Introduction

It can be difficult to make good decisions. Suboptimal decision-making is well-documented and explanations for it have spawned extensive research in behavioral economics, cognitive psychology, and behavioral neuroscience (Ariely, 2008; Damasio, 2006; Kahneman, 2011). At the same time, “optimal” decisions are traditionally defined without considering constraints such as time, cognitive load, unfamiliarity with decision-tasks, genetics, and developmental history (Deck & Jahedi, 2015; Henrich et al., 2001; Kenrick et al., 2009; Krugel et al., 2009; Nadler & Zak, 2016; Ordonez & Benson III, 1997; Oud et al., 2016; Smith, 2003; Zak et al., 2005a).

Psychological and biological factors have been shown to influence both individual and interpersonal decisions, resulting in a more nuanced view of decision-making than labeling people as “irrational” if their decisions do not conform to a theoretical optimum (McKenzie, 2009; Oullier & Basso, 2010; Michel-Kerjan & Slovic, 2010; Zak, 2022). Yet, studies of the influences on decisions have nearly always focused on a single factor affecting a single decision. Due in part to this approach, a swath of the published findings have failed to replicate (Camerer et al., 2016, 2018; Nosek et al., 2022). Multiple central and peripheral neurophysiologic signals influence how organisms behave moment-by-moment so a one-to-one approach very likely misses the variety of influences on choices while also producing many false positives (Bhalla & Iyengar, 1999; Downes, 2005; Nosek et al., 2022).

Herein we take a different tack by seeking to catalog how a large set of physiological and psychological factors influence seven standard decisions studied by behavioral scientists. Our goal is to identify factors that should be included to explain the variation in decisions, as well as to document those that can be excluded. For example, do cognitive abilities impact risk-taking or cooperation? And, does testosterone influence trustworthiness or patience?

Clearly, not every physiological factor can be measured in one study. Moreover, some potential influences, such as cognitive abilities, are more effectively measured using psychological assessments. Thus, a combination of nine physiological factors and 15 psychological measures were obtained before and while participants made choices. Below we briefly discuss the rationale for the measures used in these exploratory analyses. It is worth noting, as detailed in Methods below, that measuring many independent variables induces a potential multiple comparisons problem. We resolve this by seeking robust individual predictors for each decision task by estimating both forward and backwards regressions, using this statistical sieve to eliminate inconsequential variables.

 Physiological Influences on Choices

The factors that influence choices are reflected in brain activity (Becker et al., 2012; Grabenhorst & Rolls, 2009; Padoa-Schioppa, 2011). The approach taken here includes both direct and indirect measures of biological states. Three groups of factors were included in this study: basal values of neurochemicals, measures of physiologic arousal, and surveys of attitudes and personality traits.

The neurochemicals measured in this study were included based on research showing their impact on decision-making. Glucose (G), the primary energy source for living creatures, was included because it affects cognition, feeding, and patience for rewards (Orquin & Kurzban, 2016; García et al., 2021) and varies over the day improving statistical reliability (Dufek et al., 1995). Cortisol (CORT) is the primary long-term arousal hormone in humans and influences cognition, aggression, risk-taking, and in some studies patience (Van den Bos et al., 2009; Duque et al., 2022). CORT has significant temporal and interpersonal variations that make it an attractive target to measure. Neuropeptide Y (NPY) is associated with resilience, stress modulation, and food consumption and interacts with CORT suggesting a possible influence on decisions (Winterdahl et al., 2017). In addition, three sexually dimorphic hormones were assayed. Estradiol (E) and progesterone (Pg) circulate at neuroactive levels primarily in women and impact metabolism, memory, and mood, thereby influencing choices, including trustworthiness (Garcia‐Segura, 2008; Stanton, 2017; Zak et al., 2004). Relevant to the present study, the effects of E and Pg vary over the menstrual cycle providing the variation needed for statistical tests. Testosterone (T) has been associated with risk-taking and overconfidence, as well as proactive and reactive aggression, predominately in males, which earns it inclusion in this study (Bashir et al., 2022; Nadler et al., 2017).

