Autonomic Interplay and the Bimodal Classification of Aggression: Baseline and Reactivity Profiles in Urban Adult Males

ABSTRACT

Biological theories propose distinct physiological profiles for premeditated aggression (PM) and impulsive aggression (IA), yet prior research has relied on single‐system markers that have yielded inconsistent findings, and no study has examined PNS‐SNS interplay in relation to these aggression subtypes in adult samples including active offenders. This study investigated whether multisystem autonomic profiles at rest and during a modified Trier Social Stress Task differentially associated with PM and IA in a heterogeneous urban adult sample. Active offenders, demographically matched controls, and college students (N = 130, all male, M age = 28 years) were recruited in Atlanta. Skin conductance and respiratory sinus arrhythmia were recorded at rest and during three mTSST components varying in social confrontation demand and cognitive load. Each aggression subtype was modeled with the other adjusted for (PM and IA correlated at r = 0.54), so all associations pertain to the variance unique to that subtype. Baseline RSA was positively associated with PM only at low to moderate baseline SCL, and both reciprocal PNS and SNS activation profiles during social confrontation were associated with higher PM. Low baseline RSA was associated with more IA regardless of arousal level, with no significant reactivity interactions. All findings are replicated using peak‐to‐valley RSA. These results indicate that PM and IA are associated with qualitatively distinct multisystem physiological profiles. PM was associated with two divergent patterns under social challenge, physiological disengagement and organized mobilization, whereas IA was associated with a single trait‐level marker of low regulatory capacity independent of situational arousal.

Impact Statement

This study is among the first to examine sympathetic and parasympathetic interplay in relation to premeditated and impulsive aggression in an adult sample including active offenders. Premeditated aggression was associated with two distinct autonomic profiles under social confrontation, whereas impulsive aggression was associated with a single trait‐level marker of low regulatory capacity. The findings indicate that planned and reactive aggression have distinct physiological correlates, with implications for how biosocial theories of aggression are specified.

1. Introduction

Research on the autonomic nervous system (ANS) has long sought biological markers that index individual differences in emotional and behavioral regulation, with the goal of illuminating why some individuals are prone to antisocial and aggressive behavior. A sizable body of evidence has established that both sympathetic nervous system (SNS) activity, indexed by skin conductance (SC), and parasympathetic nervous system (PNS) activity, indexed by respiratory sinus arrhythmia (RSA), are associated with conduct problems and aggression. However, a comprehensive meta‐analysis of 101 studies and 769 effect sizes found that these associations are heterogeneous and bidirectional, and that single physiological markers may be most useful not in predicting antisocial behavior globally but in differentiating between its functionally distinct subtypes (De Looff et al. 2022). This pattern of findings underscores a fundamental limitation of single‐system approaches: the relevant biological signal may lie not in any individual marker examined in isolation, but in the coordinated interplay of arousal and regulatory systems across different behavioral demands.

A growing body of research has begun to address this limitation by adopting a multisystem approach that simultaneously examines SNS and PNS activity and their interaction (Buss et al. 2018; Chen and French 2024). Grounded in the seminal framework of the autonomic space developed by Berntson et al. (1991), Berntson et al. (1993), this approach recognizes that PNS and SNS do not operate in simple reciprocal fashion but can coactivate, coinhibit, or exhibit various coupling patterns in response to internal and external demands, each carrying distinct functional implications for behavior. Prior research in children and adolescents has demonstrated that disorganized PNS‐SNS profiles confer elevated risk for externalizing problems, particularly in the context of environmental adversity(El‐Sheikh et al. 2009; Gordis et al. 2010; Ter Harmsel et al. 2022). However, research extending this multisystem logic to adult samples with heterogeneous offending histories, and specifically to the bimodal classification of aggression, remains exceedingly sparse.

1.1. The Bimodal Classification of Aggression and Its Physiological Underpinnings

The bimodal classification distinguishes premeditated aggression (PM), characterized by controlled, calculated, goal‐directed behavior, from impulsive aggression (IA), characterized by reactive, emotionally charged, and poorly regulated behavior (Barratt et al. 1997; Stanford et al. 2003). Although this distinction maps onto the proactive‐reactive distinction widely used in child and adolescent research, the two terminological frameworks share substantial empirical convergence, and findings have replicated across measurement approaches (Babcock et al. 2014). This conceptual overlap justifies drawing on the proactive‐reactive literature to inform theoretical expectations for PM and IA, while recognizing that the two frameworks are not identical and that findings may not replicate perfectly across terminological systems. Theoretically, PM has been associated with physiological under‐arousal, consistent with its empirical links to psychopathic traits, reduced fear, and sensation‐seeking (Belfry and Kolla 2021; Raine 1993; Zuckerman 1979). IA, by contrast, has been theorized and observed to reflect over‐arousal and emotion dysregulation, consistent with a frustration‐aggression model in which hostile, impulsive reactions arise from perceived threat and blocked goals (Berkowitz 1962; Ramirez and Andreu 2006; Seo et al. 2008).

A recent scoping review of 74 studies on the biological underpinnings of proactive aggression confirmed a robust association with reduced resting heart rate and, to a lesser extent, dampened SNS reactivity as indexed by skin conductance (Belfry and Kolla 2021). Crucially, associations involving RSA, examined in isolation, were largely non‐significant. This null pattern for RSA is consistent with the broader theoretical argument that RSA functions as a transdiagnostic, non‐specific marker of emotion dysregulation whose behavioral implications are contingent upon the motivational and situational context in which it is assessed, and upon the other regulatory systems with which it is coupled (Beauchaine 2015; Beauchaine and Thayer 2015; Chen and French 2024). In support of this argument, Chen and French (2024) demonstrated in an adult sample that the association between RSA reactivity and externalizing problems was fundamentally altered by whether RSA was measured during a stress task or a reward task, and by whether it was examined alone or in interaction with sympathetic reactivity indexed by the pre‐ejection period. The present study applies a parallel logic to the bimodal classification of aggression, with a focus on the coordinated interplay of SC and RSA during a social stress paradigm.

1.2. SNS and PNS Indicators: Baseline and Reactivity

It is critical to distinguish between baseline and reactivity indices of SNS and PNS activity, as they capture fundamentally different physiological constructs and carry different interpretive implications for the biology of aggression. Baseline SC level (SCL‐B) reflects overall tonic arousal, indexing the general level of SNS activation at rest. Low baseline SCL has been linked to reduced orienting responses, fearlessness, and stimulation‐seeking behavior in the antisocial behavior literature, consistent with the proposition that individuals with low tonic arousal are less reactive to environmental signals that typically inhibit aggressive behavior (Lorber 2004; Raine 1993). This pattern is theorized to reflect a constitutionally low sensitivity to environmental input rather than a motivationally driven compensatory process, and it has been examined primarily in relation to broad antisocial behavior and conduct problems rather than aggression subtypes specifically. Baseline RSA (RSA‐B) reflects an individual's regulatory capacity, or the energetic reserve from which the organism draws during metabolic and behavioral challenge (Grossman and Taylor 2007; Thayer et al. 2009). Research has accumulated to support baseline RSA as an index of emotion regulation capacity and executive functioning, with higher baseline RSA associated with better behavioral regulation in social and cognitively demanding contexts (Appelhans and Luecken 2006; Beauchaine 2015; Thayer and Lane 2009).

Unlike baseline indices that capture trait‐level arousal and regulatory capacity, task reactivity scores reflect the dynamic engagement of physiological systems in response to specific situational demands. RSA withdrawal during stress indicates the degree to which the PNS relinquishes cardiac control to allow mobilization of metabolic resources in response to challenge, and its adaptive versus maladaptive character is heavily contingent on the nature of the task (Beauchaine et al. 2019; Chen and French 2024; Holzman and Bridgett 2017; Porges 2007). SCL reactivity during tasks with social confrontation and evaluative threat reflects the level of SNS mobilization and emotional engagement with the stressor (Dawson et al. 2007). Interpretation of PNS‐SNS reactivity profiles must therefore incorporate key features of the tasks used to elicit them, including their degree of social confrontation, vocalization demand, and cognitive load.

