Psychopathic individuals exhibit but do not avoid regret during counterfactual decision making

Significance

Psychopathic individuals display a chronic and flagrant disregard for social norms through their callous behavior and lack of regret for its consequences. Although psychopathy research largely attributes this to deficits in affective responsiveness, recent proposals suggest that value-based decision making may also contribute to the maladaptive behavior of psychopathic individuals. Using a counterfactual decision-making paradigm, we found that higher scores on psychopathy were associated with higher levels of retrospective regret. Despite this, however, individuals higher on psychopathy made riskier choices and were less influenced by prospective regret when making decisions. These findings support the idea that the maladaptive behavior of psychopathic individuals is related to deficits in domain-general cognitive processes, such as counterfactual decision making, rather than a primary affective deficit.

Abstract

Psychopathy is associated with persistent antisocial behavior and a striking lack of regret for the consequences of that behavior. Although explanatory models for psychopathy have largely focused on deficits in affective responsiveness, recent work indicates that aberrant value-based decision making may also play a role. On that basis, some have suggested that psychopathic individuals may be unable to effectively use prospective simulations to update action value estimates during cost–benefit decision making. However, the specific mechanisms linking valuation, affective deficits, and maladaptive decision making in psychopathy remain unclear. Using a counterfactual decision-making paradigm, we found that individuals who scored high on a measure of psychopathy were as or more likely than individuals low on psychopathy to report negative affect in response to regret-inducing counterfactual outcomes. However, despite exhibiting intact affective regret sensitivity, they did not use prospective regret signals to guide choice behavior. In turn, diminished behavioral regret sensitivity predicted a higher number of prior incarcerations, and moderated the relationship between psychopathy and incarceration history. These findings raise the possibility that maladaptive decision making in psychopathic individuals is not a consequence of their inability to generate or experience negative emotions. Rather, antisocial behavior in psychopathy may be driven by a deficit in the generation of forward models that integrate information about rules, costs, and goals with stimulus value representations to promote adaptive behavior.

The ability to establish, transmit, and enforce social norms is a signature of our species. Indeed, maintaining our uniquely high degree of stable, large-scale cooperation requires widespread norm compliance (1). However, although norm compliance is common, it is far from universal. Throughout history and across cultures, there have been those who would threaten social peace and community prosperity through their persistent violation of social norms. Psychopathic individuals, who exhibit a chronic and flagrant disregard for moral and legal norms, exemplify this type of person. Compared with nonpsychopathic individuals, they commit two to three times more violent and nonviolent crime and recidivate at a much higher rate (2). This persistent antisocial behavior comes at a high cost to society, with psychopathic individuals responsible for a disproportionate share of the estimated $2.34 trillion in annual costs associated with crime in the United States (3).

Psychopathy is defined by a combination of superficial charm, blunted empathy and punishment sensitivity, shallow emotional experiences, persistent antisocial behavior, and marked sensation seeking and impulsivity (2). Whereas many of the behavioral and lifestyle features of this disorder (e.g., sensation seeking, criminal offending) are shared with other antisocial subtypes, psychopathy is distinguished by the presence of deficits in emotional arousal, empathy, and affective responsiveness (2, 4). The behavioral manifestations of such deficits in psychopathic individuals are diverse, encompassing pathological lying, interpersonal manipulation, and the absence of guilt, remorse, and regret following decisions that cause harm to themselves or others. Such symptoms are considered by many to be defining features of the disorder (4); however, the cognitive and neurobiological mechanisms that produce them remain the subject of debate.

Dominant explanatory models of psychopathy attribute such symptoms to core deficits in emotion processing that prevent psychopathic individuals from generating negative affect responses to aversive stimuli, and that limit their capacity for empathic experience sharing with others (5). Consistent with this suggestion, psychopathic individuals show deficits in fear conditioning (6, 7), face emotion processing (8), and emotion-modulated startle (9). Additionally, psychopathic individuals show reduced functional and structural connectivity between amygdala and medial orbitofrontal cortex (mOFC) (10), accompanied by blunted corticolimbic engagement during moral decision making (11), aversive conditioning (6, 7), affective perspective taking (12), and in response to empathogenic (13) and facial emotion stimuli (14). Moreover, this association between corticolimbic dysfunction and psychopathy appears largely to be driven by interpersonal-affective symptoms rather than by antisocial-lifestyle features (13–15). Taken together, this work supports a model in which the affective deficits so central to psychopathy arise from dysfunction within brain networks that support the generation and evaluation of emotional states, and that link such states to social cues through associative mechanisms.

Although psychopathy research largely has focused on the basic social and affective processes detailed above, recent work highlights a potentially significant role for value-based decision making as well (16). For example, psychopathic individuals exhibit heightened ventral striatal responses to reward (17), as well as increased striatal gray matter volume (18). These data, considered in light of the mOFC findings above, have led some to hypothesize that many of the most problematic behaviors in psychopathy result from a deficit in the ability to represent and integrate information about the costs and benefits of actions (16, 19). According to this view, psychopathy symptoms that are apparently affective in nature—such as the absence of guilt, remorse, and regret—may instead arise from dysfunction within domain-general valuation systems. The experience of regret—characteristically diminished or absent in psychopathic individuals—provides a particularly instructive example for considering how affective symptoms in psychopathy could arise from aberrant value-based decision making.

According to prominent accounts, psychopathic individuals “never look back with regret or forward with concern” (20). What is regret, precisely? Decision science situates regret in the context of counterfactual reasoning and offers a useful operationalization: an aversive emotional state that is elicited by a discrepancy in the outcome values of chosen vs. unchosen actions. Put simply, the experience of regret is triggered when an agent is informed that the outcome of their choice is worse than what they would have obtained had they chosen differently (21). A wealth of evidence suggests that people are generally regret-avoidant (22). When faced with a multioption choice problem, decision makers estimate the likelihood of experiencing regret for each option in the choice set and, all other things being equal, select the one with the lowest anticipated regret (23, 24). This process requires the ability to generate and compare outcome value representations for both chosen and unchosen actions; therefore, counterfactual reasoning is thought to be fundamental to regret-sensitive decision making (22). Before action selection, counterfactual processes generate a forward model of action–outcome relationships by prospectively simulating outcome values for each choice option. At feedback, retrospective counterfactual comparisons signal the difference between outcomes for the chosen vs. unchosen action; the aversive state of regret is triggered when the counterfactual outcome is better than the obtained outcome.

Notably, counterfactual thinking and regret engage strikingly similar neural circuitry (25, 26). The strongest overlap is in mOFC, a region where structural and functional alterations are consistently found in psychopathy. mOFC damage produces a syndrome that includes social and affective symptoms similar to those seen in psychopathy (27). Recent work suggests that the presence of such symptoms in mOFC patients may be due to lesion-induced alterations in the representation of information—including counterfactual signals—during value-based decision making (28–30). Together, these studies raise the possibility that diminished regret in psychopathy may result from a deficit in the ability to generate forward action–outcome models and/or perform retrospective counterfactual comparisons. To date, however, the use of counterfactual information during decision making in psychopathy remains unclear.

