyrus, paracingulate and cingulate gyrus, frontal pole and dorsal precuneus (FIGURE 4A and Supplementary Table S1A). In contrast, average activity was significantly greater in more ventral portions of the precuneus and posterior cingulate during experience than description trials (FIGURE 4A and Supplementary Table S1B). Note that these contrasts testing for differences in mean activity across trial types included the participants’ SVO angles as a continuous covariate (z-scored across participants) to account for variation in social preferences.
Next, we investigated whether brain activation patterns during experience and description trials differed in participants who were more versus less prosocial. We divided our participants into two groups based on the median SVO in our sample (45.30 points). Previous fMRI studies investigating how BOLD activity during social decisions relates to SVO reported greater activity in medial orbitofrontal cortex (mOFC) and dorsal medial prefrontal cortex (dmPFC) for more selfish (i.e., low SVO) relative to prosocial individuals during a social choice task using the typical choices from description 22. We found a similar difference in the mOFC (MNI: -8, 44, -14; TFCE t-stat: 5.00), with low SVO individuals showing greater BOLD activity during description choices than high SVO participants. However, there were no significant differences in dmPFC activity in our sample after correcting for multiple comparisons. In contrast to description trials, we did not find any significant differences between high and low SVO participants’ BOLD responses during experience choices. A direct comparison between description and experience choices showed that there were significant interactions in the association between mean activity and SVO between experience and description trials within the set of brain regions listed in Supplementary Table S2, including the occipital cortex, posterior cingulate gyrus, posterior insula and the posterior portion of the temporoparietal junction (pTPJ, as defined by the Mars et al., atlas55; FIGURE 4B). In general, this interaction was driven by low SVO individuals showing greater activity during description than experience trials, while, in contrast, high SVO individuals showed more activity in experience than description trials. FIGURE 4C shows the patterns of BOLD responses underlying the interactions between decision type and social value orientation in the set of voxels that partially overlap with the posterior TPJ.
Trial-wise levels of BOLD activity were more closely associated with HDDM evidence accumulation rates in description than experience trials. Note that our DDM model was specified such that positive evidence accumulation rates promoted prosocial choices while negative rates promoted selfish choices, and that the evidence accumulation rates on each trial were participant-specific and thus incorporated participants’ social preferences. During description choices, the evidence accumulation rate was positively correlated with BOLD activity in regions including the caudate, dorsolateral prefrontal cortex, insula, putamen, and thalamus (Supplementary Table S3). No regions showed significant negative correlations with trial-wise mean evidence accumulation rates during description trials, and there were no significant positive or negative correlations in any regions during experience trials after correcting for multiple comparisons. Furthermore, the association between evidence accumulation rates and BOLD activity was significantly more positive during description than experience trials in the caudate nucleus, cerebellum, cingulate cortex, middle temporal and frontal gyri, and thalamus (Supplementary Table S4).