Interoceptive ability moderates the effect of physiological reactivity on social judgment

Abstract

Social judgments—that others are kind or cruel, well-intentioned, or conniving—can ease or disrupt social interactions. And yet a person’s internal state can alter these judgments—a phenomenon known as affective realism. We examined the factors that contribute to, and mitigate, affective realism during a stressful interview. Using data collected between 2015 and 2019, we hypothesized and found that individuals’ (N=161; 57.6% Female; 57.6% European American, 13.6% African American, 13.6% Asian American, 6.4% Latin(x), 6.0% biracial, and 2.8% that identified with none or 1+ of the races presented; Mage= 19.20 yrs) ability to accurately perceive their own internal sensations (i.e., heartbeats) influenced whether they attributed their own heightened stress reactions (i.e., sympathetic nervous system reactivity) to the behavior of two impassive interviewers. Participants who were poor heartbeat detectors perceived their interviewers as less helpful, polite, or professional, and more apathetic, judgmental, and aggressive when experiencing heightened levels of cardiovascular sympathetic nervous system reactivity during their interview. Being aware of one’s internal state may be one pathway to reducing bias in social perceptions in circumstances where such biases may lead us astray.

People constantly engage in “mind perception” by making best guesses about others’ intentions and dispositions (Waytz et al., 2010). A neighbor might seem “unfriendly,” a spouse “grumpy,” or a coworker “standoffish.” In contrast, a store clerk may seem “warm,” a boss “friendly,” or a friend “caring” (Abele et al., 2021; Feldman-Hall & Shenhav, 2019; Thornton & Tamir, 2020; Todorov et al., 2015). When these judgments are more accurate, they ease social interactions by enabling a perceiver to adjust their behaviors accordingly (e.g., comforting a grumpy spouse). However, when these judgments are less accurate, they can lead to inappropriate behaviors that derail social interactions (e.g., being aggressive to a coworker who means you no harm). Intervening on social judgments, however, is incredibly difficult. This is because social judgments are multiply determined and influenced by aspects of the context, target, and perceiver (Hehman et al., 2019).

For instance, it is well known that stored knowledge about social behavior (e.g., memories, stereotypes, internalized cultural norms) is an important source of information that both facilitates and subverts social interactions. These “social priors” interact with incoming external sensory information (e.g., a target’s facial morphology, the physical context; Aviezer et al., 2012; Hehman et al., 2019; Uleman et al., 2008; Wieser & Brosch, 2012) to shape social perception (Brooks et al., 2019; Fiske, 1998; see Hinton, 2017). For example, one’s stored knowledge about emotion concepts predicts which emotion category they perceive as emanating from another’s facial muscle movements (Brooks et al., 2019). This type of stored knowledge is acquired through learning (Doyle & Lindquist, 2018), and so one pathway to circumventing biased social perceptions may be by altering a person’s social knowledge (Lee et al., 2018; MacCormack et al., 2020).

Yet still other sources of information serve as “priors” that influence social perceptions. Another line of research demonstrates that altering a person’s moment-to-moment affective state may also exert unseen influence on social perception (Anderson et al., 2012; Clore & Huntsinger, 2007; Dutton & Aron, 1974; Wormwood et al., 2018). For example, individuals who are presented with pleasant stimuli outside reportable awareness perceive neutral target faces as more intensely smiling (vs scowling), warm, likable, reliable, and trustworthy (Anderson et al., 2012; Siegel et al., 2018; Wormwood et al., 2018). Although it has long been known that a person’s mood can bias their judgments and perceptions, little is known about the individual differences that produce such an effect. The present study examines for the first time in a laboratory-based social interaction, whether an individual’s ability to detect their own internal state interacts with their stress-induced physiological reactivity to alter social perceptions.