1.3. PNS‐SNS Interplay and Aggression Subtypes

The few studies that have simultaneously examined PNS and SNS in relation to aggression subtypes have been conducted primarily with children or with college students reporting on relational aggression (Moore et al. 2018; Murray‐Close et al. 2017; Scarpa et al. 2010; Wagner and Abaied 2015), and research with adult samples involving clinical or forensic populations remains exceedingly sparse. Scarpa et al. (2010) found that children's proactive aggression was associated with heightened baseline RSA and SCL, while reactive aggression was associated with reduced RSA only, suggesting distinct physiological profiles at the trait level. Moore et al. (2018) found that task RSA and SCL did not interact to predict proactive aggression in children, while reactive aggression was predicted by an interaction in which RSA moderated the link between skin conductance and reactive aggression, such that higher skin conductance was associated with more reactive aggression among children with low RSA. Murray‐Close et al. (2017) found that in college students reporting on relational aggression, reciprocal PNS activation was associated with lower levels of both subtypes. Wagner and Abaied (2015) further found that the relationship between RSA withdrawal and proactive relational aggression depended on relational victimization history.

Four studies have extended this question to adult samples without the focus on relational aggression, with mixed results. Armstrong et al. (2019) examined resting skin conductance and skin conductance reactivity in a large college sample, finding sex‐differentiated associations in which resting skin conductance was positively associated with proactive aggression among females while skin conductance reactivity was positively associated with reactive aggression and negatively associated with proactive aggression among males. Thomson et al. (2021) examined RSA and skin conductance reactivity to immersive virtual reality fear induction in young adults, finding that proactive aggression in males was associated with coinhibition of both PNS and SNS during fear induction, consistent with the fearlessness hypothesis, while reactive aggression was associated with heightened SNS reactivity in both sexes. Puhalla and McCloskey (2020) examined RSA and electrodermal activity during a laboratory provocation paradigm, in which college adults were angered by a confederate and given the opportunity to respond aggressively, and found that blunted EDA reactivity to provocation was associated with higher aggression across both subtypes, while the direction of RSA effects depended on whether individuals were high or low on emotion dysregulation. This suggests that the PNS‐SNS interplay relevant to aggression in adults may be conditioned on regulatory capacity rather than operating uniformly across individuals. Ter Harmsel et al. (2022) took a different approach and examined HR, RSA, and SCL responses to passive viewing of scenes depicting aggressive versus neutral interactions in a sample of delinquent young adults, finding that no neurobiological measure, examined additively, significantly predicted either reactive or proactive aggression within the delinquent group.

Together, these findings offer partial but inconsistent support for distinct physiological profiles of proactive and reactive aggression in adults. SNS indicators show the most consistent pattern, with dampened sympathetic responding associated with proactive aggression and heightened sympathetic responding associated with reactive aggression, particularly among males. The role of PNS activity is less consistent: RSA was unrelated to either aggression subtype when examined additively in delinquent young adults (Ter Harmsel et al. 2022), its effects on aggression were conditioned on emotion dysregulation rather than operating uniformly across individuals in a college adult sample (Puhalla and McCloskey 2020), and coinhibition of both PNS and SNS was associated with proactive aggression only in males and only under immersive fear induction (Thomson et al. 2021). Critically, existing studies have relied on passive viewing paradigms, laboratory provocation tasks, or virtual reality fear induction, none of which directly engage the social‐evaluative confrontation that is most theoretically central to physical premeditated and impulsive aggression. Whether the interactive interplay of RSA and SCL during active social confrontation differentiates PM from IA in adults remains an open question.

1.4. Current Study

The present study examined the interplay of PNS and SNS activity at rest and during a social stress task in relation to PM and IA in a heterogeneous adult urban sample comprising active offenders, demographically matched community controls, and college students recruited from a major metropolitan area. This sample design provides substantially greater range on both aggression outcomes and physiological indicators than is available from college convenience samples alone, and to my knowledge no prior study of multisystem autonomic profiles and aggression subtypes has included active offenders.

Physiological data were collected during a modified Trier Social Stress Task with three components varying in social confrontation demand, vocalization, and cognitive load, allowing examination of whether PNS‐SNS interplay relevant to PM and IA is specific to social challenge or generalizes across task demands. Separate regression models were estimated for PM and IA with each form of aggression adjusted for the other, following the recommendation that unique physiological signatures of correlated aggression subtypes can only be recovered when their shared variance is accounted for.

Based on the theoretical frameworks reviewed above, I expected baseline RSA to be positively associated with PM at low to moderate levels of baseline SCL, reflecting the combination of strong regulatory capacity and low tonic arousal that may enable calculated, purposeful aggression. For IA, consistent with the emotion dysregulation framework, I expected that low baseline RSA would be associated with IA regardless of general arousal level. Regarding reactivity, the autonomic space framework provides theoretical grounds for examining the interplay of RSA and SCL reactivity in relation to PM and IA, but prior research with adults is too limited to support fully specified directional predictions. I therefore treated these reactivity analyses as theoretically motivated but exploratory, examining whether specific PNS–SNS coupling profiles during social evaluation distinguished the two aggression subtypes. All primary findings were tested for robustness using peak‐to‐valley RSA as an alternative PNS index.

2. Method

2.1. Participants

The data used in the present analyses were drawn from the project on physiological correlates of offending behavior in which interview, psychobiological and survey data were collected (2017–2019). This project was approved by the Institutional Review Board of the researcher's home institution. College student participants were recruited in large introductory classes. Participants who were not college students were recruited by street recruiters from Central West Atlanta, a community that has suffered high rates of drug trafficking, serious street crime and violence. The recruitment strategy of using street recruiters for active offenders was successfully used in prior research (Brezina and Topalli 2012; Topalli et al. 2013). Enrollment was restricted to males because serious street offending of the kind that defined the offender group is heavily male‐dominated in this setting, and a female offender subsample of comparable size and offending profile could not be assembled through the same channels; extending enrollment to women in the control groups alone would have confounded sex with recruitment group. This restriction also avoided modeling sex differences within a sample already stratified across three recruitment groups, preserving power for the focal interaction tests. To supplement the recruitment, a snowball sampling and word‐of‐mouth referral was conducted, through which potential participants were identified via coworkers and social networks within the same communities. Active offenders are defined as those who commit crime and rely on the profit derived from the crime as their main source of income. Demographically matched controls of the active offenders (hereafter as demographic controls) were recruited from the same neighborhoods as active offenders and were similar to active offenders with regard to their residing neighborhoods, education levels, race and age but differed in that they did not derive their primary income from criminal activities. This group‐level matching was achieved through the same street recruitment and snowball sampling strategy used for the offender group. As the distinguishing criterion between active offenders and demographic controls was based on current income source rather than offending history per se, demographic controls were not excluded on the basis of prior offending history, which is consistent with community‐recruited samples from high‐risk urban neighborhoods where some prior contact with the criminal justice system is not uncommon. Exclusion criteria for all groups were female, younger than 18, currently on probation, diagnosis of a psychotic disorder, intellectual disability, or current medication use with the potential to interfere with the measurement of salivary analytes such as steroid‐based anti‐inflammatory (Granger et al. 2009). A total of 133 participants enrolled in this research project.

Of the 133 male participants, 2 opted out of the stress task (described below in measures), and 1 had missing data on the aggression measure, the key outcome variables. These three participants were subsequently excluded from the analysis resulting in a final analytic sample of 130 participants, with 47 college students, 45 active offenders, and 38 demographic controls. 82% of the participants aged 18 and 36 (M = 28.11, SD = 10.93). Participants self‐identified as African American (n = 107), White (n = 6), or other/mixed race (n = 17).