To examine the relationships between psychopathy, counterfactual decision making, and regret, we administered a counterfactual decision-making paradigm in a community-based sample that was significantly enriched for antisocial behavior. All participants received a clinical battery that assessed psychopathy and other antisocial trait subtypes. A marker of “real-world” antisocial behavior (prior incarcerations) was obtained for each participant. We measured affective responses to regret-inducing counterfactual outcomes, as well as behavioral sensitivity to prospective regret signals, which were tested for association with clinical and real-world indices of antisocial behavior.

Materials and Methods

Participants.

Participants included 62 male adults (18–55) recruited through flyers soliciting risk-taking (e.g., crime, substance use, gambling, impulsive behavior, bullying) individuals in New Haven County, Connecticut, a high-crime region (see Table S1 for sample characteristics). Participants who performed below the fourth-grade level on a standardized measure of reading, had an IQ of <70, or met criteria for psychotic disorders were excluded (see SI Materials and Methods for detailed exclusion criteria). Participants earned $10/h (regardless of performance on the task) for their completion of the self-report measures and the experimental task. All participants provided written informed consent according to the procedures set forth by the Yale University Human Subjects Committee.

Table S1.

Sample characteristics

Characteristic	N	Min	Max	M	SD
Age	62	19	55	38.65	11.34
Sex (male)	62
Race
White	23
Black	35
Biracial	4
Highest level of education		2	8	3.94	1.36
1. Grade 6 and below	0
2. Grade 7–12	6
3. Graduated high school or equivalent	18
4. Some college	27
5. Graduated 2-y college	2
6. Graduated 4-y college	5
7. Some graduate/professional school	2
8. Completed graduate/professional school	2
Annual Income		1	5	1.81	1.32
1. $0–$15,000	38
2. $15,001–$30,000	13
3. $30,001–$45,000	3
4. $55,001–$60,000	1
5. $60,001+	7
Employment status		1	6	2.68	1.38
1. Full time	15
2. Part time	9
3. Unemployed	30
4. Retired	0
5. Disability	4
6. Full-time student	4
Number of arrests		0	40	4.19	6.61
0	16
1–5	31
6–10	9
11+	6
Number of incarcerations		0	11	0.94	1.71
0	30
1–5	30
6–10	1
11+	1
IQ	62	85	125	106.85	10.80
SRP-III total	62	100	236	163.35	28.83

Measures.

Clinical assessment.

We used the Self-Report Psychopathy-III (SRP-III) scale to measure psychopathy (31). Notably, the SRP-III is sensitive to aspects of behavior that are common to multiple antisocial subtypes (e.g., criminal behavior, sensation seeking, impulsivity) (32, 33). Therefore, to assess the degree to which psychopathy per se was associated with affective response modulo these general aspects, we used participants’ scores on the Externalizing Spectrum Inventory-Brief (ESI-Brief) (34) as a covariate (see SI Materials and Methods for more information and Table S2 for zero-order correlations among assessments).

Table S2.

Zero-order correlations among task variables, key individual difference variables, and covariates

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Choice	1	0.073**	0.006	0.384**	0.323**	−0.131**	0.032*	0.008	−0.026	0.082**	−0.025	0.071**	−0.017	0.005
2. Rating 1 (partial)		1	0.535**	0.137**	−.065**	−0.077**	0.002	−0.025	−0.051**	0.065**	0.015	0.056**	−0.031*	−0.069**
3. Rating 2 (complete)			1	0.016	−0.059**	−0.007	0.040**	0	0.009	0.034**	−0.014	0.048**	−0.025	−0.047**
4. Regret (R)				1	0.331**	−.607**	0	0	0	0	0	0	0	0
5. Expected value (E)					1	0.159**	0	0	0	0	0	0	0	0
6. Disappointment (D)						1	0	0	0	0	0	0	0	0
7. SRP-III total score							1	0.750**	−0.140**	0.082**	−0.393**	0.333*	0.214**	0.224
8. ESI total score								1	−0.024	0.041**	−0.288**	0.203**	0.382**	0.146**
9. Age									1	−0.008	−0.051**	−0.043**	−0.175**	−0.050**
10. Education, y										1	−0.143**	0.513**	−0.158**	0.166**
11. Race/ethnicity											1	−0.244**	−0.097**	−0.044**
12. IQ (Shipley)												1	−0.476**	0.264**
13. Trait anxiety (STAI)													1	0.298**
14. Substance use, y														1

Note: Race/ethnicity was recoded into a dichotomous variable with White (1) and Non-White (0). *P < 0.05, **P < 0.01.

Criminal behavior.

All participants were asked how many times they had been incarcerated. This self-report was confirmed using the State of Connecticut Department of Correction inmate database. Approximately 51% of the sample had been incarcerated at least once before participation.

Counterfactual gambling task.

The present task was based on paradigms previously reported in the literature by Gillan et al. (35) and Camille et al. (23). The goal of the task was to earn as many points as possible. On each of 80 trials, participants had to choose between two wheels (gambles) that differed in expected value (Table S3). Each wheel offered two of the four possible outcomes: −210, −70, +70, and +210; respective probabilities were indicated by the proportion of the wheel occupied by a given outcome (0.25, 0.5, and 0.75). After a brief viewing period, participants selected one wheel via button press. On 50% of the trials, participants had the chance to change their mind and switch wheels before proceeding. Once the final selection was made, a red ball began to move within each wheel. After 1.5 s, the ball stopped on one of the sections of the selected wheel (the unselected wheel was occluded), indicating participants’ obtained outcome for that trial. Participants then rated their affective response to that outcome using a rating bar that ranged from “Very Disappointed” (0) to “Neither Pleased Nor Disappointed” (50) “Very Pleased” (100) (Fig. 1). The outcome of the nonselected wheel was then revealed and participants made another affective rating. At the end of each trial, the participant’s cumulative score was presented onscreen.

Table S3.