Affective Realism

Affect refers to simple feelings of pleasantness/unpleasantness (valence) and activation/deactivation (arousal) (Barrett, 2017; Barrett & Bliss-Moreau, 2009; Posner et al., 2005). These feelings are considered basic features of consciousness that imbue all objects of perception – including social others – with value. The influence of affect on perception is called “affective realism” (for description of effect see Barrett & Bar, 2009, term first introduced in Anderson et al., 2012). The result is that people, places, and things seem more unpleasant when one is experiencing unpleasant feelings and more activating when one is experiencing aroused feelings (Schwarz & Clore, 1983; Siegel & Stefanucci, 2011; Wormwood et al., 2018). Affect has long been linked to physiological states. Unsurprisingly then, body sensations (which themselves carry affective meaning) like hunger, inflammation, sexual arousal, and the timing of heartbeats can also contribute to experiences of affective realism. Individuals perceive strangers more negatively when hungry v. satiated (MacCormack & Lindquist, 2019) and are more likely to perceive sexual interest in others when sexually aroused v. calm (Bouffard & Miller, 2014). Social judgments even differ moment-to-moment within an individual. For example, when images of posed canonical fear expressions are time-locked to different parts of the cardiac cycle, faces presented during cardiac contraction (associated with greater subjective arousal) are judged as more threatening (Garfinkel et al., 2014).

Affective realism may seem maladaptive, but it likely evolved in service of allostasis. Allostasis is predictive regulation of the internal milieu. The brain efficiently enacts allostasis by anticipating (i.e., predicting) future bodily needs and preparing to meet them before they arise (Sterling, 2004, 2012; Sterling & Laughlin, 2015). For example, the brain constantly monitors the body’s energy stores and releases neuropeptides to generate feelings of hunger that motivate food intake well before blood glucose drops below dangerous levels. Predictive regulation also extends to perception. When we’re hungry, food is more appetizing and attention-grabbing (Reents et al., 2020). Using information about the current state of the body to drive perceptions of the outside world is advantageous in these non-social contexts because it aligns behavior with the body’s needs. Affective realism may also be helpful in social contexts. For example, when the immune system is actively attacking a virus, trusted others who can offer care seem especially appealing, but strangers are seen as especially threatening (Eisenberger et al., 2017). Sick individuals may therefore seek situations that offer resources and aid recuperation while also avoiding potential harm while physically vulnerable (Hennessy et al., 2014). However, social affective realism can also be destructive. For instance, law enforcement officers are biased to perceive citizens as more threatening and are more likely to unnecessarily use lethal force when they are experiencing elevated levels of cardiovascular activation (Andersen et al., 2018; Haller et al., 2014). It is thus important to know which factors moderate the influence of affective realism on social perception so people can harness it for good and engage proactively in harm-reduction strategies where affective realism may lead them astray.

Interoceptive Ability

Growing evidence suggests that an individual’s ability to detect their own internal bodily sensations – called interoceptive ability – influences their tendency to experience affective realism. For example, individuals who are less attuned to interoceptive signals are more likely to use their own emotional states when inferring the emotions of others (von Mohr et al., 2021). More generally, such individuals have difficulty differentiating their own bodily sensations from those of others’. They are more likely to subjectively experience a fake rubber hand as their own, to perceive someone else’s face as their own, or to perceive someone else’s body as their own when these ‘fake’ body parts are stroked synchronously with their own (e.g., Rubber Hand Illusion, Enfacement Illusion, Body Swap Illusion; Babo-Rebelo et al., 2019; Tajadura-Jiménez & Tsakiris, 2014; Tsakiris et al., 2011). Taken together, these findings suggest that individuals with poor interoceptive ability may be more likely to use affect as information when making other-focused judgments.