Physiologic stress had been shown to affect cognition and may increase or decrease risk aversion (Lupien et al., 2007; Kandasamy et al., 2014; Coates & Herbert, 2008; Riis-Vestergaard et al., 2018). Stress is also reported to inhibit patience, causing individuals to choose smaller more immediate rewards rather than larger delayed ones (Delaney et al., 2014; Riis-Vestergaard et al., 2018; Haushofer et al., 2013, Veszteg et al., 2021). In addition, some studies show that physiologic arousal increases cooperative behaviors in two-person strategic choices (Von Dawans et al., 2012), while others show increased self-interested behaviors (Hohman et al., 2018; Vinkers et al., 2013; Starcke & Brand, 2012). These findings warrant including measures of arousal in our analyses.

The extant research has measured physiologic arousal/stress in several ways. As a result, we assessed three electrophysiologic measures of arousal to determine which, if any, are robust predictors of decisions. The cardiac R-R interval was obtained from participants using an an electrocardiogram (ECG). The R-R interval is a more precise measure of physiologic arousal than is heart rate. ECG data were also used to derive high-frequency heart rate variability (HF-HRV) that captures activity of the vagus nerve and is inversely associated with arousal (Kreibig, 2010; Shaffer & Ginsberg, 2017). The third measure of arousal used electrodermal activity to calculate the change in skin conductance levels (SCL) from baseline (Figner & Murphy, 2011; van Dooren & Janssen, 2012).

Psychological Measures

Personality traits, affect, and attitudes have broad and varying effects on decisions (DeNeve & Cooper, 1998; Ifcher & Zarghamee, 2011; Phillips et al., 2016; Thielmann et al., 2020). As a result, a large number of measures were included in the analysis herein. These included measures of affect, reasoning abilities, and personality traits. We also obtained data on trait empathy, theory of mind, and prosocial preferences with a scale assessing interpersonal trust and greed (Kramer, 1999; Yamagishi & Yamagishi, 1994), as well as satisfaction with life.

The inclusion of multiple candidate predictors associated with nine standard behavioral decisions with a strict criterion for variable selection was designed to create a roadmap for researchers that seeks to avoid the nonreplication often found when testing the effect of one variable on one decision (Anderson & Burnham, 2004). The analysis in this study thereby generates a typology that identifies the factors that contribute to the variation in each decision as well as establishes a focused agenda for future research.

Methods

Participants and Procedures

A total of 75 participants (57% female) were recruited from the Claremont Colleges and surrounding community. Participants ranged in age from 18 to 66 (M = 27.17, SD = 11.69 ). The study was approved by the Institutional Review Board of Claremont Graduate University (#3290) and all participants gave written informed consent prior to inclusion. An alphanumeric code was assigned to each participant to anonymize their data. The study lasted approximately one and one-half hours and participants were paid between $30 and $50 (M=$46.16, SD=$5.21) for their time with earnings depending on participants’ decisions. There was no deception of any type. The experimental design exploits natural variations in neural activity across individuals to assess the contribution of each measure to variations in decisions. Age and sex were included as control variables.

After obtaining written consent, a qualified phlebotomist used a lancet to obtain a few drops of blood from participants’ fingers to measure glucose. After this, blood was drawn from an antecubital vein maintaining a sterile field using a Vacutainer blood collection kit (BD, Franklin Lakes, NJ) using two 8 ml serum-separator tubes to establish basal hormone levels. Next, participants were fitted with autonomic neurophysiology sensors to collect data using a Biopac MP150 system (Biopac Systems Inc., Goleta, CA). This included a 3-lead electrocardiogram (ECG) on the torso, and two Ag–AgCl electrodes fitted on participants’ distal phalanx surfaces of the middle and index fingers of the non-dominant hand to capture electrodermal activity (EDA). Next, participants were seated at partitioned computer stations where they completed surveys that included demographics as well as state and trait measures described below. All sessions included four participants.

After completing surveys, participants were provided with written instructions before each decision task and a chance to ask questions prior to making decisions. Each decision task was preceded by a five minute quiet period to acquire basal nervous system activity for cardiac and EDA responses. These were measured throughout each decision task to capture the change in activity from the pre-task baseline to that during the task itself. The order of the tasks was counterbalanced. After completion of decision tasks, participants were paid their earnings in private and dismissed.