List of trials

	Wheel 1					Wheel 2
Trial	x1	p	y1	1 − p	Outcome	x2	q	y2	1 − q	Outcome
1	210	0.50	−70	0.50	210	70	0.75	−70	0.25	−70
2	−70	0.75	−210	0.25	−210	210	0.25	−210	0.75	−210
3	70	0.75	−210	0.25	70	210	0.25	−70	0.75	−70
4	70	0.50	−70	0.50	70	210	0.50	−210	0.50	210
5	210	0.25	−70	0.75	−70	210	0.50	−210	0.50	−210
6	70	0.50	−210	0.50	70	70	0.25	−70	0.75	−70
7	70	0.75	−70	0.25	70	210	0.50	−70	0.50	210
8	70	0.50	−70	0.50	−70	210	0.50	−210	0.50	210
9	210	0.25	70	0.75	210	210	0.50	−70	0.50	−70
10	210	0.50	70	0.50	210	210	0.75	−210	0.25	210
11	70	0.75	−210	0.25	70	210	0.50	−210	0.50	−210
12	70	0.75	−210	0.25	70	210	0.50	−210	0.50	−210
13	210	0.25	−70	0.75	210	70	0.50	−70	0.50	−70
14	70	0.75	−70	0.25	−70	210	0.25	70	0.75	70
15	−70	0.50	−210	0.50	−210	210	0.25	−210	0.75	210
16	210	0.25	70	0.75	210	210	0.50	−70	0.50	210
17	210	0.25	−210	0.75	210	−70	0.75	−210	0.25	−70
18	−70	0.50	−210	0.50	−210	70	0.25	−210	0.75	70
19	−70	0.50	−210	0.50	−210	70	0.25	−210	0.75	−210
20	70	0.75	−210	0.25	−210	210	0.25	70	0.75	210
21	−70	0.75	−210	0.25	−70	210	0.25	−210	0.75	−210
22	210	0.50	70	0.50	70	210	0.75	−210	0.25	210
23	70	0.50	−210	0.50	−210	70	0.25	−70	0.75	−70
24	70	0.50	−70	0.50	70	70	0.75	−210	0.25	−210
25	70	0.75	−70	0.25	70	70	0.75	−70	0.25	−70
26	210	0.50	−210	0.50	−210	210	0.25	−210	0.75	210
27	210	0.25	−70	0.75	−70	70	0.50	−70	0.50	70
28	70	0.50	−70	0.50	70	70	0.75	−210	0.25	70
29	210	0.25	70	0.75	70	210	0.75	−210	0.25	210
30	210	0.50	70	0.50	210	210	0.75	−70	0.25	210
31	70	0.75	−210	0.25	70	210	0.25	−70	0.75	−70
32	70	0.50	−70	0.50	70	210	0.25	−70	0.75	−70
33	210	0.75	−210	0.25	210	70	0.25	−210	0.75	−210
34	210	0.50	70	0.50	70	210	0.75	−70	0.25	−70
35	70	0.75	−210	0.25	70	210	0.25	−70	0.75	−70
36	70	0.75	−70	0.25	70	210	0.50	−210	0.50	−210
37	70	0.50	−210	0.50	−210	70	0.25	−70	0.75	70
38	−70	0.50	−210	0.50	−70	210	0.25	−210	0.75	−210
39	70	0.75	−70	0.25	70	210	0.25	−70	0.75	−70
40	70	0.25	−210	0.75	−210	−70	0.50	−210	0.50	−70
41	70	0.50	−70	0.50	70	70	0.75	−210	0.25	−210
42	70	0.50	−70	0.50	−70	210	0.25	−70	0.75	−70
43	210	0.50	−210	0.50	210	210	0.25	−210	0.75	−210
44	210	0.25	−70	0.75	−70	210	0.50	−210	0.50	210
45	70	0.75	−70	0.25	70	210	0.50	−70	0.50	−70
46	210	0.25	−210	0.75	210	70	0.50	−210	0.50	−210
47	210	0.25	−210	0.75	−210	70	0.50	−210	0.50	70
48	−70	0.75	−210	0.25	−70	210	0.50	−210	0.50	210
49	−70	0.50	−210	0.50	−70	70	0.25	−210	0.75	−210
50	210	0.50	70	0.50	70	210	0.75	−70	0.25	−70
51	210	0.50	70	0.50	210	210	0.75	−70	0.25	−70
52	210	0.50	−210	0.50	−210	210	0.25	−70	0.75	−70
53	210	0.50	−210	0.50	210	70	0.25	−210	0.75	−210
54	70	0.75	−70	0.25	70	210	0.50	−70	0.50	210
55	210	0.25	70	0.75	70	210	0.50	−70	0.50	210
56	210	0.50	−70	0.50	−70	210	0.75	−210	0.25	210
57	70	0.75	−210	0.25	70	210	0.25	−70	0.75	−70
58	−70	0.75	−210	0.25	−70	210	0.25	−210	0.75	−210
59	70	0.50	−70	0.50	−70	70	0.75	−210	0.25	70
60	70	0.75	−210	0.25	70	70	0.25	−70	0.75	−70
61	210	0.25	−210	0.75	−210	−70	0.50	−210	0.50	−70
62	210	0.75	−210	0.25	−210	210	0.25	70	0.75	210
63	210	0.50	70	0.50	210	210	0.75	−210	0.25	−210
64	210	0.25	−210	0.75	210	210	0.25	−210	0.75	−210
65	70	0.50	−210	0.50	70	70	0.25	−70	0.75	−70
66	210	0.25	70	0.75	70	210	0.75	−210	0.25	210
67	70	0.75	−70	0.25	−70	210	0.50	−70	0.50	−70
68	210	0.25	−210	0.75	210	70	0.75	−70	0.25	−70
69	210	0.50	70	0.50	210	210	0.50	−210	0.50	−210
70	210	0.25	70	0.75	210	210	0.50	−70	0.50	210
71	70	0.25	−210	0.75	70	70	0.50	−210	0.50	−210
72	−70	0.50	−210	0.50	−210	210	0.25	−210	0.75	−210
73	70	0.50	−70	0.50	70	210	0.75	−210	0.25	−210
74	210	0.25	−70	0.75	−70	210	0.50	−210	0.50	210
75	70	0.75	−210	0.25	−210	210	0.50	−210	0.50	210
76	210	0.50	70	0.50	70	210	0.75	−210	0.25	210
77	210	0.25	70	0.75	70	210	0.75	−210	0.25	210
78	210	0.75	−210	0.25	−210	210	0.50	−70	0.50	210
79	210	0.50	−210	0.50	−210	210	0.75	−210	0.25	210
80	70	0.50	−70	0.50	−70	210	0.50	−210	0.50	−210

Fig. 1.

Trial structure of task. Participants were presented two wheels (gambles) that differed with respect to the magnitudes and probabilities of outcomes, and asked to select one. After wheel choice, participants were shown how many points they won or lost and were asked to rate how pleased or disappointed they felt by that outcome (rating 1; partial feedback). Then, participants were shown how many points they could have won or lost if they had chosen differently (i.e., selected the other wheel) and were asked to rate how pleased or disappointed they felt knowing their outcomes compared with the other wheel (rating 2; complete feedback). The complete list of trials presented to participants is provided in Table S3. Equations represented here illustrate the mapping of these parameters onto information about each wheel provided to participants. d, Disappointment; e, expected value (EV); r, regret.

Data Analysis.

Affect ratings.

Separate analyses were conducted for ratings obtained after partial and complete feedback using mixed-effects linear regression models in STATA 13. For rating 1 (partial feedback), obtained outcome (i.e., the outcome value of the participant’s selection), chance counterfactual (i.e., the difference between the obtained outcome and what the participant could have obtained had the ball landed elsewhere within the chosen wheel), and SRP-III total score were included as continuous fixed-effect predictors; participant was modeled as a random effect. Interaction terms for SRP-III*chance counterfactual and SRP-III*obtained outcome were included in the same model, to examine the incremental effect of the counterfactual outcome over and above the obtained outcome. For rating 2 (complete feedback), obtained outcome, agent counterfactual (i.e., the difference between outcomes for the chosen vs. unchosen wheel), and SRP-III total score were included as continuous fixed-effect predictors; participant was modeled as a random effect. Interaction terms for SRP-III*agent counterfactual and SRP-III*obtained outcome were included in the same model, to examine the incremental effect of the counterfactual outcome over and above the obtained outcome. In keeping with prior work (26, 30, 35), we considered rating 1 and rating 2 to reflect reported disappointment and regret, respectively.