To date, only one study we are aware of has tested the hypothesis that individuals with worse interoceptive ability are more likely to experience affective realism (Feldman et al., 2022). In this study, interoceptive ability was measured using the gold-standard heartbeat detection task wherein participants are asked to judge the coincidence of their heartbeats with auditory tones (Whitehead et al., 1977). To measure affective realism, researchers used an ostensibly “subliminal” priming task – not unlike common sequential priming tasks (e.g., the affect misattribution procedure; Payne & Lundberg, 2014)—in which structurally neutral faces (i.e., faces devoid of canonical emotion expression) were paired simultaneously with positive or negative images suppressed from reportable awareness. Suppression was achieved using a method from vision science in which a static image presented to one eye is consciously suppressed while the other eye processes a set of continuously flashing stimuli (Tsuchiya & Koch, 2005). Affective realism was operationalized as participants’ tendency to rate target faces displaying neutral expressions as relatively more pleasant vs unpleasant when paired with suppressed positive vs negative images. Individuals in this study with worse interoceptive ability were more biased by suppressed affective images while making social judgments. These initial findings were interpreted as evidence that individuals with poor interoceptive ability are more susceptible to experiences of affective realism. Here we built off this previous laboratory-based task using static images to test the effect of interoceptive ability on affective realism during a more ecologically valid social interaction between a participant and two interviewers.

The Present Study

In the present study, we measured participants’ cardiovascular interoceptive ability, their cardiovascular responding during a stressful interview, and their social perceptions of two interviewers. Specifically, participants completed the Trier Social Stress Test (TSST; Kirschbaum et al., 1993) in which they were asked to give an impromptu speech in front of a panel of impassive interviewers. We measured continuous cardiovascular reactivity throughout and then assessed participants’ self-reported negative high arousal emotions, perceptions of task difficulty, and social perceptions of their interviewers immediately after. While the TSST is designed to be overtly difficult and stressful, interviewers’ behaviors are trained to be ambiguous. Prior work has suggested that affective realism effects are most extreme under conditions of source and target ambiguity. These conditions are thought to lead subjects to “use” their “affect as information” (see Payne & Lundberg, 2014). Given this, we hypothesized that those worse in interoceptive ability would be more likely to attribute their affective reactions during the stressful task to the interviewers’ behaviors and dispositions. In contrast, we reasoned that individual differences would not influence perceptions of the task itself (e.g., difficulty) or participants’ self-reported emotions (e.g., unpleasant, high arousal affect).

Methods

The present investigation is a secondary analysis of data collected between 2015 and 2019 as part of a larger study with a total sample of 250 young adults. Participants were recruited in the United States from the University of North Carolina at Chapel Hill (UNC-CH) Department of Psychology and Neuroscience introductory psychology course participant pool (57.6% Female; 57.6% European American, 13.6% African American, 13.6% Asian American, 6.4% Latinx, 6.0% biracial, and 2.8% that either identified with more than one race or with none of the races presented; Mage= 19.20 years, SDage= 1.29 years ranging from 17 to 29 years old). Additional items were added to the experimental procedure after data had already been collected from about one fourth of the final sample. Consequently, of the original 250 participants, 185 participants completed all measures relevant to the present analyses and of those, 161 participants provided complete and useable data. Final sample demographics can be found in Table 1.

Table 1.

Final participant demographics (N = 161)

Variable	n (%) or mean (SD)
Gender 1
Female	95 (59.0%)
Male	66 (41.00%)
Age (years)	19.25 (1.31)
BMI (kg/m2)	22.86 (3.03)
Race
American Indian & Alaskan Native	0 (0.00%)
Asian American	18 (11.20%)
Native Hawaiian or other Pacific Islander	0 (0.00%)
Black/African American	23 (14.3%)
White/European American	101 (62.7%)
Latin American	8 (5.00%)
More than one race	10 (6.20%)
Other race	1 (0.60%)

All participants were screened for conditions that could impact stress reactivity or autonomic physiology (e.g., psychiatric illness, heart conditions or pacemakers, eating disorders, or body mass index (BMI) greater than 33). Participants completed two laboratory sessions. Prior to each session, participants were given instructions to avoid certain foods, substances, and health behaviors that could impact their autonomic physiology. Participants who failed to follow these instructions, who were currently ill, or who were taking any sort of medication were either excluded or rescheduled (full eligibility criterion can be found in the supplemental online materials; SOM).

Open Practices Statement.

The analyses reported in this article are secondary analyses of a data set reported in (MacCormack, Bonar, & Lindquist, under review). Hypotheses were developed based on the findings of Feldman et al. (2022) but were not formally pre-registered. Materials have been made available in the SOM. Data and full code for analyses, diagnostics, and plots are available at https://osf.io/8ys5a/.