Surveys

Trait and attitude data were obtained using the Positive and Negative Affect Schedule (PANAS) that measures current affect, the cognitive reflection test (CRT; Frederick, 2005) that quantifies reasoning abilities, and the five factor measure of personality traits using the mini-IPIP in which the factor “Openness” is denoted “Imagination” and we will use the latter term here (Donnellan et al., 2006). We also measured trait empathy using the interpersonal reactivity index (IRI), theory of mind with the Reading the Mind in the Eyes test (RME), prosocial preference with a scale assessing interpersonal trust and greed (Kramer, 1999; Yamagishi & Yamagishi, 1994), as well as satisfaction with life (SWL, Diener, Emmons, Larsen & Griffin, 1985). We anticipated that both physiological measures and self-reported items would be associated with participants’ decisions.

Decision Tasks

A set of standard decision tasks was used to provide comparability to the existing literature. All decisions were made privately in partitioned computer stations with full information and were incentivized by moderate monetary returns following standard practice in experimental economics (Smith, 1976). The decision tasks are briefly described here with the more detail in the Appendix. Participants made three individual decisions that measured risk-taking, intertemporal discounting, and altruism. Risk aversion was derived from a series of binary lottery choices (Holt & Laury, 2002) and finding the indifference point. Patience was determined by comparing a present gain to a larger future payment. Choice indifference was used to identify a parameter for standard geometric discounting (β) and a present-oriented hyperbolic discount factor (δ) following the behavioral economics literature (Laibson, 1997). Participants also made a unilateral decision to transfer some part of a $10 endowment to another anonymous participant to measure altruism in a task known as the Dictator Game (DG; Camerer & Thaler, 1995). The DG transfer decision was made twice after rematching in order to increase the number of observations.

Strategic decision-making was measured using (i) the prisoner’s dilemma game (PD), (ii) the ultimatum game (UG, Thaler, 1988), and (iii) the trust game (TG, Berg et al., 1995) using zTree software (Fischbacher, 2007) with random rematching of dyadic partners and neutral language in instructions. Simultaneous PD decisions were made three times, while sequential UG and TG decisions were made four times, twice each as decision-maker 1 (DM1) and as DM2. Random rematching was employed before each decision to eliminate the effects of reputation while DM1 and DM2 assignments were made randomly.

In the PD, two participants decide simultaneously whether to cooperate or defect and can earn up to $9 per round if one party chooses to cooperate and the other defects. Participants earn $6 if both cooperate and $2 if both defect. The PD captures expectations about the likely cooperation of others using theory of mind. The UG also depends on theory of mind as DM1 was endowed with $10 and DM2 had nothing. DM1 was instructed to offer a split of the $10 to DM2. Both DMs were informed that if DM2 does not accept the offer, s/he can reject it and both DMs would earn nothing. The “strategy method” was used in which DMs made decisions as DM1s and were ask to report their minimum acceptable (Zak, Stanton  & Ahmadi,  2007) offers as DM2s. This approach also allowed us to calculate UG generosity defined as DM1 offer minus DM2 minimum acceptance. Prosociality in the UG is measured as larger DM1 offers, a lower minimum acceptable offer as DM2, and greater generosity.

In the TG, both DMs were endowed with $10 and after instruction, software directed DM1 to choose an integer amount, including zero, of his/her endowment to send to DM2. The chosen amount was removed from DM1’s account and tripled in DM2’s account. Software reported the tripled amount DM2 received from DM1 and the total in DM2’s account. DM2 is then prompted by software to send an amount back to DM1, including zero; the return transfer was removed from DM2’s account and transferred one-to-one into DM1’s account. The consensus view is that the DM1 to DM2 transfer measures trust and the DM2 return transfer captures trustworthiness (Johnson & Mislin, 2011).