Decision making.

Choice models.

We examined the effects of three trialwise parameters on decision making (i.e., wheel choice): expected value (e), anticipated disappointment (d), and prospective regret (r) (Fig. 1). These parameters represent prospective estimates of e, d, and r, which are used to generate action values for each wheel on each trial. Potential outcomes and associated probabilities were ascribed the following notation: x₁ and y₁ correspond to the possible outcomes of wheel 1 (W1), where x₁ > y₁. Similarly, x₂ and y₂ refer the two possible outcomes of wheel 2 (W2), with x₂ > y₂. p and 1 − p are the respective probabilities of earning x₁ and y₁ and likewise q and 1 − q are the respective probabilities associated with earning x₂ and y₂.

The expected value (e) associated with choice of W1 was modeled by subtracting the expected value of W2 from that of W1, using Eq. 1. When e is positive, value-maximizing participants should select W1:

$$e=\mathrm{E}{\mathrm{V}}_{\mathrm{W}1}-\mathrm{E}{\mathrm{V}}_{\mathrm{W}2}=\left[\mathrm{p}{\mathrm{x}}_1+\left(1-\mathrm{p}\right){\mathrm{y}}_1\right]-\left[\mathrm{q}{\mathrm{x}}_2+(1-\mathrm{q}){\mathrm{y}}_2\right].$$ [1]

The anticipated disappointment (d) parameter takes into account both the probability of realizing the worst possible outcome, and the difference between the worst and best possible outcomes for that same wheel. The anticipated disappointment estimate for W1 is subtracted from that of W2, as detailed in Eq. 2. When d is positive, disappointment-avoiding participants should choose W1:

$$d=\left({\mathrm{x}}_2-{\mathrm{y}}_2\right)\left(1-\mathrm{q}\right)-\left({\mathrm{x}}_1-{\mathrm{y}}_1\right)\left(1-\mathrm{p}\right).$$ [2]

The prospective regret (r) calculation (Eq. 3) considers the size of the difference between the lowest and the highest outcomes across both wheels. This is grounded in the assumption that the magnitude of the difference between the obtained outcome and what could have been obtained had one chosen the other wheel determines the magnitude of reported regret/relief. Thus, regret-avoidant participants should choose the wheel that minimizes this difference, that is, W1 when r is positive and W2 when r is negative:

$$r=\left[\left({\mathrm{y}}_1-{\mathrm{x}}_2\right)-({\mathrm{y}}_2-{\mathrm{x}}_1)\right].$$ [3]

Based on these three parameters, individual trial-by-trial estimates were calculated for the probability of choosing wheel 1 (Pw1_it), where t denotes trial number and i denotes individual participant: P(Wheel 1_it) = 1 − P(Wheel 2_it) = F(e_it, d_it, r_it). F is the inverse logit function, F(θ) = e^θ(1 + e^θ) and θ is the logit predicted by the individual values of e, d, and r in the logistic regression.

Individual difference analyses.

We constructed several mixed-effects logistic regression models to estimate the main effects of e, d, and r on wheel choice [Choice ∼ e + d + r + (1|Participant)] as well their interactions with psychopathy [Choice ∼ e + d + r + SRP-III + SRP-III:e + SRP-III:d + SRP-III:r + (1|Participant)]. In these models, choice was a binary outcome variable (coded 1 for wheel 1 and 0 for wheel 2), SRP-III total score and each decision-making parameter (e, d, r) were considered as continuous fixed-effect predictors, and participant was treated as a random effect. Additional models presented in SI Results tested for symptom domain selectivity (e.g., externalizing vs. psychopathy; fearless dominance vs. impulsive-antisociality) and controlled for variation in age, education, IQ, substance abuse history, trait anxiety, and race/ethnicity.

Real-world behavior.

To determine whether variability in behavioral regret sensitivity predicted real-world choice behavior, we first obtained subjectwise estimates of the strength of association between r and choice behavior (i.e., unstandardized regression coefficients for r, from the main effect model described above). Next, we similarly calculated individual estimates of the strength of association between affective ratings and the chance/agent counterfactual magnitude for each trial. These values were entered into a negative binomial regression as a predictor of the number of prior incarcerations. SRP-III score and SRP-III*rating interaction terms were also included in the model.

SI Materials and Methods

Participants.

A prescreen phone interview and in-person assessment materials were used to exclude individuals who had performed below the fourth-grade level on a standardized measure of reading (Wide Range Achievement Test-III); who scored below 70 on a brief measure of IQ (Shipley); who had diagnoses of schizophrenia, bipolar disorder, or psychosis, not otherwise specified (Structured Clinical Interview for DSM Disorders); or who had a history of medical problems (e.g., uncorrectable auditory or visual deficits; head injury with loss of consciousness greater than 30 min) that may impact their comprehension of the materials or performance on the task.

Measures.

Self-Report Psychopathy-III (31).

The Self-Report Psychopathy-III (SRP-III) is a 64-item self-report questionnaire that is intended to measure features (e.g., criminal tendencies, erratic lifestyle, interpersonal manipulation, and callous affect) of psychopathy similar to those assessed by the Psychopathy Checklist-Revised (48). Items are scored on a 5-point Likert scale ranging from 1 (disagree strongly) to 5 (agree strongly). In the present study, the SRP-III displayed good internal consistency (Cronbach’s α = 0.910).

Externalizing Spectrum Inventory-Brief (34).

The SRP-III is sensitive to aspects of behavior that are common to multiple antisocial subtypes (e.g., criminal behavior, sensation seeking, impulsivity) (32, 33). To assess the degree to which the construct of psychopathy per se was associated with affective and behavioral regret sensitivity modulo these general aspects, we used participants’ score on the Externalizing Spectrum Inventory-Brief (ESI-Brief) to control for variation in trait externalizing. Externalizing can be considered a superordinate taxon encompassing poor response inhibition, threat hypersensitivity, and heightened negative affect, and is markedly elevated in antisocial and substance use syndromes. Of note, although externalizing and psychopathy produce similar behavioral manifestations (e.g., crime), they are characterized by unique neurocognitive profiles and distinct etiopathophysiological mechanisms. We therefore used the ESI as a phenotype control variable to assess whether any association to SRP scores was driven by psychopathy-specific variance as opposed to the more general aspects of antisocial behavior linked to externalizing.