Based on the open‐ended interview responses, 49% of college students reported no offending history, 38% reported their most serious offense involved minor or status offenses such as underage drinking, smoking weed, shoplifting, or trespassing, and 6% reported property crimes such as burglary, auto theft, or breaking and entering. Only 2% each reported drug offenses or violent crime as their most serious offense. Similarly, 39% of demographic controls reported no offending history; 32% reported minor or status offenses as their most serious offense; 13% reported property crimes; and 8% each reported drug offenses or violent crimes such as assault or robbery. In contrast, no active offenders reported having no offending history. Among offenders, 44% reported violent crimes (e.g., assault, robbery, armed robbery, carjacking, or murder) as their most serious offense, 31% reported drug offenses (e.g., selling drugs or marijuana), 11% reported property crimes, and 11% did not specify or chose not to say. Only 2% of offenders reported minor or status offenses. Qualitative examination of the open‐ended interview responses further revealed that the offenses reported by college students and demographic controls were predominantly non‐recurring and isolated incidents, whereas active offenders frequently described patterns of repeated and concurrent involvement in multiple serious offense types (e.g., robbery combined with carjacking and drug sales). Together, these distributions highlight meaningful differences in offending severity across groups: college students and demographic controls were concentrated in the no‐history and minor‐offense categories, whereas active offenders were overwhelmingly concentrated in the violent crime and drug offense categories (see Table S1).

2.2. Measures

Impulsive aggression and premeditated aggression are measured with the Impulsive/Premeditated Aggression Scale (IPAS), a self‐report measure designed to characterize aggressive acts occurring in the past 6 months (Stanford et al. 2003). Participants were given the written instruction that aggressive acts were defined as hitting, pushing, or verbally insulting another person or breaking or throwing objects because you were angry or frustrated. Items are rated on a 5‐point scale (1‐strongly disagree; 5‐strongly agree). The validity and psychometric property were tested in the initial study of physically aggressive men (Stanford et al. 2003), in college student samples (Haden et al. 2008), and in forensic samples (Kockler et al. 2006). Due to the mixture of college students, offenders and demographic controls in this study, I used items that loaded on the same factors in at least two of the three aforementioned studies (see Table S2 for the items). PM was measured with 8 items (Cronbach's α = 0.84) and IA with 7 (Cronbach's α = 0.84). Scale scores were computed as the mean of the items, with higher scores indicating greater levels of each aggression subtype. Observed minimum and maximum values for both scales are reported in Table 1.

TABLE 1.

Partial correlation among skin conductance level (SCL), respiratory sinus arrhythmia (RSA), and other main variables controlling for the non‐focal aggression.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1. Aggression a		−0.14	−0.01	−0.03	−0.15	−0.13	0.03	−0.07	−0.11	0.00	0.05	0.26**	0.02	−0.03	0.00
2. Age	0.22*		0.28**	−0.14	−0.12	0.09	−0.24**	−0.06	−0.19+	−0.03	0.07	−0.30**	−0.01	0.15	0.15
3. Housing assistance in youth	0.08	0.29**		0.15	0.00	−0.16	−0.15	0.05	−0.03	0.04	0.06	0.04	−0.03	−0.05	−0.02
4. Respiration rate at rest	0.04	−0.14	0.15		0.62**	−0.03	0.28**	0.07	0.00	0.13	0.10	−0.19+	0.19+	0.28**	0.21*
5. Respiration rate‐ preparation	0.05	−0.13	0.00	0.62**		−0.04	0.51**	−0.07	0.11	0.04	0.05	−0.13	−0.19+	0.06	−0.08
6. Respiration rate‐speech	0.21*	0.12	−0.14	−0.03	−0.05		0.21*	0.17*	0.06	0.09	0.05	−0.16	−0.07	−0.10	0.01
7. Respiration rate‐arithmetic	−0.07	−0.25**	−0.15	0.28**	0.51**	0.20*		0.19+	0.11	0.06	0.09	−0.10	−0.05	−0.10	−0.36**
8. SCL‐Baseline	−0.12	−0.10	0.04	0.06	−0.08	0.13	0.20*		0.04	0.26**	0.27**	0.02	−0.04	−0.13	−0.13
9. SCL‐reactivity preparation	−0.02	−0.21*	−0.04	0.00	0.10	0.04	0.11	0.04		0.55**	0.48**	0.14	−0.02	−0.07	−0.13
10. SCL‐Reactivity speech	−0.07	−0.04	0.03	0.13	0.04	0.08	0.07	0.27**	0.55**		0.88**	−0.04	0.11	0.05	0.01
11. SCL‐reactivity arithmetic	−0.06	0.07	0.06	0.10	0.06	0.05	0.09	0.28**	0.49**	0.88**		−0.08	0.08	0.06	0.02
12. RSA‐Baseline	−0.26**	−0.32**	0.02	−0.19*	−0.10	−0.18+	−0.09	0.07	0.17+	−0.03	−0.07		−0.13	−0.40**	−0.16
13. RSA‐reactivity preparation	0.07	0.01	−0.02	0.19+	−0.18+	−0.05	−0.05	−0.04	−0.02	0.11	0.07	−0.15		0.48**	0.47**
14. RSA‐reactivity speech	0.17+	0.18+	−0.04	0.28**	0.06	−0.06	−0.11	−0.15	−0.08	0.04	0.05	−0.43**	0.48**		0.61**
15. RSA‐reactivity arithmetic	0.11	0.16+	−0.01	0.22*	−0.08	0.03	−0.36**	−0.15	−0.13	0.00	0.01	−0.19+	0.47**	0.62**
Mean	2.44 b ; 2.48 c	28.11	0.38	16.36	17.00	13.89	18.48	10.67	0.74	2.00	1.57	6.75	−0.22	−0.19	−0.51
Standard deviation	0.97 b ; 0.99 c	10.93	0.49	3.56	3.15	3.04	3.73	4.80	1.73	1.97	1.77	1.17	0.60	0.81	0.91
N	130	130	130	129	130	130	128	112	105	104	104	130	130	130	128
Minimum	1	18	0	5.06	7.21	7.19	9.63	3.02	−7.97	−2.83	−3.99	3.38	−2.36	−2.44	−3.22
Maximum	5	60	1	24.74	26.95	23.89	26.16	27.51	6.82	9.75	7.27	9.89	1.38	2.10	1.78

^a All values are partial correlations from two separate partial correlation matrices. Values above the diagonal control for impulsive aggression; values below the diagonal control for premeditated aggression. The control variable applies to every pair within each triangle, including pairs in which neither variable is an aggression measure, which is why the same physiological pair can differ slightly above and below the diagonal. Row and column 1 accordingly contain the partial correlations of premeditated aggression with each variable controlling for impulsive aggression (above the diagonal) and of impulsive aggression with each variable controlling for premeditated aggression (below the diagonal). Zero‐order correlations among all study variables are reported in Table S4.

^b The mean or standard deviation for PM.

^c The mean or standard deviation for IA. Respiration rate is measured in the unit of Breaths per minute (BPM), SCL in μs (microsiemens), and RSA is measured as power in the high frequency band (0.15–0.40 Hz) in the unit of ms².

⁺ p < 0.1, *p < 0.05, **p < 0.01.

The Stress Task was modified from the widely used Trier Social Stress Task (TSST; Kirschbaum et al. 1993). Participants completed the task in the laboratory that was air‐conditioned to 72 to 75° F in order to maintain a consistent ambient temperature. In the modified TSST (mTSST), participants were told that they would deliver a speech and engage in arithmetic work during which time they would be evaluated by an expert on their extraversion traits. They were given 2.5 min to prepare the speech for a mock job interview for their dream job and then 2.5 min to deliver it in front of an expert (experiment accomplice) for evaluation purposes. Participants were made aware that they were being audio recorded to increase their stress level. Next, participants were to count backward from 749 in 6 s as quickly as possible without making mistakes in their head for 2 min, and they would report the number out loud whenever prompted by the computer. The experiment accomplices who assumed the role of the expert were trained to maintain a stoic appearance and gave verbal prompts (e.g., “you are not going fast enough, start over”) at standard intervals to increase social evaluative threat and stress.