Neurochemicals

Glucose was measured using disposable tests strips and an electronic glucose meter (Care Touch, Future Diagnostics LLC, Brooklyn, NY). Blood samples for the remaining neurochemicals were rocked and sat at room temperature for 30 min to facilitate coagulation. Blood tubes were centrifuged at 1500 RPM for 12 min following our previous protocols and serum extracted (Zak et al., 2004). Serum was pipetted into 2 ml microtubes with screw caps (FisherBrand, Thermo-Fisher Inc., Waltham, MA). Microtubes were immediately placed on dry ice and transferred to an − 80 C freezer until they were assayed. All assays were performed by the Endocrine Research Laboratory at the University of Southern California (Los Angeles, CA) using commercial radioimmunoassay kits (CORT, E, T, Pg: DiaSorin, Stillwater, MN; NPY: Millipore, Burlington, MA). Inter- and intra-assay coefficients of variation were less than 15% for all analytes. There are several forms of testosterone (total, free, and dihydrotestosterone) and all are highly correlated. We obtained the most typical measure of testosterone, total testosterone (T).

Autonomic Physiology

Cardiac (sampling rate 1 kHz) and electrodermal activity (sampling rate 250 Hz) were collected with BioNomadix® transmitters and recorded with AcqKnowledge® software version 4.2 (Biopac Systems Inc., Goleta, CA) after participants washed their hands using supplied non-detergent bar soap. Research assistants monitored electrophysiology during the study to ensure data collection and created “flags” using AcqKnowledge software to denote task onset and completion. R-R signals were baseline corrected for all decisions then passed through a band-pass finite impulse response (FIR) filter to remove high- and low-frequency noise. HF-HRV was measured using the standard deviation of the interbeat interval during each decision (Kreibig, 2010; Shaffer & Ginsberg, 2017). Decision times varied from 13 s to other 30 s per decision (M = 26.07 SD = 24.31). This measure of vagal tone is typically used for short measurements (< 60 s; Salahuddin et al., 2007). Electrophysiologic data were down-sampled to 1 Hz to reduce noise (Luck, 2014) and were manually inspected for anomalies.

Statistical Approach

We used an exploratory data mining technique (Tan et al., 2016) to identify robust predictors for decision tasks rather than seeking to test hypotheses for each candidate predictor. This approach leans into individual variation while at the same time using a strict criterion that eliminates false positives (Lopez-Miguel, 2021). For example, CORT has diurnal variation and, by design, study sessions were run at various times of day to capture this variation. The use of controls for time of day or meal times, in the case of glucose levels, would decrease the likelihood of finding robust relationships which is the goal of this study (Lenz & Sahn, 2021). While there are multiple ways to perform model selection analyses (Zhao & Yu, 2006; Johnson & Rossell, 2012), we used an analytical approach focused on clarity and reliability (Genell et al., 2010).

The analysis begins by testing for bilateral relationships for each independent variable with each decision task. As this is exploratory, no correction is made for multiple comparisons. Then, forward and backward stepwise regressions were estimated to identify significant predictors while eliminating others. Forward regressions start with a null model and then add predictors based on their p-values until no additional variables meet the inclusion criterion. The backward stepwise regression model begins with all candidate predictors after which variables are excluded until only those with significant p-values remain. Robust predictors were those that were statistically significant in both forward and backward regressions. This approach balances Type I and Type II errors because each dependent variable is independent of the other dependent variables, with our analysis excluding variables using p ≤ .05 (Rothman, 1990; Perneger, 1998; Heinze et al., 2018). All models will be tested for multicollinearity using variance inflation factors (VIFs).

Results

The significant and robust variables related to each decision component and their signs are reported in summary form in Table 1. The detailed findings are discussed in the text below. Table 2, in the Appendix, presents the summary statistics for the 10 dependent variables, the nine neurophysiologic independent variables, 15 trait and attitude measures, and two controls.

Table 1 Robust predictors and their signs for each decision task are variables that were significant in both forward and backward stepwise regressions. Other predictors are independent variables that were significant in either the forward or backward regressions but not both
Full size table

Risk Preferences

Risk aversion was significantly correlated with participants’ basal E (r = − .325 p = .004) and self-reported personal distress. (r = .279 p = .015). Consistent with the bivariate findings, the forward and backward stepwise regressions selected E (-) and personal distress (+) as significant predictors of risk aversion with opposite signs. The backward stepwise linear regression also included RME (+), T (-), male sex (+), and neuroticism (-), with a higher variance explained (R2 = 0.309, F(6, 68) = 5.08 p = .0002) compared to the forward stepwise linear model (R2 = 0.194, F(2, 72) = 8.66 p = .0002). Neither model suffered from multicollinearity (VIFs < 1.01).