The ESI-Brief is a 100-item self-report questionnaire that assesses a range of behavioral and personality characteristics associated with the externalizing spectrum of psychopathology. The items consist of statements regarding specific behaviors and qualities, and participants are asked to choose the response that best describes them on a 4-point Likert scale: True (1), Mostly True (2), Mostly False (3), or False (4). Before scoring, the appropriate items were reverse coded. Total scores range from 100 to 400, with higher scores corresponding to higher levels of externalizing. For the present sample, the internal consistency (Cronbach's α) was 0.981.

SI Results

Behavioral and Affective Regret Sensitivity.

In line with previous research (35), behavioral regret sensitivity (operationalized as the strength of the relationship between prospective regret and choice behavior) was not significantly correlated with affective regret sensitivity (operationalized as the strength of the relationship between the magnitude of the agent counterfactual and the magnitude of reported affective response) (P = 0.716). Likewise, behavioral regret sensitivity was not significantly correlated with affective disappointment sensitivity (i.e., strength of relationship between chance counterfactual and affective response; P = 0.434), or with affective outcome sensitivity (i.e., strength of relationship between obtained outcome and reported affect) for either partial (P = 0.160) or complete (P = 0.450) feedback trials.

Task Effects: Change of Mind by Opting to Switch.

As noted in Materials and Methods, participants had the opportunity to change their minds and switch their wheel selections on 50% of trials. This opportunity to switch wheels previously was shown to intensify the regret sensitivity among healthy participants (23), but not among individuals with psychopathology, such as obsessive–compulsive disorder (35), suggesting healthy participants were more sensitive to the emotional impact of personal responsibility. Follow-up analyses were conducted to examine the relationship between psychopathy and the opportunity to change one’s mind on affective responses.

The majority of individuals did not switch wheels (M = 3.23; SD = 3.15 of 40 trials). However, the total number of wheel switches did correlate with SRP-III score [r₍₆₀₎ = 0.30, P = 0.019], such that a higher level of SRP-III was associated with more wheel switches. Consistent with Gillan et al. (35), the change-of-mind opportunity did not exacerbate emotional responses to obtained outcomes (P = 0.779) or to the regret/relief (agent counterfactual) (P = 0.525). Additionally, no two-way interactions with SRP-III (P = 0.883) and obtained outcome or regret/relief (P = 0.814) were observed.

Although individuals higher on SRP-III were not differentially emotionally affected by the opportunity to change their minds, they did display greater wheel switching on trials that allowed choice switches. It may be that individuals higher on psychopathy reacted more to the information presented in the moment. Although speculative, this is consistent with attention-based models of psychopathy that suggest psychopathic individuals fail to integrate contextual information and react strongly to goal-relevant information in the moment (45). In the counterfactual paradigm, the pattern of points within wheels and across trials provided context for choices. This information requires parallel integration of information, a process that is deficient in psychopathic individuals. Instead, individuals high on psychopathic traits may have seen the opportunity to switch wheels as relevant to their ultimate goal of earning more points, without consideration of other previously presented information.

Individual Difference Analyses: Symptom Domain Selectivity.

Affective and regret sensitivity in psychopathy vs. externalizing.

Two antisocial subtypes, individuals with psychopathy and externalizing traits, are associated with significantly higher rates of antisocial activity and substance abuse than other individuals. Although individuals with psychopathy and those with externalizing traits have similar phenotypic expressions, including violent behavior, impulsivity, and substance abuse, they are associated with distinct psychobiological dysfunctions. We next examined subtype-specific associations to affective and behavioral regret sensitivity.

To assess the degree to which psychopathy was associated with affective responses (i.e., retrospective regret) controlling for other antisocial traits (i.e., externalizing), we used participants’ score on the ESI-Brief to construct a model that simultaneously considered affective rating-by-psychopathy and affect rating-by-externalizing interactions. For partial feedback ratings, psychopathy did not significantly modulate the effect of obtained outcome or chance counterfactual on reported affect after controlling for r-by-externalizing interactions (obtain outcome: P = 0.294; chance counterfactual: P = 0.522). However, we did observe a significant interaction between obtained outcome and externalizing (B = 0.0001, SE = 0.021, 95% CI = 0.0000025–0.0003, z = 1.99, P < 0.047), such that individuals with higher ESI-Brief scores reported more negative affect in response to negative outcomes.

For complete feedback, psychopathy significantly moderated the impact of obtained outcome on affective ratings (B = −0.006, SE = 0.002, 95% CI = −0.0101 to −0.0017, z = −2.76, P = 0.006), with less negative affect reported in response to the most negative outcomes for individuals with high vs. low SRP-III scores, and no difference in reported affect for positive outcomes. After controlling for variation in externalizing, psychopathy did not significantly moderate the relationship between agent counterfactual and affective ratings (P = 0.101). By contrast, a significant interaction was observed for externalizing and agent counterfactual (B = 0.0001, SE = 0.00004, 95% CI = 0.0003–0.0002, z = 2.67, P = 0.008); individuals exhibiting high vs. low levels of externalizing reported increased negative affect to the most negative counterfactuals. However, externalizing did not significantly moderate the effect of obtained outcome on affective ratings (P = 0.124). On the whole, these findings suggest that externalizing-specific variance moderates the degree of reported negative affect when an individual learns that the outcome of their choice is much worse than it would have been had they chosen differently. Importantly, although psychopathy-specific variance predicted weaker affective responses to negative obtained outcomes, it was not associated with differential affective regret sensitivity.

To determine the selectivity of diminished prospective regret sensitivity for psychopathy, we constructed a model that simultaneously considered decision variable-by-psychopathy and decision variable-by-externalizing interactions. We observed a significant r-by-psychopathy interaction even after controlling for variation in externalizing (B = −0.0004, SE = 0.0002, 95% CI = −0.0007 to −0.0001, z = −2.91, P = 0.004); however, the r-by-externalizing interaction was not significant (P = 0.16). Together, these results confirmed that psychopathy is associated with decreased behavioral regret sensitivity, even after adjusting for variation in aspects of antisocial behavior that may be present in, but not specific to, psychopathy.

Regret sensitivity in fearless dominance and impulsive antisociality.

Given the specific association with psychopathy and prospective regret sensitivity, as a convergent test of selectivity, we also examined trait-specific associations with this measure. Prior work using the Psychopathic Personality Inventory has identified two underlying factors: “fearless dominance” (FD) that is thought to preferentially index the interpersonal-affective facets of psychopathy, and “impulsive antisociality” (IA) is linked to substance abuse, aggression, impulsivity, and criminality (49). Using the Multidimensional Personality Questionnaire-Brief Form (50), FD and IA subscales (which were a derivative of the Psychopathic Personality Inventory) were calculated as linear combinations of specific standardized (i.e., z-scored) primary trait scales. Specifically, FD was calculated as (0.34 × zSocial Potency) + (−0.42 × zStress Reaction) + (−0.21 × zHarm Avoidance). IA was calculated as (0.16 × zAggression) + (0.31 × zAlienation) + (−0.13 × zTraditionalism) + (−0.29 × zControl) + (−0.15 × zSocial Closeness). Statistical analyses included decision variable-by-FD and decision variable-by-IA interactions as predictors of choice behavior. Results confirmed that lower behavioral regret sensitivity was specific to FD (r-by-FD interaction: B = −0.002, SE = 0.0003, 95% CI = −0.0026 to −0.0015, z = −7.83, P < 0.001; P = 0.449 for r-by-IA interaction).