Procedure.

This study was approved by an Institutional Review Board at the University of North Carolina at Chapel Hill. Compensation was four hours’ worth of course-credit. During session one, participants reported on their current health and completed informed consent before being instrumented for electrocardiographic (ECG) measures for a 5-minute resting baseline period. Participants then completed three counterbalanced tasks: the modified Whitehead heartbeat detection task (Whitehead et al., 1977), a survey evaluating interoceptive awareness and beliefs (reported in MacCormack, Bonar, & Lindquist, under review), and a behavioral reaction time task (unrelated to the present investigation).

During session two, participants were again asked to report on their current health before completing an open-ended feelings report. Next, participants were instrumented for electrocardiographic and impedance cardiographic measurements during a second 5-minute resting baseline period. As required by the UNC-CH IRB, participants were then provided a second informed consent document. This document told subjects that they would be completing a series of cognitive behavioral tasks that included public speaking. After providing consent, participants completed the gold-standard Trier Social Stress Test (Kirschbaum et al., 1993) followed by a second open-ended feelings report and a final set of questionnaires in which they were asked to report on their own emotions and make situational and social judgments. Participants were then debriefed and compensated for their time.

Measures

Heartbeat Detection (HBD) Task.

We assessed cardiovascular interoceptive ability using the modified Whitehead heartbeat detection task (HBD; Kleckner et al., 2015; Whitehead et al., 1977) programmed using a custom software in MATLAB (developed by Kleckner et al., 2015) in combination with the Mindware heartbeat detection software module (v. 3.0.13). On each of sixty trials, participants heard ten tones through headphones. Tones were either played coincident with the subject’s heartbeat (approximately 200 ms after ventricular depolarization) or not coincident with the subject’s heartbeat (approximately 500 ms after ventricular depolarization). After each trial, participants indicated “yes” or “no” as to whether the tones did or did not coincide with their heartbeat. Prior research has demonstrated that 60 trials of a heartbeat detection task should be sufficient to detect a medium effect size for our sample size (Kleckner et al., 2015). Trials where reaction times were faster than 200 ms or slower than 700 ms were excluded from analyses (5.11% total trials). Four of the 185 subjects who completed sessions one and two and who completed all study components were excluded from the final sample for missing HBD data due to equipment malfunction. Interoceptive ability was operationalized as the percent of correct trials, or accuracy (see SOM for complementary models run using signal detection d’ and a’).

Trier Social Stress Test (TSST).

The Trier Social Stress Test (TSST) is a motivated performance task that is known to robustly increase subjective experiences of unpleasantness as well as cardiovascular activity via activation of the sympathetic nervous system (SNS; Kudielka et al., 2007). The TSST consists of two components: a speech task and a surprise mental math task. Prior to beginning these tasks, research assistants invited two “interviewers” into the testing room. These interviewers were described as experts in nonverbal communication, public performance, and cognitive ability. Interviewers dressed professionally and wore white laboratory coats over their clothing. Interviewers informed participants that they would be giving a ten-minute speech as part of a hypothetical interview for their dream job. Participants were given 2 minutes to mentally prepare. Afterwards, interviewers re-entered the room and instructed participants to begin. Participants were required to speak for the entire ten minutes as interviewers watched with neutral expressions–providing no feedback or encouragement. Following the speech task, participants were told they would be completing a mental math task. Specifically, participants were asked to count out loud from the number 996 backwards in steps of seven as fast as they could while making as few errors as possible. If participants provided an erroneous answer, they were asked to start again from 996. This task took five-minutes total and was modified systematically for participants who found the math too easy or too difficult. Continuous physiology was acquired throughout.

Physiological Data Acquisition and Processing.