Time Preferences

Bivariate correlations showed the coefficient for the geometric discount factor for time preference (β) was negatively correlated with E (r=-.281, p = .015) and positively correlated with CRT (r = .422, p < .001). Stepwise regressions in both directions confirmed the effect of CRT(+) on β and added agreeableness (-). The forward model explained 23% of the variation in present bias time preference (R2 = 0.229, F=(2,72) = 10.665 p < .001) while the backward model also included perspective taking (+) which increased the R2 to 26% (F=(3,71) = 8.29 < 0.001, p < .001; VIFs < 1.03).

The impatience parameter (δ) was correlated with CRT (r = .447 p < .001) and E (r=-.271 p = .019). The forward and backward stepwise regressions confirmed that δ was statistically associated with CRT (+). The backward regression also identified G (+), PG (-), and age (+) as related to δ (ps < 0.001). The model with only CRT explained 20% of the variation in impatience (R2 = 0.20 F(1,73) = 18.20, p < .001) while adding G, PG, and age to the model drove up the explained variation (R2 = 0.289, F(4,70) = 7.12, p < .001). Neither model suffered from multicollinearity (VIFs < 1.10).

Altruism

Bivariate correlations for DG transfers were found for T (r = − .266, p = .021), age (r = .301, p = .009), CRT (-0.234, p = .044), and distrust (r=-.239, p = .039). Forward (R2 = 0.13 F(2,72) = 5.72, p = .005) and backward (R2 = 0.302, F(8,66) = 3.57, p = .002) stepwise regressions found that T (-) and age (+) were robustly associated with altruism. The backward regression also added female sex (-) imagination (−), CRT (-), positive affect (-), negative affect (+), and greed (-). The two variable regression explained 16% of the variation in altruism and did not suffer from more than moderate multicollinearity (VIFs < 3.5, F(3,71) = 4.56, p = .006).

Cooperation

Average cooperation rates in the PD were associated with T (r=-.419 r < .001), male sex (r = .333 p = .004), and CRT (r=-.300 p = .009). Unsurprisingly, average profits were inversely correlated with the cooperation rate (r=-.531, p < .001). Forward stepwise regressions (R2 = 0.279, F(3,71) = 9.16, p = .0000) and backward stepwise regressions (R2 = 0.343, F(5,69) = 7.23, p = .0000) revealed that cooperation was influenced by T (-), HF-HRV (+), and CRT (-) with low VIFs (<1.40). The backward stepwise regression added two additional variables, positive affect (+) and perspective taking (-).

Ultimatum Game

DM1 offers in the UG were negatively correlated with CORT (r=-.233, p = .044), T (r=-.242, p = .037), and self-reported distrust (r=-.274, p = .017). The forward (R2 = 0.075, F(1,73) = 5.94, p = .0173) and backward stepwise regressions (R2 = 0.364, F(8,66) = 4.71, p = .0001 all VIFs < 4.4) confirmed the effect of distrust (-). The backward stepwise regressions included seven additional variables related to DM1 offers, female sex (-), imagination (-), T (-), CRT (-), positive affect (-), CORT (-) and NPY (-).

The average minimum acceptable offer by DM2 in the UG was associated with T (r=-.3141 p = .006), CRT (r=-.312, p = .006), and neuroticism (r=-.296, p = .011). The forward (R2 = 0.213, F(3, 71) = 6.42, p = .0007) and backward (R2 = 0.229, F(4,70) = 5.19, p = .0010; all VIFs < 1.15 stepwise regressions only identified T (-), CRT(-), and neuroticism (-) as robust predictors while the forward estimation added SCL (-) and the backward included extraversion (+).

Generosity in the UG, the difference between DM1 offers and DM2 minimum acceptable offers, was associated with positive affect (r=-.252, p = .029) and imagination (r=-.231, p = .046). The forward regression (R2 = 0.128, F(2,72) = 5.28, p = .007) replicated the bivariate findings but the backward regression (R2 = 0.299, F(7,67) = 4.10, p = .0008, all VIFs < 1.25) screen only let through imagination (-). The backward estimation also included RME (-), DM1 R-R (-), neuroticism (+), DM1 HF-HRV (+), CORT (-), and conscientiousness (+).