In addition, this analysis revealed significant interactions between FD and both expected value and disappointment sensitivity (e-by-FD: B = 0.004, SE = 0.0009, 95% CI = 0.0023 to −0.0059, z = 4.65, P < 0.001; d-by-FD: B = 0.004, SE = 0.0009, 95% CI = 0.0023 to −0.0059, z = 4.65). The e-by-FD interaction is especially noteworthy, as it shows that individuals who score higher on this trait behave as rational utility maximizers. In other words, their choices are strictly yoked to the difference in expected value between the two wheels. If one considers prospective regret information as signaling an action cost, the observed pattern of behavior suggests that this cost signal is not able to appropriately modulate the representation of action values during decision making.

Individual Difference Analyses: Covariates.

For all analyses, several covariates were considered. Specifically, we examined age, IQ, education, substance abuse history, trait anxiety, and race/ethnicity as covariates. Each of these covariates was selected based on previous research that documented associations between these variables and counterfactual decision making, psychopathy, and/or criminal behavior. For example, impulsive-antisocial behavior tends to decrease with age (51); differences related to IQ/education indirectly relate to decision making and antisocial behavior; prolonged substance abuse can impact decision making and also has shared variance with psychopathic traits and criminal behavior (52); anxiety has been linked to differences in counterfactual decision making, as well as distinct etiological manifestations of psychopathy [e.g., primary and secondary psychopathy (53)]; and, finally, some previous work has shown specific racial/ethnic differences in cognitive mechanisms in psychopathy (54). Inclusion of these covariates did not alter any of the reported affect, decision-making, or real-world behavior (prior incarceration) results (i.e., psychopathy-related effects remained significant).

Results

Affect Models.

Partial feedback.

Consistent with previous research, we found a significant main effect for obtained outcome [B = 0.083, SE = 0.004, 95% confidence interval (CI) = 0.077–0.090, z = 23.77, P < 0.001] on affective response, such that higher outcome values were associated with more positive affective ratings, and lower outcomes were associated with more negative affective (disappointment) ratings. Chance counterfactual also had a significant main effect on affective ratings (B = 0.015, SE = 0.002, 95% CI = 0.012–0.019, z = 8.11, P < 0.001), with more positive affect reported when the difference between obtained and counterfactual outcomes was positive and more negative affect reported when the difference was negative. There was no main effect of SRP-III on rating 1 (P = 0.98). The interaction between obtained outcome and SRP-III was significant (B = 0.007, SE = 0.002, 95% CI = 0.003–0.01, z = 3.90, P < 0.001), with higher psychopathy scores predicting stronger negative and stronger positive affective responses to negative and positive obtained outcomes, respectively. However, we did not observe significant interactions between SRP-III and chance counterfactual (P = 0.359) (Fig. 2 A and C), showing that psychopathy did not modulate affective responses to disappointment-inducing outcomes. After controlling for variation in externalizing, psychopathy did not significantly modulate the effect of either obtained outcome or chance counterfactual on reported affect (obtained outcome: P = 0.294; chance counterfactual: P = 0.522; SI Results).

Fig. 2.

Affective ratings for partial and complete feedback. During partial feedback, individuals high on psychopathy reported more extreme affective responses to the values of what they won or lost (A), but not to chance counterfactual information (C). During complete feedback, the affective ratings for high psychopathy individuals were less influenced by obtained outcome (B), but more influenced by agent counterfactual values (D). Together, these show that psychopathy is associated with typical, or perhaps even enhanced, emotional responses to regret. Shading around lines represents 95% CI for point estimates.

Complete feedback.

At the complete feedback stage, participants were shown what they would have obtained had they chosen a different wheel. In line with prior work, there were main effects of obtained outcome (B = 0.013, SE = 0.003, 95% CI = 0.008–0.018, z = 4.71, P < 0.001), with larger loss and larger gain outcomes predicting stronger negative and stronger positive affective responses, respectively. Additionally, there was a main effect of agent counterfactual (B = 0.041, SE = 0.002, 95% CI = 0.037–0.004, z = 22.35, P < 0.001) on affect ratings, with larger negative values (indicating better outcomes for the unchosen wheel) and larger positive outcomes (indicating better outcomes for the chosen wheel) predicting stronger negative and positive affective responses, respectively. The main effect of SRP-III was not significant (P = 0.215); however, we did observe significant interactions between SRP-III and both obtained outcome (B = −0.004, SE = 0.001, 95% CI = −0.006 to −0.0008, z = −2.51, P < 0.012) and agent counterfactual (B = 0.005, SE = 0.0009, 95% CI = 0.0034–0.0070, z = 5.62, P < 0.001) (Fig. 2 B and D). These results show that higher SRP-III scores predicted lower reported negative affect in response to negative obtained outcomes, and exaggerated negative and positive affective responses to negative and positive counterfactual feedback (i.e., more reported regret/relief). After controlling for variation in externalizing, the SRP-III*obtained outcome interaction remained significant (P = 0.006); however, the SRP-III*agent counterfactual interaction did not (P = 0.101), indicating that, although psychopathy-specific variance predicted weaker affective responses to negative outcomes, it was not associated with differential affective regret sensitivity (SI Results).

Decision-Making Models.

Main effects of decision variables e, d, and r.

There was a significant main effect for e, whereby all participants showed a tendency to choose wheels based on anticipated economic utility (B = 0.012, SE = 0.001, 95% CI = 0.010 to −0.014, z = 12.51, P < 0.001). The main effect of r was also significant (B = 0.004, SE = 0.0003, 95% CI = 0.004–0.005, z = 15.01, P < 0.001). However, the main effect of d was not significant (B = 0.0008, SE = 0.0004, 95% CI = −0.00001–0.002, z = 1.93, P = 0.054). These results are largely consistent with prior work showing that individuals make choices to minimize the likelihood of experiencing regret (21, 24).

Decision Variable-by-Psychopathy interactions.

The results described above indicate that trial-by-trial variation in choice behavior is driven, in part, by trialwise variation in the magnitude of prospective regret signals. However, this linkage was significantly weaker in individuals with higher SRP-III scores (B = −0.0005, SE = 0.0001, 95% CI = −0.0008 to −0.0002, z = −3.71, P < 0.001; Fig. 3A), suggesting that, although the magnitude of anticipated regret drives choice at low levels of psychopathy, this relationship becomes uncoupled as psychopathy severity increases. In other words, participants with higher SRP-III scores appeared relatively insensitive to the prospect of regret during value-based decision making, despite the fact that they reported typical negative affective responses to regret-inducing outcomes. The relationship between SRP-III and prospective regret sensitivity remained significant after controlling for variation in externalizing (P = 0.016; SI Results), indicating that psychopathy is associated with decreased behavioral regret sensitivity even after adjusting for variation in nonspecific aspects of antisocial behavior.

