Do tufted capuchin monkeys, Sapajus spp., experience regret in decision making under risk?

.

In their daily lives, nonhuman animals face situations requiring a choice, in which they must account for both the potential payoffs and the potential costs associated with pursuing a certain course of action (Santos & Rosati, 2015). According to classical economic theory, rational decision makers should maximize their own expected utility, which refers to the magnitude of satisfaction received from consuming a good or service (Baron, 2000;Von Neumann & Morgenstern, 1947). However, it is well known that several factors lead decisions to deviate from this assumption, not only in humans but also in nonhuman animals. In decision making under risk, where the choice options imply a probabilistic variation in their payoffs, several factors affect individual preferences and make human and nonhuman animals less economically rational than expected, as for instance feeding ecology (Rosati, 2017;De Petrillo & Rosati, 2019;De Petrillo & Rosati, 2021), contextual factors (Heilbronner et al., 2013), social influences (Rosati & Hare, 2012;Zoratto et al., 2018) and an individual's energy budget (Kacelnik & Bateson, 1996;Stephens, 1981).
Emotions experienced after choice play a crucial role in shaping decision making under risk. Humans actively shift their preferences when they experience negative states as a consequence of unfavourable outcomes (i.e. disappointment; Camille et al., 2004;Coricelli et al., 2005). Moreover, humans make decisions not only based on what they receive, but also by comparing the outcome of their chosen option with the outcome they could have obtained if they had chosen differently, a phenomenon called counterfactual reasoning (Byrne, 2002;Coricelli et al., 2005;Coricelli & Rustichini, 2010). When this comparison is unfavourable, because the outcome of the forgone option is better than the outcome of the chosen option, people often experience regret (Camille et al., 2004;Coricelli et al., 2005Coricelli et al., , 2007. Empirical data confirmed that, when making a series of choices between two gambles, people not only adjust their choices when they are provided with feedback about the forgone option (and discover that they could have received a larger reward if they had chosen differently, Coricelli et al., 2007), but also anticipate that they will experience regret and take this possibility into account when making decisions (Camille et al., 2004;Coricelli et al., 2005;Coricelli & Rustichini, 2010).
Of course, humans can experience various types of regretful experiences, with profoundly different consequences and impact on their life. Yet, as proposed by Kim et al. (2015), regret-like emotions would be expected to be widespread also in nonhuman animals, as the ability to adjust future choices according to hypothetical outcomes would allow an animal to make adaptive choices quickly, without directly experiencing all possible choice outcomes. Although adjusting future choices on the basis of both actual and hypothetical outcomes may require sophisticated cognitive capacities, current empirical evidence suggests that both rodents and macaques react to hypothetical outcomes and may experience emotional states that are analogous to regret and consequential behavioural responses (Abe & Lee, 2011;Hayden et al., 2009;Lee et al., 2005;Steiner & Redish, 2014;Sweis et al., 2018). For example, when playing a computerized version of the game rockepaperescissors, rhesus macaques, Macaca mulatta, adjust their choice on the basis not only of what they received, but also of what they would have received if they had chosen a different option (Abe & Lee, 2011;Lee et al., 2005). In addition, rhesus macaques showed evidence of counterfactual reasoning (i.e. comparing the outcome of their chosen option with the outcome they could have obtained if they had chosen differently) when presented with a computerized risky decision-making task (Hayden et al., 2009). Here, monkeys faced eight possible options: seven always provided a small juice reward, and one sometimes provided a much smaller reward and sometimes a much larger reward. Crucially, after choosing, monkeys were always shown what outcome the risky option would have provided, and they preferentially selected the risky option after seeing that it would have provided a larger, missed reward in the previous trial.
Although there is neural and behavioural evidence of regret-like emotions in both rodents and macaques, these studies were carried out in constrained laboratory environments, where the subjects were often physically restrained and underwent food or fluid restrictions to increase their motivation, which may have increased their sensitivity to potential reward losses and subsequent emotional responses. Yet, to what extent regret-like emotions occur in zoo animals housed in less constrained conditions than the typical neurophysiological laboratory, and how these emotions affect their decision process, remains unclear. It is also not known whether nonhuman animals, like humans (Camille et al., 2004), show more intense responses when comparing the outcome of their chosen option with the outcome they could have obtained if they had chosen differently (regret-like response) rather than when just receiving an unfavourable outcome (disappointmentlike response).
In previous research, both apes (chimpanzees, Pan troglodytes, and bonobos, Pan paniscus) and capuchin monkeys, Sapajus spp., showed behavioural markers of negative emotions in response to unfavourable outcomes in a risky decision-making task (Rosati & Hare, 2013;De Petrillo et al., 2017; see also Davies et al., 2015 for similar data in chickens, Gallus domesticus). Specifically, subjects showed more negative behaviours, such as scratching and banging, and switching attempts (to revise their previous choice) after gambling and receiving a low-value outcome than they did after gambling and receiving a high-value outcome or after choosing the safe alternative. However, these studies measured individuals' emotional and behavioural response only when the outcome of the chosen option was revealed (partial feedback procedure) but not when the outcome of the unchosen option was revealed (complete feedback procedure). This makes it difficult to distinguish between the disappointment-like response for receiving a low-value outcome and the regret-like response for not having received the high-value outcome if choosing differently.