The modifications to the TSST were made to accommodate my study samples and practical constraints. First, the task was shortened to fit within the time frame of the larger assessment plan of the project and to ensure that the offender and their demographic controls can engage the task to the maximal level without lapse of attention. Second, participants were asked to report the number out loud in the arithmetic session whenever prompted by the computer (four times in total), which was intended to ensure that the participants engaged while minimizing the talking during this session. Third, instead of a panel of three experts, only one expert along with a research assistant was present in the room because of the small room size. However, the confined space may help elicit a higher stress level. Last, throughout the task, a timer was presented on the computer screen with a loud ticking sound to create a pressing feeling.

Skin conductance (SC) at rest and during the mTSST was recorded using disposable snap electrodes with an 11 mm diameter Ag/AgCl contact surface that are pre‐gelled with isotonic gel. The electrodes were attached to the distal phalanges of the first and second fingers of the non‐dominant hand (Scerbo et al. 1992). Skin conductance signal was recorded using a low pass filter of 3 Hz, transmitted via a wireless transmitter to the amplifier Bionomadix PPGED, and sampled at 2000 Hz in MP160 with AcqKnowledge version 4.1 software (BIOPAC Systems Inc.). Potential artifacts were visually inspected and removed by interpolating the SC level over the artifact period using the data immediately prior to and after the artifact.

Respiratory Sinus Arrhythmia (RSA) was indexed by the power from the high frequency band in heart rate variability (HF‐HRV) following guidelines (Berntson et al. 1997). ECG was recorded using a single channel with a Lead II vector using three silver/silver chloride (Ag/AgCl) adhesive disposable electrodes placed at the right clavicle, left lower rib, and left clavicle as the ground electrode. Skin was prepared using abrasive skin prepping paste, and isotonic recording gel was used as the electrolyte medium. Impedance for ECG was checked to ensure it was below 5 kΩ before any data were collected. Respiration was recorded with the respiration belt transducer placed around the chest. Both ECG and respiration signals were digitized and sampled at 2000 Hz in MP160 with AcqKnowledge version 4.1 software (BIOPAC Systems Inc.). Signals were filtered with a bandpass FIR filter (ECG: 0.5 Hz and 35 Hz; Respiration: 0.5 and 1 Hz). ECG data were visually inspected for accurate R peak identification, and manual edits were conducted to correct misidentified or missed beats prior to HRV analysis. A fast Fourier transform was applied to each segment with a Hanning window, and the HF‐HRV was calculated by summing power spectral density values in the measurement unit of ms² over the bandwidth of 0.15–0.40 Hz, and then was log transformed with the natural number as the base. RSA was computed from the full duration of each task component that are 2 min and longer, consistent with minimum epoch length recommendations for short‐term frequency‐domain HRV analysis (Berntson et al. 1997; Malik et al. 1996).

The baseline SCL (SCL‐B) and baseline RSA score (RSA‐B) were computed by averaging SCL and RSA over a two‐minute period at the start of the laboratory session prior to all experimental tasks, during which participants were instructed to sit still and to look at a fixation point on the screen. The SCL reactivity (SCL‐R) and RSA reactivity (RSA‐R) were computed by subtracting the baseline from the reactivity task following the guideline for reactivity scores (Berntson et al. 1997). The higher SCL‐R scores indicate bigger SNS responses whereas negative values of RSA‐R indicate RSA withdrawal and positive values indicate augmentation.

2.3. Analytic Plans

Analyses were conducted in Mplus 8.4 (Muthén and Muthén 1998–2024) using maximum likelihood estimation with robust standard errors, and missing data were handled with full information maximum likelihood. Missing data rates varied across key variables. Aggression outcomes, age, housing assistance, and RSA variables had complete or near‐complete data (0–2 missing cases out of 130). SCL variables had the highest rates of missing data, ranging from 13.8% for baseline SCL to 20.0% for SCL reactivity variables, due to electrode displacement, broken leads, or signal artifacts identified during post hoc data inspection. Missing data rates were distributed similarly across the three recruitment groups (see N reported by groups in Supplemental Table s3), suggesting that missingness was not systematically related to group membership. To account for differences in aggression and in other predictors across the three recruitment groups, group membership was included as dummy variables with active offenders as the reference group, with college students and demographic controls as the two indicator variables.

Six separate regression models were estimated, three with PM as the outcome and three with IA as the outcome, one for each mTSST component: speech preparation, speech delivery, and arithmetic. In each model, the non‐focal form of aggression was included as a predictor to isolate the unique physiological correlates of each aggression subtype after accounting for their shared variance. The baseline predictors included in all six models were RSA‐B, SCL‐B, and their product term. The task reactivity predictors included RSA‐R, SCL‐R, and their product term, where the reactivity indices corresponded to the specific mTSST component serving as the outcome model's task context. All RSA and SCL indices were mean centered prior to computing product terms to facilitate interpretation of the coefficients and reduce multicollinearity.

All models were adjusted for age, housing assistance in youth as a proxy for early financial hardship, resting respiration rate, and the task‐matched respiration rate corresponding to the mTSST component of interest, following the same component‐matching logic described for the task reactivity indices above. Respiration rate was included as a covariate because respiratory rate influences HF‐HRV independently of vagal activity wherein slower breathing increases RSA amplitude through mechanical effects on cardiac cycle timing, and controlling for it helps ensure that RSA more accurately reflects vagally mediated parasympathetic activity (Berntson et al. 1997; Grossman and Taylor 2007). Task‐matched respiration rates were used because respiration rate varies across task components and its influence on HF‐HRV is component‐specific. The robustness of all primary findings was examined using peak‐to‐valley RSA as an alternative PNS index, and by re‐estimating models without adjusting for the non‐focal aggression.

Although the interaction term treats RSA and SCL symmetrically, for the purposes of probing the interaction, RSA was designated as the focal predictor and SCL as the moderator, reflecting the theoretical roles assigned to each system: RSA indexes vagally mediated regulatory capacity, which is the theoretically primary variable in differentiating PM from IA, while SCL indexes arousal level, which provides the motivational context within which regulatory capacity operates (Beauchaine 2015; Beauchaine and Thayer 2015). Reversing these designations would not alter the interaction coefficient or its significance. To probe significant interactions, regions of significance were identified by computing the conditional association between the focal predictor and the outcome across the observed range of the moderator using the MODEL CONSTRAINT command in Mplus (Muthén and Muthén 1998–2024), with regions of significance defined as the range of the moderator within which the 95% confidence interval for the conditional association excluded zero (Johnson and Fay 1950; Johnson and Neyman 1936). Simple‐effect slopes were computed by evaluating the association between the focal predictor and the outcome at high (mean + 1 SD) and low (mean −1 SD) values of the moderator and plotted in figures.

3. Results

Prior to the primary analyses, one‐way ANOVAs were conducted to examine group differences in all key study variables across the three recruitment groups (college students, demographic controls, and active offenders). Descriptive statistics and ANOVA results by group are presented in Supplemental Table s3. Impulsive aggression differed significantly across groups, with college students reporting lower IA than both demographic controls and active offenders. Premeditated aggression did not differ across groups. Groups did not differ significantly in baseline or reactivity physiological measures, with the exception of arithmetic respiration rate, which was higher among college students than the other two groups, and RSA reactivity during arithmetic, which showed greater withdrawal among college students relative to demographic controls.