Fig. 3.

Behavioral sensitivity to regret, expected value, and disappointment. Panels illustrate the logit model-predicted probability of choosing wheel 1 at varying levels of prospective regret (A), expected value (B), and disappointment (C), for high (red line)- and low (blue line)-psychopathy individuals. Shading around lines represent 95% CI for point estimates. Choice behavior in individuals higher on psychopathy were less affected by prospective regret (A) and disappointment (C) signals.

The interaction between d and SRP-III was also significant, albeit weaker in magnitude (B = −0.0005, SE = 0.0002, 95% CI = −0.0009 to −0.0003, −z = 2.39, P = 0.017; Fig. 3C). We did not observe a main effect of SRP-III on choice behavior (P = 0.143), nor was the interaction between SRP-III and e (P = 0.118; Fig. 3B) significant. SRP-III scores did not significantly predict the total number of points won at the end of the task (P = 0.834). These findings, considered together with the affective rating data above, show that individuals scoring higher on a measure of psychopathy exhibited typical (or perhaps even more extreme) negative emotional responses to regret-inducing outcomes, yet their ability to use prospective regret signals to guide choice behavior appeared relatively compromised. This pattern of results suggests that psychopathy is associated with a selective deficit in the ability to use forward action–outcome models to guide action selection; retrospective counterfactual evaluation, on the other hand, appears to be preserved.

Real-world behavior.

Lower behavioral regret sensitivity predicted a greater number of prior incarcerations (B = −65.48, SE = 26.64, 95% CI = −117.69 to −13.27, z = −2.46, P = 0.014). Behavioral regret sensitivity remained a significant negative predictor of number of prior incarcerations, even after controlling for affective outcome sensitivity, affective regret sensitivity, and affective disappointment sensitivity (P = 0.04). Although these results show that behavioral regret sensitivity selectively predicts incarceration, the relevance of psychopathy to that relationship remained unclear. To confirm a direct link between psychopathy, behavioral regret sensitivity, and real-world decision making, we modeled the effect of interactions between SRP-III score and behavioral regret sensitivity, affective regret sensitivity, affective outcome sensitivity, and affective disappointment sensitivity on number of prior incarcerations. The psychopathy-by-behavioral regret sensitivity interaction was the only significant predictor of incarceration in this model (B = −22.94, SE = 11.06, 95% CI = −44.62124 to −1.259966, z = −2.07, P = 0.038) (Fig. 4). This result shows that the highest risk for incarceration results from a combination of high SRP-III scores and low behavioral regret sensitivity. Psychopathy was not associated with the number of prior incarcerations at higher levels of behavioral regret sensitivity, suggesting that higher behavioral regret sensitivity mitigated the impact of SRP-III score on incarceration.

Fig. 4.

Behavioral regret sensitivity moderates the relationship between psychopathy and number of prior incarcerations. Individuals with a combination of high SRP-III scores and low behavioral regret sensitivity reported more prior incarcerations, compared with individuals with high SRP-III scores and high behavioral regret sensitivity. For display, SRP-III and behavioral regret sensitivity values were separately split into two groups, based on median values. Error bars indicate 95% CIs. Although not used for inference, the 2 × 2 group factor interaction is also significant (P < 0.01).

Discussion

Psychopathic individuals display a striking lack of remorse and regret when faced with the profoundly adverse consequences of their frequent and flagrant norm violations. Some influential theories suggest that this is due to an inability to generate aversive emotional responses to negative outcomes. The current data suggest that an alternative viewpoint merits consideration. Using a counterfactual decision-making paradigm, we found that psychopathy was not associated with blunted emotional responses to retrospective regret signals. Individuals with high SRP-III scores reported significant negative affect when informed that they could have received a better outcome had they chosen a foregone option. By contrast, psychopathy did affect the degree to which prospective regret signals were used to guide choice behavior. These signals were largely ignored in individuals who scored high on SRP-III, whose decisions were driven primarily by the expected value of each option irrespective of the potential for postdecision regret. Moreover, behavioral regret sensitivity predicted participants’ incarceration histories, a real-world marker of maladaptive decision making, and moderated the impact of psychopathy on criminal behavior. On the whole, these findings raise the intriguing possibility that maladaptive decision making in psychopathy is not the result of a basic emotion deficit, but may instead arise from a problem with generating forward models that estimate values for a range of potential actions by simulating likely outcomes associated with their execution, or in using such models to guide online action selection.

The finding that behavioral regret sensitivity is reduced in psychopathic individuals is especially interesting given that psychopathy is associated with structural and functional deficits in brain circuitry important for counterfactual decision making. Several lesion studies have shown that patients with mOFC damage exhibit behavioral regret insensitivity (25, 30), and functional imaging findings confirm that mOFC activity predicts regret-avoidant decision making (24). Likewise, electrophysiological data indicate that mOFC neurons maintain a representation of the reward value of unchosen options following action selection (36)—consistent with mOFC’s suggested role in prospectively mapping action options to outcome values (37)—and that mOFC counterfactual value signals are transmitted to other regions more directly involved in action selection (e.g., dorsolateral prefrontal cortex, medial caudate) (36, 37). With regard to psychopathy, structural imaging studies have found reduced surface area, cortical thickness, and gray matter volume within OFC (38). Accordingly, psychopathic individuals exhibit blunted OFC activation during reward learning, empathy, and emotion appraisal paradigms (12, 13, 15), and show reduced functional and structural connectivity between OFC and subcortical zones (e.g., amygdala and striatum) (10, 39). Taken together, findings from these parallel literatures suggest that diminished behavioral regret sensitivity may be a consequence of OFC dysfunction or, more likely, dysconnectivity in psychopathic individuals.

Although the link between OFC function and behavioral regret sensitivity is relatively uncontroversial, less is known about the specific mechanisms that support this link. mOFC (particularly areas 10r and 10m) is one component of a distributed network that supports the construction of self-relevant mental simulations. mOFC activation is consistently seen across otherwise disparate tasks—such as prospection, autobiographical episodic recall, and theory of mind—that require a mental model of hypothetical actions, states, or outcomes (40–42). The use of such models to represent actions, states, and outcomes is a hallmark of “model-based” decision systems. Whereas “model-free” valuation relies on the directly experienced outcomes of an agent’s actions, model-based valuation uses prospective simulation to integrate counterfactual reward information into stimulus/action value representations. Recent work suggests that connectivity between mOFC and striatum may be important for the model-based updating of action–value associations via counterfactual outcome simulation. Although mOFC has been shown to represent counterfactual outcome values (36), striatal signals encode action values that are directly used to guide action selection (37). Recent work by Kishida et al. (43) shows that striatal dopamine transients reflect the integration of reward prediction error and counterfactual prediction error signals. Together, these data raise the possibility that model-based value information represented in prefrontal cortex may shape choice behavior, in part, by directly modulating subcortical action value signals (i.e., “model-based modulation”).