To bridge this gap, in the current study we examined whether capuchin monkeys exhibit regret-like responses, and how this experience affects their future decisions, in a risky decision-making task where subjects received information on both the outcome of their choice and the outcome of the alternative, unchosen option. Capuchin monkeys are well positioned for this investigation because they are omnivorous and opportunistic foragers and can flexibly adapt to a wide variety of different habitats (Fragaszy et al., 2004). In an ecological rationality framework (e.g. Gigerenzer et al., 1999), according to which feeding ecology is considered as an important selective pressure shaping decision making under risk (e.g. Rosati 2017), opportunistic foragers such as capuchins should be more likely to be prone to risk than species exploiting more constant food sources (Heilbronner et al., 2008). Moreover, in natural contexts capuchins often have to choose as a group between visiting alternative feeding patches, and within those patches they have to make individual decisions about where to forage (e.g. Fragaszy et al., 2004). Thus, they should be very careful in this decision-making process not to incur the loss of a resource through within-group scramble competition, and it is plausible that they could know the outcome of their decision relative to alternative choices. Additionally, in captive studies, capuchins share with humans several decisional biases (Lakshminarayanan et al., 2011;Watzek & Brosnan, 2020;Marini et al., 2022) and, as reported above, they exhibit behavioural markers of negative emotions when receiving an unfavourable outcome in a risky decisionmaking task (De Petrillo et al., 2017).
In the present study, capuchins faced a series of choices between a safe, constant option and a risky, variable option that provided either a higher or a lower food quantity with different probabilities. In each trial we scored, for each individual, scratching behaviour (a self-directed behaviour that indexes emotional states related to motivational conflict or anxiety, Maestripieri et al., 1992;Troisi, 2002) and switching behaviour (i.e. the attempt to modify the subject's initial choice by selecting the previously unchosen option). Crucially, we scored these behaviours after both the outcome of the chosen option and the outcome of the alternative, forgone option was revealed (complete feedback procedure). We tested the following two hypotheses and associated predictions.
Hypothesis 1: capuchins can flexibly modify their choices according to the probability associated with the risky option and their choices are influenced by their previous reward history (De Petrillo et al., 2015, 2017. Under this hypothesis, we expected capuchins, in the present study, to increase their preference for the risky option as the probability associated with obtaining the larger reward increased and to modify their choices according to the outcome of both the chosen and forgone options. Hypothesis 2: capuchins exhibit both disappointment-(De Petrillo et al., 2017) and regret-like responses when presented with risky choices. Under this hypothesis, we expected capuchins to exhibit more scratching behaviour and attempt to switch their initial choice after gambling and receiving a less favourable outcome than after gambling and receiving a good outcome or playing it safe. Moreover, we expected capuchins to exhibit more negative behavioural responses also when witnessing that the outcome of the forgone option was more favourable than the outcome of the chosen option. Hypothesis 3: humans update their choices in response to feeling regret, suggesting that regret might have an adaptive function. Because being able to adjust future choices according to hypothetical outcomes would be beneficial for other species as well (Kim et al., 2015), we hypothesized that experiencing disappointment-and regret-like emotions would influence capuchin monkeys' future choices. Under this hypothesis, we expected that both the behavioural responses shown after witnessing the choice outcome and those shown after witnessing the outcome of the forgone option would alter capuchins' subsequent choices.

Subjects and Housing
We tested 12 tufted capuchin monkeys (10 Sapajus cay: four males and six females; one Sapajus nigritus male; one Sapajus macrocephalus male; age range 9e33 years, mean ¼ 21.42; see Table A1 for detailed subject information) belonging to four social groups, hosted at the Primate Centre of the ISTC-CNR in Rome, Italy. Each group was housed in its own indooreoutdoor compartment, with the outdoor area measuring 53.2e374.0 m 3, depending on the group size, and the indoor area measuring 25.4 m 3 . All indooreoutdoor compartments were enriched by wooden perches, branches and tree trunks. All subjects were tested individually in the indoor compartments, between 0930 and 1330 hours. Capuchins had ad libitum access to water during testing and received their main meal (fresh fruits, vegetables and monkey chow) after the testing was concluded. Subjects completed no more than one test session per day. Eight capuchins had previously participated in a similar study involving risky choices (De Petrillo et al., 2015, 2017 ;  Table A1), while four were naïve to risky choice tasks.

Apparatus
We used a 'choice apparatus' to present the subjects with two choice options (Fig. 1). The apparatus consisted of a dark platform, measuring 65 Â 40 cm and 16 cm high, with two transparent boxes (9.5 Â 20 cm and 15 cm high), placed 28 cm apart. Each transparent box had a hole with a diameter of 2 cm on the subject's side: in all phases of the study, subjects could choose the preferred option by inserting their finger in the box hole.