Sixty percent of participants had RSA withdrawal in speech preparation, 65% in speech delivery, and 75% in arithmetic. Sixty‐seven percent of participants showed an increase in SCL in speech preparation, 94% in speech delivery, and 89% in arithmetic. On average, participants had a significant RSA withdrawal from baseline (M = 6.75, SD = 1.17) in speech preparation (M = 6.53, SD = 1.26; t = −4.16, p < 0.01, Cohen's d = 0.37), speech delivery (M = 6.56, SD = 1.12; t = −2.70, p < 0.01, Cohen's d = 0.23), and arithmetic (M = 6.24, SD = 1.34; t = −6.38, p < 0.01, Cohen's d = 0.57). Participants' SCL also increased significantly from baseline (M = 10.82, SD = 4.82) in speech preparation (M = 11.56, SD = 4.92; t = 4.41, p < 0.01, Cohen's d = 0.43), speech delivery (M = 12.60, SD = 5.42; t = 10.38, p < 0.01, Cohen's d = 0.93), and arithmetic (M = 12.23, SD = 5.61; t = 9.04, p < 0.01, Cohen's d = 0.81).

SCL‐B was significantly correlated with PM (r = −0.19) and IA (r = −0.23), and RSA‐B was significantly correlated with IA (r = −0.17) but not PM (r = 0.10). The PM and IA were strongly correlated (r = 0.54). The full bivariate correlation table is in Supplemental Table s4. Controlling for IA, PM was positively correlated with RSA‐B, whereas controlling for PM, IA was inversely correlated with RSA‐B (see Table 1).

Because PM and IA were correlated at r = 0.54 and each model adjusted for the non‐focal subtype, all coefficients reported in Tables 2 and 3 pertain to the variance unique to the focal aggression subtype after the variance shared with the other subtype has been removed. Adjusting for the effect of IA, the interaction between RSA‐B and SCL‐B was always significant in the association with PM, and RSA‐R interacted significantly with SCL‐R during speech delivery and arithmetic but not in speech preparation in association with PM (Models 1–3, Table 2). Because both terms were mean‐centered, the RSA‐B coefficient estimates the association at the sample mean of SCL‐B (10.67 μs) rather than an unconditional association; it was positive and comparable in magnitude across models but reached p < 0.05 only in the arithmetic model (β = 0.15, p = 0.047), remaining marginal in speech preparation (β = 0.13, p = 0.090) and speech delivery (β = 0.13, p = 0.085). This is consistent with the region‐of‐significance analysis reported below, in which the mean of SCL‐B falls near the upper bound of the significant region. Adjusting for the effect of PM, none of the RSA and SCL interactive terms was significant, but the negative RSA‐B coefficient was always significant in association with IA (Models 4–6, Table 3).

TABLE 2.

Interaction of skin conductance level (SCL) and respiratory sinus arrhythmia (RSA) on premeditated aggression.

Predictors	Model 1: speech preparation			Model 2: speech delivery			Model 3: arithmetic
	β (SE)	95% CI	p	β (SE)	95% CI	p	β (SE)	95% CI	p
Impulsive aggression	0.61** (0.08)	[0.46, 0.76]	< 0.001	0.63** (0.07)	[0.48, 0.77]	< 0.001	0.61** (0.08)	[0.46, 0.76]	< 0.001
Age	−0.12 (0.07)	[−0.27, 0.02]	0.099	−0.08 (0.07)	[−0.21, 0.06]	0.251	−0.13 (0.07)	[−0.28, 0.01]	0.069
Housing assistance	−0.09 (0.07)	[−0.22, 0.04]	0.183	−0.04 (0.08)	[−0.20, 0.12]	0.646	−0.04 (0.07)	[−0.18, 0.10]	0.611
College students (vs. offenders)	−0.11 (0.11)	[−0.32, 0.10]	0.313	−0.09 (0.10)	[−0.29, 0.11]	0.371	−0.12 (0.10)	[−0.32, 0.08]	0.232
Demographic controls (vs. offenders)	−0.06 (0.07)	[−0.18, 0.07]	0.396	−0.02 (0.06)	[−0.14, 0.10]	0.751	−0.04 (0.07)	[−0.17, 0.09]	0.536
Baseline respiration rate	−0.02 (0.10)	[−0.20, 0.17]	0.862	−0.05 (0.09)	[−0.23, 0.12]	0.555	−0.08 (0.09)	[−0.26, 0.10]	0.387
RSA‐baseline (RSA‐B)	0.13+ (0.08)	[−0.02, 0.27]	0.090	0.13+ (0.08)	[−0.02, 0.28]	0.085	0.15* (0.08)	[0.00, 0.30]	0.047
SCL‐baseline (SCL‐B)	−0.06 (0.07)	[−0.19, 0.09]	0.444	−0.03 (0.08)	[−0.18, 0.12]	0.694	−0.08 (0.09)	[−0.25, 0.09]	0.382
RSA‐B × SCL‐B	−0.18* (0.08)	[−0.32, −0.03]	0.017	−0.16* (0.07)	[−0.30, −0.02]	0.024	−0.18** (0.07)	[−0.32, −0.04]	0.011
Task respiration rate	−0.06 (0.09)	[−0.25, 0.12]	0.502	−0.17* (0.07)	[−0.31, −0.03]	0.017	0.08 (0.10)	[−0.12, 0.27]	0.430
RSA‐reactivity (RSA‐R)	0.01 (0.05)	[−0.09, 0.12]	0.828	−0.02 (0.06)	[−0.14, 0.09]	0.703	0.00 (0.06)	[−0.12, 0.12]	0.994
SCL‐reactivity (SCL‐R)	−0.10 (0.09)	[−0.27, 0.07]	0.228	0.04 (0.08)	[−0.12, 0.19]	0.643	0.01 (0.09)	[−0.16, 0.19]	0.880
RSA‐R × SCL‐R	0.11 (0.08)	[−0.04, 0.26]	0.138	−0.17* (0.07)	[−0.31, −0.02]	0.022	−0.15* (0.07)	[−0.28, −0.02]	0.029

Note: For Models 1–3, the task respiration rate covariate corresponds to the speech preparation, speech delivery, and arithmetic components of the mTSST respectively. SCL in μs, RSA in ms² (HF band 0.15–0.40 Hz). Active offenders serve as the reference group for group membership dummy variables.

Abbreviations: CI, confidence interval; SE, standard error.

+p < 0.10, *p < 0.05, **p < 0.01.

TABLE 3.

Interaction of skin conductance level (SCL) and respiratory sinus arrhythmia (RSA) on impulsive aggression.

Predictors	Model 4: speech preparation			Model 5: speech delivery			Model 6: arithmetic
	β (SE)	95% CI	p	β (SE)	95% CI	p	β (SE)	95% CI	p
Premeditated aggression	0.61** (0.07)	[0.48, 0.75]	< 0.001	0.63** (0.07)	[0.50, 0.77]	< 0.001	0.62** (0.07)	[0.48, 0.75]	< 0.001
Age	0.10 (0.08)	[−0.05, 0.25]	0.195	0.05 (0.08)	[−0.10, 0.21]	0.485	0.11 (0.08)	[−0.05, 0.27]	0.168
Housing assistance	0.02 (0.08)	[−0.13, 0.18]	0.767	0.01 (0.09)	[−0.16, 0.19]	0.879	−0.01 (0.08)	[−0.17, 0.15]	0.925
College students (vs. offenders)	−0.12 (0.10)	[−0.32, 0.08]	0.251	−0.13 (0.09)	[−0.31, 0.05]	0.146	−0.11 (0.10)	[−0.31, 0.09]	0.272
Demographic controls (vs. offenders)	−0.01 (0.07)	[−0.16, 0.13]	0.858	−0.05 (0.07)	[−0.19, 0.10]	0.526	−0.03 (0.08)	[−0.18, 0.13]	0.725
Baseline respiration rate	0.03 (0.09)	[−0.15, 0.22]	0.722	0.04 (0.08)	[−0.12, 0.20]	0.611	0.06 (0.09)	[−0.11, 0.22]	0.503
RSA‐baseline (RSA‐B)	−0.17* (0.07)	[−0.31, −0.02]	0.022	−0.18* (0.07)	[−0.32, −0.04]	0.012	−0.19** (0.07)	[−0.33, −0.04]	0.011
SCL‐baseline (SCL‐B)	−0.12 (0.08)	[−0.26, 0.03]	0.113	−0.11 (0.08)	[−0.27, 0.05]	0.174	−0.08 (0.09)	[−0.27, 0.10]	0.370
RSA‐B × SCL‐B	0.07 (0.07)	[−0.07, 0.21]	0.347	0.04 (0.07)	[−0.10, 0.18]	0.562	0.07 (0.07)	[−0.07, 0.21]	0.328
Task respiration rate	0.01 (0.10)	[−0.18, 0.20]	0.930	0.19* (0.07)	[0.04, 0.33]	0.011	−0.06 (0.09)	[−0.23, 0.12]	0.542
RSA‐reactivity (RSA‐R)	−0.01 (0.06)	[−0.13, 0.11]	0.869	−0.01 (0.06)	[−0.13, 0.11]	0.825	0.00 (0.06)	[−0.12, 0.13]	0.985
SCL‐reactivity (SCL‐R)	0.06 (0.08)	[−0.09, 0.20]	0.453	−0.09 (0.07)	[−0.22, 0.05]	0.212	−0.04 (0.09)	[−0.22, 0.13]	0.624
RSA‐R × SCL‐R	−0.09 (0.06)	[−0.21, 0.04]	0.168	0.09 (0.07)	[−0.05, 0.22]	0.196	0.10 (0.07)	[−0.05, 0.24]	0.180