Trust

The amount sent in the TG was unrelated to any of the independent variables in bivariate correlations (ps > 0.05) though NPY (r = .224, p = .0534), DM1 HF-HRV (r = .218, p = .0605) and DM1 SCL (r = .216, p = .0626) trended towards significance. Confirming this, the forward stepwise regressions had no significant variables. Backward elimination (R2 = 0.116, F(3,71) = 3.21, p = .0282) included distrust (-), DM1 SCL (+), and conscientiousness (-).

Trustworthiness

The average percentage returned by DM2s was only significantly correlated with SWL (r = .241, p = .037). The forward regression (R2 = 0.058, F(3,71) = 3.21, p = .037) and backward elimination (R2 = 0.125, F(3, 71) = 3.37; VIFs < 1.05) confirmed the effect of SWL (+) on trustworthiness. The backward estimation also included extraversion (-) and CRT (-).

Discussion

Our exploration of robust predictors used multiple simultaneous measurements of physiologic activity, personality traits, and attitudes to understand which factors affected 10 different decisions across six behavioral tasks. While measuring multiple possible predictive variables is logistically complicated, our findings show there is value in this omnibus approach. As expected, most of the nine physiologic variables and 15 trait and attitude variables were not robust predictors of the behaviors included in this study, although many of them were significant in one of the stepwise regressions. We believe this approach is valuable to the research community because it limits the factors that need to be measured in future studies and identifies those that need additional study to understand how they impact brain and behavior. By design, we made little use of control variables for time of day or meal consumption in order to decrease the likelihood that variables such as CORT or G would be significant. Yet, if one were to test the hypothesis that CORT decreases generosity in the UG based on the findings reported here, one would want to include a time of day control to add statistical precision.

Hormones

Basal levels of two hormones, E and T, were robust predictors of several decision types. E was negatively associated with risk aversion which is consistent with women generally being more risk-averse than men (Byrnes et al., 1999; Croson & Gneezy, 2009; Zethraeus et al., 2009; Nelson, 2015). E peaks twice a month for naturally cycling women and our findings suggest that women’s risk-taking will vary over the course of a month (Dreher et al., 2007; Diekhof, 2015; Op de Macks et al., 2016; Lazzaro et al., 2016). This findings is consistent with male sex being a significant positive but nonrobust predictor of risk aversion, although the negative sign on T in the backward stepwise regression is contradictory. More research is warranted on how E, T and sex interact to influence variations in risk aversion.

We also found that T robustly reduced cooperation in the PD and altruism in the DG, while it decreased the minimum acceptance threshold for offers in the UG. The first two are indicators of antisocial behaviors that have been well-established in the behavioral literatures for both endogenous T (Zak et al., 2005b; Burnham, 2007) and T administration studies (Zak et al., 2009; Bashir et al., 2022). T decreasing the UG acceptance threshold could be viewed as prosocial but recent findings suggest this may be a power or status demonstration (Dreher et al., 2016). As the T in males starts to decline after age 30 (Melmed et al., 2015), our findings suggest that prosocial behaviors are expected to increase when men age as others have shown (Cutler et al., 2021). Consistent with this, the analysis here identifies age as increasing altruism in the DG consistent with a large literature (Zak et al., 2022b). Future studies should include participants in wider age ranges in order to quantify if and when T falls sufficiently to change social decisions. A limitation of this study was the collection of T only in men and E and Pg only in women which induces a collinearity with sex that the present study is unable to fully disentangle.

Basal levels of the other hormones measured did not meet the criterion to be robust predictors of the behaviors in this study. Yet, we found that other hormones were significant in either the forward or backward stepwise regressions but not both. Of note is that G was shown to decrease the hyperbolic patience parameter δ. Low glucose is known to motivate feeding behaviors to maintain energy levels and our finding is partially supported by a meta-analysis of 42 studies showing that glucose influences intertermporal discounting and does so more strongly when the future reward is food (Orquin & Kurzban, 2016). Progesterone levels may also decrease patience though there is no existing literature or mechanism to our knowledge that supports this finding, but it should be investigated in future studies, for example, in women on birth control pills that typically contain Pg as a pharmacologic intervention.