Experimental Procedure
Each capuchin was tested in three different phases: numerical discrimination, familiarization and main experiment.
Numerical discrimination: to evaluate whether subjects could discriminate between the different food quantities involved in the main experiment, we presented them with numerical discrimination sessions of 26 trials each. In each session, capuchins completed six exposure trials (only one option available at a time: one, four or seven food items) and 20 binary-choice trials (in which they chose between either one versus four or four versus seven food items). In this phase, as well as in the other two phases of the study, each food item was an eighth of an entire peanut seed, a high-preferred food. Capuchins received three exposure trials at the beginning of the session to introduce the reward contingencies, while the three remaining exposure trials were interspersed within the session. There were 10 binary-choice trials for each type of comparison (one versus four and four versus seven). The lefteright position of the two comparisons was pseudorandomized, with the restriction that the same type of comparison could not be presented more than twice in a row within the same session. At the beginning of each exposure or binary-choice trial, the experimenter showed the option(s) to the subject for 1 s, covered it/them with the corresponding upside-down bowl(s) (see below, Familiarization and Main experiment) and then allowed the subject to choose. The numerical discrimination sessions were repeated until subjects, in Figure 1. Schematic illustration of the experimental set-up (Neutral condition). Capuchins chose between a safe, constant option covered by a white upside-down bowl (on the left) and a risky, variable option covered by a red upside-down bowl (on the right). The assignment of the two bowls to either the safe or risky option was counterbalanced across subjects. After capuchins made their choices, the outcomes of the chosen and forgone options were shown for 3 s each (see main text); then, the choice outcome was provided. the binary-choice trials, were able to select the larger option in at least eight of 10 trials for each type of comparison. We carried out the Numerical discrimination phase once, by using the bowls employed in the Neutral condition (see below, Familiarization and Main experiment). Capuchins completed a minimum of one session and a maximum of six sessions (mean ¼ 2.17, SE ¼ 0.44) before reaching the criterion and proceeding to the Familiarization phase.
Familiarization: once capuchins reached the criterion in the Numerical discrimination phase, they received one block of three familiarization sessions before each of the three conditions of the Main experiment (Neutral, Advantageous, Disadvantageous). In each condition, there was a pair of bowls, differing in colour and shape, each assigned to either the safe option (always four food items) or risky option (one or seven food items with a different probability, according to the experimental condition; see below); within each condition, the assignment of each bowl to the safe or risky option was counterbalanced across subjects. In each block of familiarization sessions, capuchins learned the association between each bowl and the amount of food corresponding to either the safe or risky option (four, one or seven food items), as well as the probability associated with the choice of the risky option (Neutral condition: same probability of receiving either one or seven food items; Advantageous condition: two-thirds probability of receiving seven food items and one-third probability of receiving one food item; Disadvantageous condition: two-thirds probability of receiving one food item and one-third probability of receiving seven food item). In each familiarization session, capuchins completed 20 exposure trials, where they were presented with a single option (one, four or seven food items, in a pseudorandomized order). In this phase, options were covered by their corresponding bowl, so that subjects could not see the number of food items associated with the option they were presented with.
Main experiment: each subject received three experimental conditions: Neutral, Advantageous and Disadvantageous. We carried out the Neutral condition first, whereas the order of presentation of the other two conditions was counterbalanced across subjects. In each condition, subjects had to choose between a safe, constant option (four food items) and a risky option (one or seven items that could be received with different probabilities according to the condition). The two options were covered by a pair of bowls, differing in colour and shape according to the condition; within each condition, the assignment of each bowl to the safe or risky option was counterbalanced across subjects. We used the same bowls and associated outcomes and probabilities as in De Petrillo et al. (2015,2017). In the Neutral condition, where subjects had the same probability of receiving either a seven-item or a one-item outcome upon choosing the risky option, we used a red, rectangular bowl and a white circular bowl to cover the two options. In the Advantageous condition, where subjects had a two-thirds probability of receiving a seven-item outcome and a one-third probability of receiving a one-item outcome upon choosing the risky option, we used an aluminium bowl shaped like a truncated cone and a green, irregular bowl. In the Disadvantageous condition, where subjects had a one-third probability of receiving a sevenitem outcome and a two-thirds probability of receiving a oneitem outcome upon choosing the risky option, we used a lightblue, circular bowl and a yellow bowl shaped like a truncated pyramid.