The current study provides evidence that model-based modulation is disrupted in psychopathy. Participants with higher scores on a measure of psychopathy showed typical levels of negative emotion in response to regret-inducing feedback (i.e., intact retrospective counterfactual simulation), but appeared unable to use counterfactual value representations to guide choice behavior (i.e., impaired prospective counterfactual simulation). One might expect that a model-based decision-making deficit per se would disrupt both processes, as is evident in OFC lesion patients (30). Instead, psychopathy-related differences were evident only when counterfactual information was required for action selection. Notably, such differences were selective for psychopathy, rather than more general aspects of antisocial behavior. This finding would appear to be at odds with the only other study to examine counterfactual decision making in an antisocial population, which reported a weak association between behavioral regret sensitivity and impulsive-antisocial traits (44). The discrepant findings highlight the utility of phenotype selectivity analyses for neural and behavioral studies of psychopathy (see SI Discussion for a detailed discussion of this issue).

These findings are particularly interesting given that psychopathy is associated with aberrant selective attention. A putative early “attentional bottleneck” leads to enhanced allocation of resources to information that is consistent with an individual’s current, goal-directed focus; however, this comes at the expense of processing information that may be highly salient, yet peripheral to that goal set (45). Here, participants were explicitly instructed to maximize the number of points won. This instruction could have generated an inflexible attentional focus on reward magnitude, producing the relative overweighting of expected value and underweighting of anticipated regret that we observed in individuals with high levels of trait psychopathy. This hypothesis is consistent with the known importance of attention processes for input selection during value-based decision making (46), and with work showing decreased regret sensitivity and diminished counterfactual elaboration in individuals with poor attentional flexibility (47). Future work should test the hypothesis that aberrant model-based value modulation in psychopathy arises from a primary deficit in the allocation of goal-directed attention.

Several methodological and conceptual limitations should be noted. First, although we used a targeted recruitment approach in a high-crime community to ensure a distribution of psychopathic traits (Table S1), the rates of psychopathy (e.g., 1% in the general population vs. 15–25% in prisons) and the measures used to assess these traits (e.g., questionnaire vs. Psychopathy Checklist interview) are different in forensic populations (48). Future studies should examine counterfactual decision-making processes in incarcerated offenders to confirm the generalizability of these findings to the full range of psychopathy severity. Finally, we used an economic decision-making paradigm to operationalize regret sensitivity. Although this conceptualization of regret has compelling support from both experimental psychology and behavioral economics, one might entertain an objection based on the distinction between regret and remorse. The former describes a negative emotion elicited by knowledge of a rejected alternative’s outcome for oneself; the latter entails an aversive state induced by information that a foregone alternative would have produced a better outcome for another. Clinical characterizations suggest that psychopathic individuals demonstrate impairments in both regret and remorse. Future work using novel tasks that consider counterfactual decision making in the context of dyadic interactions will be useful for clarifying the relationship between remorse sensitivity and psychopathy, and in particular, to determine the degree to which regret and remorse sensitivity exert independent and interactive effects on real-world behavior in psychopathic individuals.

In sum, the present study identifies a specific deficit in the ability of individuals with psychopathic traits to integrate prospective counterfactual signals into decision making. By contrast, their ability to perform retrospective counterfactual comparisons appears to be preserved, as evidenced by their self-reported negative affect when faced with regret-inducing counterfactual information. The current data provide additional support for the idea that maladaptive behavior in psychopathy may result from deficits in domain-general cognitive processes, such as counterfactual representation, rather than a primary emotion deficit. Specifying the mechanisms that account for the striking disconnect between affective experience and decision making observed here will be crucial for advancing our understanding of the cognitive and neurobiological roots of psychopathy.

SI Discussion

Power.

Although the current sample size is comparable to or larger than those of other studies examining individual differences in counterfactual decision making (23, 35, 44), there is increasing recognition that compelling statistical inference requires explicit consideration of power. Power analyses indicated that the current sample size provided sufficient power (i.e., >80%) for all models, at an average of 97%, to detect a moderate effect size (d = 0.50 with a two-tailed α of 0.05).

Discussion of Hughes et al. (44).

It is useful to discuss the relationship between the findings reported here and those reported in a prior study by Hughes et al. (44). Briefly, the authors observed that affective (but not behavioral) regret sensitivity was lower in a group of 60 incarcerated offenders compared with 20 healthy volunteers, and reported weak evidence of an association between impulsive-antisocial (but not psychopathic) traits and behavioral regret sensitivity. This is in contrast to the present finding that lower behavioral regret sensitivity is specifically lower in psychopathic (but not impulsive-antisocial) individuals. There are several methodological differences between the two studies that account for the seemingly discrepant findings. First, the Hughes et al. primary comparison involved a contrast between 60 incarcerated offenders and 20 undergraduate controls. Despite demographic differences between these samples, they failed to consistently control for the potentially confounding effects of age, socioeconomic status, environmental exposures, substance abuse, education, ethnicity, and psychiatric illness. Post hoc analyses found a weak association between impulsive-antisocial traits and behavioral regret sensitivity. However, these analyses should be interpreted with caution given the authors’ design and analysis choices. In particular, Hughes et al. used only 16 trials in their experimental design, compared with 80 in the current study [n.b. of the nine extant studies using the Camille et al. counterfactual paradigm (23), the median number of trials is 80; range: 60–192]. Furthermore, in their analyses of affective regret sensitivity, as few as 0 trials were included in their statistical model for some participants. Finally, Hughes et al. used mixed-effect linear/probit regression models with separate trialwise regret, disappointment, and expected value predictors. In all of the previous research that used this task, panel logit or linear mixed models were used to estimate how strongly participants relied on prospective regret, disappointment, and expected value signals. Critically, given how the regret, disappointment, and expected value predictors are constructed, each of these shares some variance with the other. Appropriate statistical modeling requires that all three of these predictors be included in the same model; likewise, any examination of individual differences requires that the model include all three predictors and an additional three terms representing interactions between these predictors and some group or individual difference variable (e.g., psychopathy). For example, the inference that impulsive-antisocial traits affected the use of prospective regret signals in Hughes et al. (44) was derived from a model that included only regret and regret X impulsive-antisocial trait predictors. This is important because—given the covariance between regret, disappointment, and expected value predictors—it is impossible to know whether this effect is specific to the regret variable or is driven by disappointment or expected value. Considering the present data, the interaction between psychopathy and expected value is significant when we use a model that only includes expected value, psychopathy, and expected value X psychopathy predictors. However, this effect disappears when we model the data appropriately, showing that this apparent effect was actually driven by variance held in common with the other parameters (e.g., disappointment, regret). The modeling approach used by Hughes et al. (44) limits the inferential power of their findings and makes it challenging to compare the results of that study to the findings reported here. Overall, it is really difficult to compare the findings from the study by Hughes et al. (44) to the findings from the present study because their design and analytic choices preclude the possibility of examining the relationship between counterfactual decision making, psychopathy, and regret (prospective and retrospective).