Stress

A number of measures of physiologic arousal/stress and factors that modulate it were used in this study, including the R-R interval, SCL, HF-HRV, CORT, and NPY. Baseline, R-R, HF-HRV and SCL are significantly correlated with each other (ps < 0.05) and CORT trends toward correlation with HF-HRV and SCL (ps < 0.08). Being relaxed, as measured by higher HF-HRV, robustly increased cooperation in the PD. Studies in a variety of fields have shown that high levels of stress are typically associated with self-focused, rather than other-focused, behaviors. High stress levels are associated with aggression (Sandi & Haller, 2015), increased risk taking (Bendahan et al., 2017) and reduced prosocial behaviors consistent with our PD finding (Schulreich et al., 2022; Vinkers et al., 2013).

Several measures of stress were significant predictors in either the forward or backward stepwise regressions. UG offers were higher for participants with lower CORT and lower NPY. NPY typically rises in response to an increase in CORT in order to modulate stress responses (Morgan et al., 2002). Our finding needs to be explored further by using multiple measures of CORT and NPY to map their time course before, during, and after decisions. Moreover, CORT and NPY are often dysregulated in trauma victims so careful screening of participants’ medical histories is warranted in future research (Winterdahl et al., 2017). Interventions such as saunas or ice baths can raise CORT and NPY and could be used as interventions to probe this finding (Zukowska-Grojec & Vaz, 1988; Huhtaniemi & Laukkanen, 2020). Even consistent attendance at religious services is associated with higher NPY, identifying another way this relationship could be examined (Tønnesen et al., 2019).

Measures of stress also decreased generosity in the UG in either backward or forward analyses. The consistency of this relationship with all the correct signs was found for R-R, HF-HRV, and CORT indicating it likely signal and not noise and should be part of future studies. CORT is easily captured in saliva rather than using a blood draw as we have done, while R-R and HF-HRV require the equipment and experience to analyze an ECG, so saliva is the most efficient way to measure basal arousal. In addition, stress levels are easily increased using social stressors such as the Trier stress test (Kirschbaum et al., 1993) or the cold pressor test in which participants hold their hands in ice water for as long as they are able (Lovallo, 1975). These interventions could help establish the causal relationship between stress and generosity. Additional confirmatory evidence for this finding is that synthetic oxytocin infusion reduces stress (Kemp et al., 2012) and increases generosity in the UG (Zak et al., 2007).

The final measure of stress in this study, SCL, was significant in one of the regressions for the UG acceptance threshold and trust but carried unexpected signs and is inconsistent with other findings. SCL can change for a variety of reasons, for example, temperature in a room, and we suggest these findings are not worth additional explorations until other more promising influences are examined.

Traits

Cognitive abilities measured by the CRT robustly increased intertemporal patience, being related to both parameters β and δ. CRT was also shown to decrease cooperation in the PD and decrease the acceptance threshold in the UG. Patience is required to delay gratification in order to obtain larger, delayed rewards (Frederick, 2005). Delaying rewards has been associated with prefrontal suppression of consumptive activity (Mitchell et al., 2011; Kable, 2014) that correlates with CRT scores (Oldrati et al., 2016). Similarly, cooperation in the PD has been associated with dorsolateral prefrontal cortex activity (Emonds et al., 2012; Rilling et al., 2004) and when this is actively disrupted using transcranial magnetic stimulation, cooperation rates fall (Soutschek et al., 2015). Our finding that CRT decreases the UG acceptance threshold is consistent with previous findings for the CRT in the UG.

Relatedly, altruism was robustly negatively related to trait imagination. This personality attribute is positively related to intelligence and creativity and is similar to the trait of openness (Harris, 2004). Such individuals are generally more prosocial on self-report (Guo et al., 2019) while another personality trait not measured here, honesty-humility, has been associated with increased DG allocations (Zhao et al., 2016; Hilbig et al., 2015). Our finding that imagination reduced altruism is supported by its robust negative relationship to generosity in the UG, a finding that also has not been reported in the literature. The relationship between imagination and prosocial behaviors is of sufficient interest that it should spark future studies that broaden the subject pool, the amount at stake, and thoroughly measure various aspects of personality and physiology to understand how imagination relates to prosociality.