Each condition comprised 10 16-trial sessions: six forced-choice trials and 10 binary-choice trials. To remind the subjects of the association between each bowl and the corresponding food amount and probability, every session started with three forcedchoice trials, where only one option was available (safe or risky); the order in which the safe and the risky options were presented was alternated between sessions. The remaining three forced-choice trials were distributed throughout the session. In binarychoice trials, subjects could choose between the safe option and the risky option, without being allowed to see the number of food items associated with each option, which was covered by the corresponding bowl before subjects were provided with the opportunity to indicate their preference. Every trial began when the experimenter removed an opaque panel that separated the subject from the apparatus. As shown in Fig. 2, once the subject had made its choice, the experimenter removed the bowl covering the chosen option so that the subject could see its content for 3 s (Post choice phase; see Supplementary Videos S1 and S2); then, the experimenter also removed the bowl covering the forgone option, allowing the subject to see its content for 3 s (Post unchosen phase; see Supplementary Videos S3 and S4), and finally the subject received the food items associated with the chosen option (Reward phase). The intertrial interval was 30 s. If the subject did not indicate its preference within 15 s from the beginning of the trial, the experimenter put the panel back in place between the subject and the apparatus, waited for 5 s and repeated the trial. If the subject kept not choosing for three consecutive times, the session was invalidated and performed again on the subsequent day.

Ethical Note
This study complied with protocols approved by the Italian Health Ministry (n. 337/2017-PR to E. Addessi) and was performed in full accordance with the Directive 2010/63/EU on the protection of animals used for scientific purposes. Monkeys were familiar with the experimental enclosure and the testing apparatus. All tests were voluntary: capuchins could refuse to participate by not entering the inside enclosure and could stop participating at any time. In this case, the session was interrupted and resumed on the following day. All capuchins were watched closely while participating and at any sign of stress testing was ended and resumed on the following day.

Behavioural Coding and Data Analysis
In each trial, we scored the choices for the risky option. Moreover, in both the Post choice and Post unchosen phases, we scored all occurrences of scratching (defined as a repeated movement of the hand or foot during which the fingertips are drawn across an individual's fur) and of switching attempts (number of finger insertions in the hole of the unchosen box, De Petrillo et al., 2017).
To assess reliability, two experimenters independently coded 25% of the video clips. The index of concordance was > 96% for all measurements. We could not score scratching and switching in 33 of 3600 trials because video recordings were not available.
To analyse trial-by-trial binary responses, we implemented generalized linear mixed models (GLMM) in R v. 4.0.24 (R Core Team, 2020) using the glmer function from the lme4 package (Bates et al., 2015). Models always included random subject intercepts (to account for repeated measures), sex (male and female), previous experience (yes or no; to account for each individual's experience in a previous risky-choice task;De Petrillo et al., 2015), trial number (within condition) and session number. We then added our main predictors of interest to subsequent models to test their importance, comparing model fit using likelihood ratio tests (Bolker et al., 2009) and additionally the Akaike information criterion (AIC), where lower AIC scores indicate better fit (Bolker et al., 2009;Burnham & Anderson, 2002). Across these models, the dependent variable was capuchins' choice (binary measure: 1¼ choice for the risky option, 0¼choice for the safe option) or behavioural response (number of scratching or switching episodes). Post hoc pairwise comparisons of model predictors were computed using the emmeans package (Lenth et al., 2019) with a Tukey correction.

Hypothesis 1: Capuchins' Risky Choice and Previous Reward History
Overall, capuchins behaved rationally, as they chose the risky option above chance only in the Advantageous condition (onesample t test: 0.81 ± 0.06, t 11 ¼ 5.253, P < 0.001; Fig. 3a), while in the Neutral condition they did not show a significant preference for either option (0.46 ± 0.06, t 11 ¼ e0.631, P ¼ 0.54) and in the Disadvantageous condition they chose the risky option below chance (0.31 ± 0.05, t 11 ¼ e3.656, P ¼ 0.004). To analyse capuchins' trial-by-trial choice, we implemented GLMMs with a binomial distribution (see Behavioural Coding and Data Analysis section). The best model for capuchins' risky choices included sex, previous experience, session number, trial number, condition, chosen option outcome and the interaction between the latter two variables (model 6, Table 1). This model showed that capuchins became more risk prone over time: their choice for the risky option increased over sessions. Post hoc comparisons also showed that capuchins chose the risky option more often in the Advantageous condition than in the Neutral or Disadvantageous conditions, and more often in the Neutral than in the Disadvantageous condition (Tukey P value adjustment; P < 0.001 for all significant comparisons; Fig. 3a).
Moreover, capuchins chose the risky option more often after choosing a risky option (i.e. after receiving either one or seven food items) than after choosing a safe option in all conditions (P < 0.05 for both the one versus four and the seven versus four comparisons; Fig. 3b and see Table A2 for parameters of this best model). We did not find a significant effect of sex, age or previous experience on capuchins' choice for the risky option.

Scratching behaviour
Capuchins performed a similar number of scratching episodes after seeing both the chosen and the forgone outcomes (paired Wilcoxon test: P ¼ 0.29; Fig. 4a and b). To analyse capuchins' scratching behaviour, we implemented GLMMs with a Poisson distribution with the number of scratching episodes as the dependent variable.