Note: The task respiration rate covariate corresponds to the speech preparation, speech delivery, and arithmetic components of the mTSST respectively. SCL in μs, RSA in ms² (HF band 0.15–0.40 Hz). Active offenders serve as the reference group for group membership dummy variables.

Abbreviations: CI, confidence interval; SE, standard error.

+p < 0.10, *p < 0.05, **p < 0.01.

As seen in Figure 1, RSA‐B was positively associated with PM but only at low to moderate SCL‐B (10.96 μs and below): when SCL‐B was lowered by 1 standard deviation unit, the association between RSA‐B and PM increased by 0.17 standard deviation unit. The relation between RSA‐R and PM depends on the SCL‐R in the speech delivery and it was a cross‐over interaction in that not only the magnitude but also the direction of the relation changed across the range of SCL‐R. The association between RSA augmentation and PM was positive and significant among those with no or little SCL‐R (0.62 μs and below) and for those with one standard deviation unit higher in SCL‐R, the association between RSA augmentation and PM continued to decrease by 0.18 standard deviation unit; at high levels of SCL‐R (4.17 μs and above), the association between RSA withdrawal and PM was positive and significant and grew stronger at higher levels of SCL‐R. The findings for RSA‐R and SCL‐R in arithmetic on PM mirrored those in speech delivery (see bottom panel in Figure 1).

FIGURE 1.

SCL region of significance (patterned shaded area) for the association between baseline RSA and premeditated aggression (top panel), between RSA reactivity and premeditated aggression at speech delivery (middle panel), and arithmetic (bottom panel).

The levels of PM at various combinations of RSA‐SCL were illustrated with the simple‐effect slopes (Figure 2). The most PM was found among those with high RSA‐B and low SCL‐B, and among those with RSA augmentation and low SCL‐R (reciprocal PNS activity) or those with RSA withdrawal and high SCL‐R (reciprocal SNS activity) in speech delivery and arithmetic.

FIGURE 2.

Simple‐effect slopes illustrating predicted levels of premeditated aggression at combinations of baseline RSA and baseline SCL (Panel 1) and at combinations of RSA reactivity and SCL reactivity during speech delivery (Panel 2) and arithmetic (Panel 3). Predicted values of premeditated aggression are plotted at high (mean + 1 SD) and low (mean—1 SD) levels of SCL. Panel 1: The combination of high baseline RSA and low baseline SCL was associated with the highest levels of premeditated aggression. Panels 2 and 3: Two distinct profiles were associated with elevated premeditated aggression during social confrontation: Reciprocal PNS activation (RSA augmentation coupled with low SCL reactivity) and reciprocal SNS activation (RSA withdrawal coupled with high SCL reactivity), suggesting heterogeneity in the physiological correlates of premeditated aggression.

Peak‐to‐valley RSA reactivity correlated well with HF‐HRV reactivity at speech preparation (r = 0.58), speech delivery (r = 0.68), and arithmetic (r = 0.64), and all primary findings were replicated using pvRSA as an alternative PNS index (see Table S5). The regions of significance and simple‐effect slopes showed high resemblance across both RSA indices. When the non‐focal aggression was not adjusted for, the pattern of baseline findings changed in informative ways (see Table S6): without adjusting for IA, SCL‐B was inversely associated with PM; without adjusting for PM, SCL‐B was inversely associated with IA and the negative RSA‐B coefficient for IA was no longer significant. In both cases, all significant interaction terms in Tables 2 and 3 remain unchanged, indicating that the interaction findings are robust to whether or not the non‐focal aggression is covaried. A small number of participants had respiration rates falling outside the 9–24 breaths per minute range corresponding to the 0.15–0.40 Hz high‐frequency band across task phases (rest: n = 3; speech preparation: n = 1; speech delivery: n = 4; arithmetic: n = 5). Primary models were re‐estimated excluding these participants and all key findings remained unchanged, supporting the robustness of the reported associations.

4. Discussion

This is the first study to document the differential association between the PNS–SNS interplay and the bimodal classification of aggression in adults using a social stress paradigm with active social confrontation, and to do so in a heterogeneous sample spanning active offenders, community controls, and college students, substantially extending the range of aggression and physiological functioning beyond what convenience samples alone can provide. Premeditated aggression was associated with higher vagally mediated regulatory capacity among participants with low physiological arousal, and with both reciprocal PNS and SNS activation profiles in response to social confrontation during speech delivery and arithmetic. In contrast, impulsive aggression was uniquely and parsimoniously associated with lower vagally mediated regulatory capacity regardless of arousal level. These findings were replicated with peak‐to‐valley RSA and were robust to whether or not the non‐focal aggression was statistically adjusted for, strengthening confidence in their reliability.

4.1. Baseline Physiological Profiles and Premeditated Aggression

The following interpretations are offered in light of the associational and cross‐sectional nature of the findings, and should be understood as theoretically motivated rather than causally established. They should also be understood as pertaining to the variance unique to each aggression subtype. Because PM and IA were correlated at r = 0.54 and each was modeled with the other adjusted for, references below to the physiology of PM and of IA are shorthand for the correlates of what is distinctive to each subtype once their shared variance is set aside; results without this adjustment are reported in the sensitivity analyses (Table S6). Both the under‐arousal hypothesis and emotion regulation framework are needed to account for the complexity of PM. Consistent with the prediction, baseline RSA was positively related to PM at low to moderate levels of baseline SCL. The interaction finding, on which this conditional interpretation rests, was robust: it remained significant when task reactivity indices were simultaneously included in the model, when another RSA index was used, and whether or not IA was adjusted for. The stability of the baseline interaction in models that also included reactivity terms is particularly noteworthy because it indicates that the baseline RSA by SCL profile reflects a trait‐level physiological configuration that is associated with PM independently of how individuals actually respond during the stress task itself. The interplay of baseline RSA and SCL on aggression is not suitable to be organized into the autonomic space, even though autonomic space has been routinely employed to interpret findings when a mixture of baseline and reactivity measures were used (El‐Sheikh et al. 2009; Keller and El‐Sheikh 2009). The baseline in the autonomic space refers to the functional state of the targeted organ when both systems produce no change (Berntson et al. 1991, 1993), a condition not met here given that RSA and SCL index different end organs. Instead, the baseline findings are better interpreted in light of arousal and regulatory capacity.

The below‐average physiological arousal, indexed by baseline SCL, may be the necessary condition for the display of premeditated aggression, consistent with the under‐arousal hypothesis. Individuals with low physiological arousal show reduced orienting to environmental signals that typically inhibit aggressive behavior and are less responsive to punishment cues and social threat in ways that reduce behavioral inhibition generally (Lorber 2004; Raine 1993). They may also be less inhibited in general and prone toward sensation‐seeking to bring their physiological level to an optimal range (Zuckerman 1979). Low tonic arousal is therefore better understood as removing a key inhibitory check rather than as a direct motivational force toward aggression.