Those with another personality trait, neuroticism, were more prosocial by demanding less to accept an offer in the UG. People who score high on neuroticism are generally anxious and less prosocial (Guo et al., 2018), yet there no significant correlations between this trait and any of our measures of arousal responses, both at baseline and during task completion. A recent meta-analysis of the effect of stress on prosociality found no overall relationship with UG decisions (Nitschke et al., 2022), but the variety of stress inductions and measures in this analysis call such findings into question. A possible mechanism for our result is that those who score high in neuroticism are plagued by worry (de Bruin et al., 2007) and they may have modulated this concern by accepting less in the UG to avoid rejection. The present study used the strategy method in which DM2s in the UG had to decide their minimum acceptance threshold in advance of receiving a DM1 offer. Consistent with this finding is the statistically significant but nonrobust positive relationship between neuroticism and generosity in the UG. Further examination of the relationship between personality traits, worry, and prosocial behaviors is certainly warranted.

Two attitudinal measures also affected prosocial behaviors. Self-reported distrust robustly reduced offers in the UG and proposals in the TG, though the latter only in the backward stepwise regression. These findings are expected and provide evidence that the measures and analytical approach used herein are reasonable. We also found that trustworthiness was robustly and positively related to participants’ SWL. There is an extensive literature relating SWL to various types of prosocial behaviors (Reich et al., 2019) providing support for this finding. A history of prosocial behaviors, as well as in-experiment charitable giving, have been positively associated with the brain’s release of oxytocin suggesting an interaction between traits and behavior that are reflected in differences in neural responses (Zak et al., 2022b). Trustworthiness itself is associated with oxytocin release (Zak et al., 2004, 2005a) and at the country level is strongly associated with economic growth so its relationship to SWL is important to understand for both individual and aggregate measures of quality of life (Zak & Knack, 2001; Knack & Zak, 2003; Zak & Fakhar, 2006).

One attitudinal measure, personal distress, robustly increased risk aversion. Distress is an aversive emotional reaction that focuses attention on immediate options so this finding is not unexpected (Schneewind & Kupsch, 2007), though it is understudied in the behavioral sciences. In the extreme, those diagnosed with clinical anxiety appear to be highly risk-averse (Lorian & Grisham, 2011). Being male and RME were positively associated with risk aversion in one-direction significance, but these have unexpected signs and may be noise. Men typically express less personal distress than women (Viertiö et al., 2021; Bilodeau et al., 2020) and are known to be less risk averse while the RME skews male (Kirkland et al., 2013). Biological sex failed to be a robust predictor of any of the behaviors studied, though in a backward stepwise regression women were less altruistic than men.

Conclusion

The analysis herein identified a set of physiologic variables, personality traits, and attitudes that were robustly related to decision tasks used by behavioral scientists in experiments. The present analysis has identified physiologic variables from which causal studies should be done. For example, estradiol can be administered to establish its causal relationship to risk aversion while testosterone administration can establish how it impacts choices in the prisoner’s dilemma game. Additionally, why those who are trustworthy have greater satisfaction with life is begging for additional research, with a recent finding indicating a role for the neurochemical oxytocin (Zak et al., 2022b). Our findings that cognitive abilities robustly influence patience as well as cooperative behaviors also needs more explanation using both surveys and neurologic measures. Perhaps the most fruitful area for future research will be to examine if the significant but nonrobust predictors from one stepwise regression reach the level of robust significance when, for example, a different sample is measured and controls are included. Indeed, we found that the ultimatum game had a large number of variables across categories that were significantly related to choices, but not robustly so, with several measures of stress/physiologic arousal appearing to decrease cooperation. Relatedly, physiologic arousal was recently shown to predict which study participants renege on promises to reciprocate in a $500 trust game suggesting that the way that stress influences cooperation is an area ripe for additional research (Zak et al., 2022a). Also left for future research is whether the findings we report here hold up in field experiments.

The most important take-away from this analysis is that labelling behaviors as “irrational” without the kinds of comprehensive measurements done here is unwarranted. More generally, the multi-modal methodology we used that relates measures of physiologic activity and psychological factors to decisions can be extended to more decision types and additional measures to more fully understand and appreciate the varieties of the human experience.