First, we examined capuchins' disappointment-like responses by analysing scratching behaviour immediately after the outcome of the chosen option was revealed (Post choice scratching). The best model for scratching episodes included sex, previous experience, session number, trial number and condition (model 2, Table 2 and  see Table A3 for parameters of this model). This model showed an effect of experience on capuchins' scratching behaviour: capuchins with experience with the risky-choice task (De Petrillo et al., 2015) showed more scratching episodes than capuchins with no  experience with the task. Capuchins' scratching episodes also increased across sessions. Finally, post hoc comparisons between conditions showed that capuchins performed more scratching in the Advantageous and Disadvantageous conditions than in the Neutral condition (P < 0.05 for all significant comparisons; Fig. 4a). Then, we examined capuchins' regret-like responses by analysing scratching behaviour after the outcome of the forgone option was revealed (Post unchosen scratching). The best model included only sex, previous experience, session number and trial number and again showed an effect of experience, with capuchins already experienced in the risky-choice task exhibiting more scratching behaviour than not-experienced capuchins (model 4, Table 2).

Switching attempts
Capuchins performed more switching attempts to modify their initial choices after seeing the outcome of the forgone option than after seeing the outcome of the chosen option (paired Wilcoxon test: P ¼ 0.016; Fig. 4c and d). To analyse capuchins' switching behaviour, we implemented Poisson GLMMs with the number of switching attempts as the dependent variable.
First, we examined capuchins' disappointment-like responses by looking at switching attempts after the outcome of their choice was revealed (Post choice switching). The best model for switching attempts included previous experience, sex, session number, trial number and the interaction outcome*condition (model 4, Table 3). This model revealed a significant effect of previous experience on capuchins' switching behaviour after their choice outcome was revealed: capuchins with no previous experience performed more switching attempts than the more experienced individuals. Moreover, we found a significant effect of the interaction between condition and outcome: while for all comparisons presented in the Advantageous condition capuchins performed more switching attempts after receiving a worse than a better outcome (one versus seven, one versus four and four versus seven, P < 0.05), in the Neutral and Disadvantageous conditions they did so only when they received one item rather than four or seven items (P < 0.05 for all significant comparisons; Fig. 4c and see Table A4 for parameters of the best-fitting model).
Then, we examined capuchins' regret-like responses by looking at the switching attempts after the outcome of the forgone option was revealed (Post unchosen switching). Because we found an unexpected effect of sex in one of our models, we included an additional interaction to further explore this effect. Specifically, we included an interaction between sex and condition to explore whether this effect was present in all or some conditions. The best model included session number, trial number and the interactions outcome of the forgone option*condition and sex*condition (model 9, Table 3  capuchins with no experience in the risky-choice task performed more switching attempts than those with previous experience. Post hoc comparisons revealed that in all conditions capuchins attempted to switch their choice more often when the outcome of the forgone option was better than their choice outcome (i.e. when the outcomes of the forgone/chosen options were seven/four and four/one, respectively) than when the outcome of the forgone option was worse than their choice outcome (i.e. when the outcomes of the forgone/chosen options were four/seven and one/four, respectively; P < 0.05 for all significant comparisons; Fig. 4d and see Table A5 for parameters of the model). From our explorative analysis, we found that in the Neutral and Disadvantageous conditions males performed more switching attempts than females (P < 0.05). Finally, we found no significant correlation between scratching and switching attempts in response to the choice outcome (Spearman correlation: rho ¼ e0.49, P ¼ 0.104). We found a significant negative correlation between scratching and switching attempts in response to the outcome of the forgone option (Spearman correlation: rho ¼ e0.72, P ¼ 0.008).

Hypothesis 3: Capuchins' Risky Choice and Emotional Response
To examine how capuchins' behavioural responses affected their future choices, in a series of models we added to the previous bestfitting model describing capuchins' risky choices (see Hypothesis 1: Capuchins' risky choice and previous reward history), which included previous experience, sex, subject (as a random factor), session number, trial number and choice outcome*condition, the number of scratching episodes and switching attempts performed when both the outcomes of the chosen and forgone options were revealed. We also explored the interaction between the significant predictors. The best model included previous experience, sex, session number, trial number, choice outcome*condition, post choice scratching and post unchosen scratching, post choice switching*condition and post unchosen switching*condition (model 4, Table 4). This model revealed a significant effect of sessions and trials: capuchins' choices for the risky option increased across trials and over sessions. Moreover, how capuchins' switching attempts were related to their choice for the risky option depended on the condition: switching attempts after the outcome of the chosen option was revealed were positively related to capuchins' risky choices in both Neutral and Disadvantageous conditions ( Fig. 5aec and see Table A6 for parameters of this best-fitting model), whereas switching attempts in response to the outcome  of the forgone option were positively related to capuchins' risky choices only in the Disadvantageous condition (see Fig. 5def).