To derive our measure of cardiovascular sympathetic nervous system reactivity, cardiovascular data were acquired using Mindware Technologies (Gahanna, OH, USA) Biolab software before, during, and after the TSST. Both electrocardiography and impedance cardiography data were obtained using pre-gelled ConMed (Westborough, MA) Cleartrace Ag/AgCL sensors connected via wires to a Mindware Bionex box. Specifically, electrocardiography data were collected from three non-invasive spot electrodes placed in a modified lead-II configuration. One electrode was placed distally on the right collarbone and two electrodes were placed on the lower-most ribs (one on either side of the ribcage). Impedance cardiography data was acquired using the four-spot electrode configuration described in Qu et al (1986). Two inner recording electrodes were placed on the participant’s front: one at the base of the neck at the top of the sternum, and one at the bottom of the sternum over the xiphisternal junction. Two outer recording electrodes were placed on the participant’s back: one approximately 4 cm above the base of the neck and one approximately 4 cm below the bottom of the sternum.

All data was visually inspected and scored by trained scorers according to field guidelines (Sherwood et al 1990, Jennings et al 1981) using Mindware Technologies’ Heart Rate Variability (v3.021) and Impedance (v3.2.4) analysis software. For all analyses, physiological measures reflect individual change scores from the 5-minute baseline taken immediately prior to the TSST. Physiological activity during the TSST was averaged across the first minutes of speech preparation, speech, and mental math (when effects of induction on heart rate were likely to be near their peak, e.g., von Dawans, Kirschbaum, & Heinrichs 2011) before being baseline corrected. Where participants were missing data during the first minute of these tasks, the second minute was substituted.

As our cardiovascular SNS index, we computed pre-ejection period (PEP) reactivity derived from the electrocardiograph and impedance cardiograph data. PEP reflects the time (in milliseconds) between the electrical impulse signaling contraction of the left ventricle and the opening of the aortic valve. Smaller PEP intervals suggest periods of increased cardiac contractility. Such changes are predominantly controlled by sympathetic innervation of the heart (Berntson et al., 2017; Cacioppo et al., 2017), making PEP a relatively pure measure of SNS activity and an intuitive measure of cardiovascular arousal (see Satpute et al., 2019). For ease of interpretation, we inverted PEP in all subsequent analyses, tables, and figures, so that higher values would signal greater increases in SNS activity. Twenty-one of the 185 subjects who completed sessions one and two and who completed all study components were excluded from the final sample for missing PEP data due to poor physiological signal quality (62%) or technical malfunction (38%).

Self-Reported Judgments and Experiences.

Before and after the TSST, participants were provided a free text field to describe their current emotion. Immediately following the TSST, participants completed a set of self-report items assessing their judgments of the interviewers, task, and their own emotional experience.

Open Ended Feelings Reports.

Before and after completing the TSST, participants were instructed to, “Please write out in your own words how you feel right now. Please be as SPECIFIC and DESCRIPTIVE as possible about how you feel.” On average, subjects wrote 43.12 words (SD = 24.71). All open-ended feelings reports were word tokenized (i.e., split into vectors of single words or “tokens”) and merged with an open-source semantic lexicon using the tidytext package (version 0.3.1) in R (Silge & Robinson, 2016). Semantic lexicons are large dictionaries of words which have been normed along various semantic dimensions. In the present analyses we used the NRC Valence, Arousal, and Dominance (NRC-VAD) Lexicon (Mohammad, 2018). The NRC-VAD Lexicon is a dictionary containing valence and arousal norms for more than 20,000 English words. Norms were generated from manual annotations and ranged from 0 (negative valence/low arousal) to 1 (positive valence/high arousal). We chose this lexicon because it had the greatest representation of words found in the feelings reports (M = 40%, SD = 10%). Valence and arousal scores were averaged across all available words in each of the feelings reports to generate a proximal measure of individuals’ mental state before and after the TSST. For ten subjects, the NRC-VAD had particularly poor coverage (i.e., normative ratings for <5 words). Feelings reports for these ten subjects, and for one additional subject who was missing data, were excluded from analyses. As these lexical measures likely underestimate experienced emotion, we used these indices as a conservative means of evaluating the efficacy of the TSST at inducing negative high arousal affect.

Social Judgments.