The calculated and deliberate aspect of PM has its physiological underpinning in vagally mediated regulatory capacity. Evidence has converged to show that baseline RSA serves well as an index of vagally mediated regulatory capacity associated with emotion regulation (Beauchaine 2015), executive functions, and cognitive performance (Thayer et al. 2009), and in regulation of energy exchange during metabolic and behavioral change (Grossman and Taylor 2007). Higher baseline RSA has been associated with various benefits including better behavioral regulation at social challenge (Hastings and De 2008; Hastings et al. 2008), better working memory and executive functions (Thayer et al. 2009), and better adjustment outcomes. Yet high baseline RSA is not solely beneficial as shown in the current study because it is associated with the premeditated form of aggression when coupled with low physiological arousal. The above‐average regulatory capacity may equip individuals who are less inhibited and prone to act aggressively to be calculated, deliberate, and purposeful in the expression of that aggression, providing the executive and regulatory scaffolding that enables aggression to be organized and directed toward a specific goal rather than discharged reactively in response to provocation. The absence of a significant main effect of baseline SCL on PM in the present study differs from Armstrong et al. (2019), who found resting skin conductance positively associated with proactive aggression in a college sample without testing its interplay with PNS activity. This discrepancy may reflect the residualization of aggression subtypes in the present analyses, which isolates the unique physiological correlates of PM after removing shared variance with IA. Consistent with this interpretation, when models were re‐estimated without adjusting for IA in our sensitivity analyses, an inverse association between baseline SCL and PM emerged, converging with Armstrong et al. (2019) and the broader under‐arousal literature. Importantly, the interaction between baseline RSA and SCL remained significant in these sensitivity models, suggesting that tonic arousal carries both independent and conditional relevance for PM depending on whether the overlap with IA is accounted for.

4.2. Reciprocal Activation Profiles and the Heterogeneity of Premeditated Aggression

The reactivity findings for PM reveal a cross‐over interaction in speech delivery and arithmetic in which both reciprocal PNS activation and reciprocal SNS activation were associated with more PM, while neither pattern emerged in speech preparation. The convergence across speech delivery and arithmetic, despite their substantial differences in vocalization demand and cognitive load, points to social confrontation and negative evaluation as the operative shared feature. Speech preparation involves neither social confrontation nor vocalization, evokes less stress, and carries lower cognitive demands, whereas both speech delivery and arithmetic place participants under direct social evaluative threat. The PNS‐SNS interplay responding to social confrontation therefore appears to be the relevant context for understanding PM at the situational level, and the absence of findings in speech preparation is not a null result but an informative boundary condition.

The basic tenets of the autonomic space are a useful framework for organizing the reactivity findings, in contrast to the baseline findings discussed above. Reciprocal activation profiles of PNS and SNS are the modal response to challenge, and it is the extreme versions of these normal profiles in response to social confrontation that were linked with PM in the present study. The two distinctive reciprocal profiles warrant separate interpretation because they reflect qualitatively different modes of physiological engagement with social threat, and they are distinguishable in their association with planned and calculated aggression.

The reciprocal PNS activation profile, characterized by RSA augmentation in the context of low or absent SCL reactivity, reflects physiological disengagement from the social confrontation. PNS augmentation in response to social challenge indicates that the organism is directing resources inward toward rest and restoration rather than mobilizing for active coping, and when combined with the absence of SNS activation this produces a state of pronounced physiological calm in the face of social threat (Porges 2007). The lower the SNS activation, the more the PNS augmentation is associated with PM, suggesting that the degree of disengagement from the social challenge is what matters rather than any single indicator examined alone. This profile bears a strong resemblance to the physiological signature associated with callous‐unemotional traits, which are characterized by blunted threat responding, reduced fear, and attenuated social orienting (Frick et al. 2014). Research on the psychophysiology of CU traits in adults has found reduced electrodermal responding to aversive and fear‐relevant stimuli (Chen et al. 2021), and more recently that individuals with high CU traits exhibit prolonged physiological engagement with immediate reward stimuli followed by rapid habituation as novelty diminishes (Chen et al. 2024), a pattern suggesting selective physiological allocation toward motivationally salient stimuli while social threat signals are filtered out. A parallel process may be operating in the present context. Individuals prone to PM who physiologically disengage during social confrontation may be preserving cognitive and motivational resources for the planned execution of aggression rather than expending them on emotional reactivity to the provocation itself. The disengaged physiological state that superficially resembles calm may in fact reflect the affective detachment that allows deliberate harmful action to proceed without the inhibitory interference of fear or empathic concern. This interpretation converges with Thomson et al. (2021), who found that proactive aggression in males was associated with coinhibition of both PNS and SNS during immersive fear induction, suggesting that physiological disengagement from threat may be a transcontextual marker of proactive aggression that generalizes across fear and social evaluative paradigms, and that the present findings extend to contexts of active social confrontation. Future research directly measuring CU traits and intermediate personality dimensions such as fearlessness and affective detachment would allow a more precise test of whether this profile is associated specifically with the callousness component of PM.

The reciprocal SNS activation profile, characterized by pronounced RSA withdrawal in the context of high SCL reactivity, reflects an organized and exaggerated mobilization of physiological resources to deal directly with the social challenge. RSA withdrawal during social confrontation is not inherently maladaptive; it typically facilitates focused attention, coordinated responding, and metabolic readiness for action (Holzman and Bridgett 2017; Porges 2007). What is notable in the present findings is that it is the extreme version of this profile, where both systems are simultaneously and strongly engaged in the classical reciprocal pattern, that is associated with more PM. The higher the SCL reactivity, the more the RSA withdrawal is associated with PM, suggesting that the degree of organized physiological mobilization is what differentiates individuals high in PM from those who show merely average reciprocal activation. This profile can be understood as an exaggerated but organized fight‐or‐flight response to social confrontation, one in which the individual is fully engaged with the challenge at a physiological level. The critical interpretive question is why this high‐arousal, high‐mobilization profile is associated with PM specifically rather than IA, which would be the more intuitive outcome for a strongly reactive physiological response to social threat. The answer likely lies in how the mobilized arousal is regulated and directed. Individuals who respond to social confrontation with strong coordinated physiological engagement but who nonetheless channel that response into planned rather than impulsive behavior may be those with an orientation toward dominance, power acquisition, and instrumental goal attainment, characteristics consistently linked to PM (Fite et al. 2010; Ramirez and Andreu 2006; Wagner and Abaied 2015). The strong physiological response to social challenge may in this case reflect not emotional dysregulation but a purposeful engagement with the confrontation as a domain for asserting dominance or pursuing a preconceived goal. The fact that this profile is associated with PM over and above IA, after their shared variance is removed, underscores that the relevant distinction is not the intensity of physiological engagement but the regulatory and motivational context within which that engagement occurs. Future studies incorporating measures of social dominance orientation, proactive coping strategies, and cold‐blooded psychopathic traits would allow a more refined test of this proposed profile and its distinction from the disengagement pathway described above.

That two physiologically opposite profiles both are associated with PM suggests that PM should not be understood as a unitary construct with a single physiological signature but as an outcome associated with at least two qualitatively distinct autonomic pathways: one characterized by disengagement and affective detachment from social threat, and another by organized, purposeful mobilization in response to social challenge. These two profiles may correspond to different developmental and personality substrates for PM, with the disengagement profile more closely linked to callous and fearless temperament and the mobilization profile more closely linked to dominance motivation and proactive coping under challenge. Both profiles are associated with the same behavioral endpoint, planned and purposeful aggression, but are distinguishable at the physiological level. This heterogeneity has both theoretical and practical implications. Theoretically, it suggests that aggregate associations between autonomic markers and PM in prior research may have been attenuated by the cancellation of opposite‐direction effects across these two subgroups, which could partly explain the inconsistent findings in the single‐marker literature reviewed earlier. Practically, it suggests that interventions targeting PM may need to address qualitatively different underlying processes depending on which physiological profile characterizes the individual, with the disengagement profile calling for approaches that build affective engagement and empathic responding, and the mobilization profile calling for approaches that develop regulatory control over goal‐directed arousal in social confrontation contexts.