DISCUSSION
In this study, we presented capuchin monkeys with a riskychoice task in which the probability of obtaining the larger payout varied across conditions. Overall, our results showed that capuchins flexibly adjusted their behaviour: they preferred the risky option in the Advantageous condition and the safe option in the Disadvantageous condition, and were indifferent between options in the Neutral condition. Additionally, they chose the risky option more in the Advantageous condition than in the Neutral and Disadvantageous conditions and more in the Neutral condition than in the Disadvantageous condition. Thus, capuchins appear to have fully understood the contingencies of the task. Interestingly, they showed a more economically rational behaviour than in a previous study in which we presented the same conditions and stimuli, but employing a partial feedback procedure (i.e. revealing only the outcome of the chosen option; De Petrillo et al., 2015). In the latter, capuchins significantly preferred the risky option in both the Advantageous and Neutral conditions and chose the two options to a similar extent in the Disadvantageous condition. It is unlikely that the discrepancy between the two studies was a mere effect of experience with the main features of the task, as previous experience (having previously participated in De Petrillo et al.'s (2015) study) did not seem to affect capuchins' risky choices. Rather, it is possible that the complete feedback procedure employed in the current study, in which both the choice outcome and the outcome of the forgone option were revealed, allowed a more accurate learning of the contingencies of the task than when only the choice outcome was revealed. Indeed, individuals can modify their behaviour not only on the basis of the actual consequences of their actions, but also by witnessing the fictive outcomes that could have resulted from alternative actions. Several findings in both human and nonhuman animals support such fictive learning, that is, the ability to adjust future choices on the basis of both actual and hypothetical outcomes (Abe & Lee, 2011;Coricelli & Rustichini, 2010;Epstude & Roese, 2008;Hayden et al., 2009;Steiner & Redish, 2014), and highlight how counterfactual reasoning can drive long-term changes in decision strategies to avoid regret (Coricelli et al., 2005;Sweis et al., 2018). Together, these findings suggest that the emotions that arise from counterfactual comparisons do not necessarily interfere with rational decision making, but they may implement it, as they are a way of evaluating past outcomes to adjust future choices. From this perspective, it is plausible that capuchins, when provided with the opportunity to gain information on both the chosen and forgone options, relied on counterfactual reasoning to facilitate the learning process of the outcome structure of the task, which in turn led them to make more rational choices.
Our study also revealed that, in all conditions, capuchins attempted to modify their initially chosen option more often after witnessing the outcome of choices with unfavourable than with positive consequences, that is, when they received the lowest amount of food (one food item) compared to the highest amount of food (seven food items) or to the safe option (four food items), and when they received the safe option compared to the highest amount of food (although the latter finding was limited to the Advantageous condition). These results mirror both our previous findings, obtained when employing the same task with a partial feedback procedure (De Petrillo et al., 2015, 2017, and similar findings obtained in chimpanzees and bonobos (Rosati & Hare, 2013). Thus, these studies confirm that capuchins experience disappointment-like behaviour when receiving an unfavourable choice outcome. Interestingly, regardless of condition and choice outcome, post choice switching occurred more frequently in capuchins without any previous experience in risky-choice tasks rather than in experienced subjects. This was probably because experienced individuals had previously learned that attempting to switch after selecting an option did not result in any consequence. However, given our small sample size, these findings need further investigation.
Capuchins also attempted to modify their initially chosen option more often after witnessing that the outcome of the forgone option was better, rather than worse, than the outcome of the chosen option. Thus, capuchins apparently showed a behavioural response compatible with regret for the choice they made. In humans, regret emerges relatively late in the course of development (at around 6 years of age, McCormack et al., 2020), whereas in other animal species regret-like responses have so far been demonstrated only in constrained laboratory environments, where the subjects underwent food or fluid restrictions to increase their motivation, which may have increased their sensitivity to potential reward losses and subsequent behavioural and emotional responses (Hayden et al., 2009;Lee et al., 2005;Steiner & Redish, 2014;Sweis et al., 2018). This is the first time that a behaviour compatible with regret has been observed in a gambling task in a zoo population.
Capuchins' switching attempts after witnessing the outcome of the forgone option were generally more frequent in the Neutral condition than in the Disadvantageous and Advantageous conditions, possibly because this condition was presented first and thus capuchins were more reactive to the novel, complete feedback procedure, and not yet aware of the lack of consequences of switching attempts. Indeed, regardless of condition, post unchosen switching behaviour was more frequent in capuchins without any previous experience in risky-choice tasks than in experienced subjects, as also reported above for post choice switching. Again, it Model selection for the variable 'switching attempts': the table reports the likelihood ratio (LRT) tests for the comparison between models and the Akaike information criterion (AIC) for each model. Models 1e4 have post choice switching as response variable and subjects as random effects. Models 5e9 have post unchosen switching as response variable and subjects as random effects. Bold indicates significance (P < 0.05) and the lower AIC.
is likely that experienced individuals had previously learned that switching attempts do not result in any consequence and extended the acquired information to the novel procedure employed in the present study. Moreover, regardless of condition, switching attempts were more frequent when the outcomes of the chosen/ forgone options were one versus four rather than four versus seven. This may be due to a higher aversiveness of the one food item option in comparison to the four food items option or, more parsimoniously, to a better capability of capuchins to discriminate between one and four than between four and seven (Addessi et al., 2008). Interestingly, our explorative analysis revealed that in the Neutral and Disadvantageous conditions males performed more switching attempts than females after witnessing the outcome of the forgone option. This suggests that males were more sensitive than females to the experience of making a disadvantageous choice, especially in the two conditions in which an unfavourable outcome was most likely to occur. Previous studies on risky decision making in nonhuman primates did not find a difference in risk preferences between males and females (De Petrillo et al., 2015; see De Petrillo & Rosati, 2021 for a review); therefore, we did not expect to find an effect in this study. One possibility is that differences in feeding ecology between males and females observed in the wild might explain this finding, as it has been proposed to explain capuchins' sex difference in self-control abilities (Addessi et al., 2011). While capuchin males in the wild forage more on small vertebrates and large invertebrates on the ground, thus hunting in the open without external protection, females prefer to forage on small invertebrates a few metres above the ground, so benefitting from more protection (Cebus olivaceus, Fragaszy 1986; Sapajus libidinosus, Verderane, 2010). This suggests that males are more exposed to risks than females and might benefit more from a mechanism that allows them to quickly react to an unfavourable outcome and adapt their future behaviour. To our knowledge, ours is the first evidence of a sex difference in regret-like behaviours in nonhuman animals, which has possibly never previously been observed due to the small sample sizes employed in previous studies. In the few studies in which sex differences in regret have been observed in humans, an opposite pattern has been found, with women being more prone to regret than men in the context of sexual decisions (Galperin et al., 2013;Kennair et al., 2018). Future research should further investigate this aspect in larger samples of nonhuman animals belonging to multiple species, to possibly elucidate its relationship with ecological and/or social species-specific features.
Although in the nonhuman literature scratching behaviour is usually related to negative emotional states (Maestripieri et al., 1992;Troisi, 2002), in the present study, surprisingly, scratching episodes did not differ significantly either after making choices with unfavourable, rather than positive, consequences or after witnessing that the outcome of the forgone option would have been better, rather than worse, than that of the chosen option. We believe it is unlikely that capuchins were not sensitive to unfavourable choice outcomes. Indeed, in our previous study (De Petrillo et al., 2017), in which we employed a 10 s time window to score capuchins' behavioural responses, they showed a significantly higher frequency of scratching after witnessing a negative choice outcome (i.e. one food item) rather than a positive outcome (seven food item) or a safe outcome (four food items), thus showing a behaviour compatible with disappointment (for a similar finding in great apes see Rosati & Hare, 2013). Rather, we hypothesize that, due to the different structure of the task and to the shorter time window available to score their behavioural responses (3 s), in the present study capuchins just had less time to produce scratching behaviour, as also evidenced by the significant, negative relationship between scratching and switching attempts, which emerged in the Post unchosen phase, where capuchins performed switching attempts most frequently.
Indeed, post choice and post unchosen scratching showed an exactly opposite pattern to that observed for switching behaviour (detailed above). Both after witnessing the outcome of their choice and the alternative, unchosen outcome, experienced subjects scratched more than those without any previous experience in risky-choice tasks. Moreover, post choice scratching was more frequent in the two conditions presented later (Advantageous and Disadvantageous) than in the condition presented first (Neutral).
Thus, it appears that, as capuchins become aware of the lack of consequences of switching attempts (by accumulating experience in the risky-choice task), they reduce their switching attempts but intensify their scratching behaviour. The parallel increase of both post choice scratching behaviour and risky choices across sessions suggests a potential role of scratching as a displacement activity performed to mitigate the anxiety associated with the selection of the risky option, which, regardless of condition, may imply a loss.
Overall, in our study, capuchins exhibited disappointment-like behaviour for receiving an unfavourable outcome and regret-like behaviour for not having made a better choice. Moreover, regretlike displays were more intense than disappointment-like displays (Zeelenberg et al., 2000;Camille et al., 2004). In fact, capuchins performed more switching attempts after witnessing that the outcome of the forgone option was better than the outcome of the chosen option than after witnessing just an unfavourable choice outcome. Similar findings have been reported for humans, where although both regret and disappointment activate anterior insula and dorsomedial prefrontal cortex, the activation is more intense when experiencing regret than disappointment (Chua et al., 2009).