4.3. No Reactivity Effects for Impulsive Aggression

In contrast to the complexity of the PM reactivity findings, no reactivity interactions reached significance for IA across any of the three task components. This absence of reactivity effects is not simply a null result to be noted and set aside but is theoretically meaningful in light of the baseline findings. IA was associated robustly and exclusively with low baseline RSA, suggesting that the relevant physiological process for IA is a stable trait‐level deficit in regulatory capacity rather than any particular pattern of situational response to social stress. Individuals prone to IA do not appear to be characterized by a distinctive way of responding to the specific features of social confrontation, whether in terms of how strongly they mobilize SNS resources or how much they withdraw PNS regulation. What distinguishes them physiologically, once the variance IA shares with PM has been removed, is a constitutional insufficiency in the vagally mediated regulatory resources that allow affective states to be modulated, deferred, and redirected, and this insufficiency expresses itself as a stable background condition rather than as a situationally triggered response pattern. This interpretation is consistent with the broader characterization of IA as driven by a general emotion regulation deficit that operates across situations rather than being selectively activated by particular provocation features in children (Card and Little 2006) and adults (Seo et al. 2008), and it reinforces the conclusion from the baseline section that low regulatory capacity is the specific and unique physiological signature of IA once its shared variance with PM is accounted for.

4.4. Implication of Sample Composition

The heterogeneous composition of the present sample warrants explicit consideration in interpreting the findings. One of the primary motivations for recruiting active offenders alongside demographic controls and college students was to extend the range of aggression and physiological functioning beyond what convenience samples alone can provide (Brezina and Topalli 2012; Topalli et al. 2013). The one‐way ANOVAs reported in Supplemental Table s3 indicate that this strategy was partially successful: impulsive aggression was significantly higher among active offenders and demographic controls than among college students, consistent with expectations given differential exposure to criminogenic environments and street culture (Raine 1993). It is worth noting that college students, who reported the lowest IA, also showed the highest baseline RSA among the three groups, a pattern directionally consistent with the negative RSA‐B and IA association reported in the primary analyses. However, group membership was included as a covariate in all models, and the association between RSA‐B and IA remained significant after accounting for group differences, suggesting the association reflects a within‐group physiological tendency rather than a between‐group compositional artifact.

PM, however, did not differ significantly across groups, suggesting that the deliberate and calculated form of aggression captured by the IPAS is distributed more broadly across the sample than its face validity might imply, consistent with evidence that PM is associated with psychopathic and callous traits that are not restricted to offending populations (Belfry and Kolla 2021; Fite et al. 2010). Importantly, the three groups did not differ significantly on any baseline or reactivity physiological measure with the exception of arithmetic respiration rate and RSA reactivity during arithmetic, which were both elevated among college students relative to the other groups and likely reflect differences in cognitive engagement with the arithmetic task rather than dispositional differences in autonomic regulation (Lorber 2004). The absence of group differences in the primary physiological predictors, including baseline RSA, baseline SCL, and most of their reactivity indices, suggests that the associations between autonomic profiles and aggression subtypes observed in the overall sample are not simply driven by group composition or confounded by systematic differences in physiological functioning across recruitment groups. The convergence of findings across a sample spanning a wide range of offending behavior therefore strengthens confidence that the reported associations reflect broader biosocial processes rather than sample‐specific artifacts, and supports the extension of PNS‐SNS interplay models of aggression to adult populations beyond the college convenience samples that have dominated prior research.

4.5. Limitations and Contributions

Several limitations should be noted. The temporal ordering of measurement limits causal inference: the IPAS captures aggressive acts occurring in the 6 months prior to the laboratory visit, while physiological measures were collected at a single subsequent laboratory session, meaning the independent variable was measured after the dependent variable. The associations reported here should therefore be interpreted as reflecting co‐occurring physiological and behavioral tendencies rather than prospective physiological predictors of aggression. The stress task is mild by design, and findings generalize only to stressors of similar intensity and social‐evaluative character. Vocalization during some task components may complicate RSA interpretation (Beauchaine et al. 2019; Berntson et al. 1997), though the convergence of key findings across speech delivery and arithmetic, which differ substantially in vocalization demand, reduces this concern. Although primary findings were robust to the exclusion of participants whose respiration rates fell outside the 9–24 breaths per minute range corresponding to the 0.15–0.40 Hz high‐frequency band, RSA estimates for these participants should be interpreted with appropriate caution. The absence of an a priori power analysis and preregistration should also be noted. The study was designed as part of a broader data collection effort prior to the current analysis plan, and the sample size was not determined based on the interaction effects that are central to the present findings. Interaction effects are generally harder to detect than main effects, and the current sample size may have been underpowered to detect smaller interaction effects. Results should therefore be interpreted with appropriate caution and replicated in larger samples. Relatedly, group membership was included as a covariate, which adjusts for mean differences across the three recruitment groups but does not test whether the physiology–aggression associations themselves differ by group. Such moderation would require three‐way interactions with subgroups of 38 to 47 participants, which the present sample cannot support. This is a meaningful gap given evidence that these associations may be sample‐dependent, with null findings in a delinquent young adult sample (Ter Harmsel et al. 2022) alongside significant associations in college and community samples (Armstrong et al. 2019; Puhalla and McCloskey 2020; Thomson et al. 2021), and testing group as a moderator is an important direction for future research. Generalizability is also limited by the all‐male sample. Because prior work reports sex‐differentiated autonomic correlates of aggression subtypes (Armstrong et al. 2019; Thomson et al. 2021), the profiles reported here should be regarded as male‐specific until tested in samples that include women. Finally, RSA and SCL index different end organs, which limits the precision with which the autonomic space framework can be applied to the baseline findings, as discussed above. Future research using cardiac‐linked PNS and SNS indices acting on the same end organ, such as RSA and pre‐ejection period, would allow a more rigorous test of autonomic space dynamics in relation to aggression subtypes.

These limitations are balanced by the contributions of the study. This is among the first studies to examine PNS‐SNS interplay at rest and during stress in relation to the bimodal classification of aggression in an adult sample including active offenders. The findings reveal that PM and IA have distinct physiological signatures: high regulatory capacity coupled with low tonic arousal characterizes PM at the trait level, while low regulatory capacity is the unique and unconditional physiological marker of IA. The extreme versions of both reciprocal activation profiles in response to social confrontation are further associated with PM, pointing to heterogeneity in its physiological correlates. Together these findings advance the bimodal classification of aggression and underscore the importance of multisystem models that distinguish arousal from regulatory capacity in understanding why aggression takes the form it does.

Author Contributions

Frances R. Chen: conceptualization, methodology, data curation, supervision, formal analysis, project administration, visualization, validation, writing – original draft, writing – review and editing, funding acquisition, investigation.

Conflicts of Interest

The author declares no conflicts of interest.

Supporting information

Table S1: Most serious crime type reported by recruitment group (open‐ended interview responses).

Table S2: Item‐level descriptive statistics for the impulsive and premeditated aggression scale by group.

Table S3: Descriptive statistics and one‐way ANOVAs for key study variables by recruitment group.

Table S4: Bivariate correlation among skin conductance level (SCL), respiratory sinus arrhythmia (RSA) and other main variables.

Table S5: Robustness analyses using peak‐to‐valley RSA: interaction of skin conductance level (SCL) and peak‐to‐valley RSA (pvRSA) on aggression.

Table S6: Robustness analyses without adjusting for the non‐focal aggression subtype: interaction of skin conductance level (SCL) and respiratory sinus arrhythmia (RSA) on aggression.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.