Finally, we evaluated whether disappointment-and regret-like responses were associated with capuchins' decision making. Risky choices were positively related to the occurrence of switching behaviour (but not scratching behaviour) both after witnessing the previous choice outcome (in the Neutral and Disadvantageous conditions) and after witnessing the previous alternative outcome (only in the Disadvantageous condition). Thus, the experience of Model selection for the variable 'risky choice': the table reports the likelihood ratio (LRT) tests for the comparison between models and the Akaike information criterion (AIC) for each model. All models have risky choices as response variable and subjects as random effects. Bold indicates significance (P < 0.05) and the lower AIC.
disappointment-and regret-like emotions did influence capuchins' choice strategies, leading to an increase in risk proneness. In humans, the anticipation of regret may promote risk seeking (e.g. Zeelenberg & Beattie 1996), risk-averse decision making (e.g. Coricelli et al., 2005) or a lack of willingness to reconsider the previous choice (Habib et al., 2012). When reconsidering the previous choice, people can become either risk averse or risk seeking, depending on whether the risky or safe gamble is the regretminimizing choice, as they attempt to avoid postdecisional regret (Zeelenberg et al., 1996;Zeelenberg, 1999). In capuchin monkeys, however, the observed increase in risk proneness following regretlike behaviour potentially involved a loss and thus another probable regretful experience. However, it should be considered that our experimental setting was a relatively low-cost situation, as capuchins did not obtain their food ration during the experiment (as for instance the mice, Mus musculus, tested by Sweis et al., 2018). Thus, being in an average positive energetic state, capuchins could accept the risk to frequently select the risky option, even if this course of action would result in a frequent loss of the food reward (for comparable findings in wild chimpanzees, see Gilby & Wrangham, 2007). Capuchins' behaviour may be due to the flexible and opportunistic feeding ecology that characterizes the genus Sapajus, including the species S. cay (Smith et al., 2022), the most represented in our sample. In the wild, capuchins often exploit unpredictable and/or potentially hazardous food sources (Fragaszy et al., 2004), occasionally hunt small vertebrate prey (S. cay, Murillo Peixoto Couto et al., 2021;S. libidinosus, Falotico et al., 2018), an activity involving a high energetic cost and a high risk of not being successful, and process cashew nuts to access the nutritious kernel notwithstanding the risk of being burned by caustic chemicals (S. libidinosus, Sirianni & Visalberghi, 2013). Although our capuchins were indifferent between options in the Neutral condition, they showed a 'win-stay'/'lose-stay' attitude, according to which they chose the risky option more frequently after choosing a risky option in the previous trial, regardless of its outcome (see also De Petrillo et al., 2015). This is in contrast with previous studies on risky decision making in different colonies of capuchin monkeys which found, instead, a decreasing attraction to risk (Rivi ere et al., 2019;Roig et al., 2022). An important question for future research is therefore to investigate whether differences across colonies may be due either to different experimental set-ups or to different experiences during their lifetime (De Petrillo & Rosati, 2021).
In summary, this study indicates that capuchin monkeys make more rational choices when they are provided with information about the outcome of the forgone option in addition to the outcome of the chosen option. This suggests that the ability to adjust future choices on the basis of both actual and hypothetical outcomes (fictive learning) is widespread across nonhuman species, as it helps individuals make adaptive choices by reducing the costs associated with directly experiencing all possible outcomes. Our work not only strengthens the finding that capuchins, like other animal species, show disappointment-like behaviour for receiving an unfavourable outcome, but also demonstrates, for the first time in a zoo population, the occurrence of regret-like behaviour for not having chosen differently. Moreover, we observed, for the first time in nonhuman animals, that males were more sensitive than females to witnessing that the outcome of the forgone option would have been better than the outcome of the chosen option. Importantly, gender differences in human risky decision making might stem from diverse causes, including cultural learning about gender roles (Croson & Gneezy, 2009 Finally, in our low-cost experimental setting, capuchins increased their preference for the risky options after showing regret-like behaviour for the previous choice, as also shown in rhesus macaques and in some human studies. Further research is needed to address whether this behaviour is prone to the contextual effects of the experimental setting or whether it is a species-specific behavioural trait.

Author Contributions
F. De Petrillo, E. Addessi conceptualized the study; F. Rossi and S. Gastaldi collected the data; F. Rossi and S. Gastaldi coded the data; F. De Petrillo and E. Addessi analysed the data; F. De Petrillo, F. Rossi and E. Addessi wrote the manuscript with input from all authors.

Declaration of Interest
None. ± 0.01 0.01 ± 0.01 0.02 ± 0.01 0.04 ± 0.02 M: male; F: female. Data are reported for each condition. Choice ¼ mean proportion of choices for the risky option ± SE (probability of obtaining seven food items is 0.5 in the Neutral condition, 0.7 in the Advantageous condition and 0.3 in the Disadvantageous condition); PCscrat ¼ mean number of scratching episodes ± SE when the outcome of the chosen option is revealed; PUscrat ¼ mean number of scratching episodes ± SE when the outcome of the option capuchins did not choose is revealed; PCswit ¼ mean number of switching attempts ± SE when the outcome of the chosen option is revealed; PUswit ¼ mean number of switching attempts ± SE when the outcome of the option capuchins did not choose is revealed.