After the TSST, participants were asked to rate their interviewers under the cover story that this information would be used by the project’s PI to evaluate interviewers’ performance in the lab. Specifically, participants rated the extent to which they felt their interviewers were “aggressive,” “helpful,” “lacking empathy,” “professional,” “polite,” and “judgmental.” All social evaluative judgments were made on 7-point Likert scales (0 = “Not at All,” 1 = “Slightly”, 3 = “Somewhat,” 5 = “Quite a Bit,” 6 = “Extremely”). Importantly, participants were informed that these social evaluative judgments were a “normal part of every experiment” and were encouraged to provide candid responses. Because we did not have specific hypotheses about individual social appraisals (e.g., “aggressiveness” versus “politeness”), all negative items were reverse scored and social judgments were averaged such that higher scores indicated more positive evaluations (std. α= 0.71).

Task Judgments.

After the TSST, participants were asked to evaluate the speech and math tasks on their difficulty and stressfulness. All ratings were made on 6-point Likert scales (0 = “Not at All,” 1 = “A Little Bit,” 2 = “Somewhat,” 3 = “Moderately,” 4 = “Very,” 5 = “Extremely”). Because we did not have specific hypotheses about individual situational appraisals (e.g., “difficulty” versus “stressfulness”), responses on these two items were averaged such that higher scores indicated more negative task judgments.

Emotion Endorsements.

After the TSST, participants endorsed which of a series of emotion words they had experienced during the task. Words were selected from an expanded 30-item version of the Positive & Negative Affect Schedule (PANAS; Watson et al., 1988) and captured emotional experiences across the four quadrants of the affective circumplex (high arousal positive, high arousal negative, low arousal positive, and low arousal negative emotions). Participants were asked to rate how intensely they were experiencing each emotion on a 7-point Likert scale (0 = “Not at All,” 1 = “A Little Bit,” 2 = “Somewhat,” 3 = “Moderately,” 4 = “Quite A Bit,” 5 = “Very Much,” 6 = “Extremely”). For the present analyses, we focus on experiences of high arousal negative affect. Participants’ negative high arousal affect was computed as their average endorsement across the emotion words: “embarrassed,” “stressed,” “annoyed,” “irritable,” “panicky,” “distressed,” “frustrated,” “afraid,” “angry,” “guilty,” “disgusted,” “anxious,” “hyperactive,” and “unhappy” (std. α= 0.91).

Data Analysis

Manipulation check.

To establish the effectiveness of the TSST at inducing stress, we first performed two-tailed paired-samples t-tests assuming equal variance to evaluate differences in text-based estimates of valence and arousal from participant’s open-ended feelings reports prior to and after the TSST.

We next performed a two-tailed paired sample t-test on SNS activity prior to and during the first minute of the TSST tasks. A significant Levene’s test for homogeneity of variance suggested that population variance for SNS activity could not be assumed equal before and during the TSST; F(1,298) = 3.88, p = 0.05. Consequently, we conducted a two-tailed paired samples t-test using a Welch modification to the degrees of freedom.

Hypothesis tests.

Models testing main hypotheses were run using hierarchical multiple regression implemented using the stats package (version 3.6.2) in R (R Core Team, 2019). All continuous predictor variables were mean centered. Models were run in three steps. In the first step, age, gender, and BMI were entered as covariates since all are known to correlate with independent or dependent variables of interest. At step two, SNS reactivity (−1*PEP reactivity) and HBD accuracy were entered as predictors with their interaction term entered at step three. Significant interactions were probed using the interactions package (version 1.1.5) in R (Long, 2019). Specifically, we evaluated the simple slopes of SNS reactivity at mean levels of HBD accuracy (M = 60%) +/− one standard deviation (+1SD = 72%; −1SD = 48%). All models met assumptions of linearity, normality, and homoscedasticity. No observations demonstrated undue statistical influence (estimated using Cook’s distance) for any of the models reported.

Specificity tests.

Affective realism is thought to be a domain-general phenomenon. While our hypotheses focus on the effects of affective realism for social judgments, it is possible that we could have observed evidence of affective realism elsewhere. For example, it is possible that individuals with greater physiological reactivity and lesser interoceptive accuracy might rate the TSST tasks as more difficult or report greater negative high arousal affect. However, prior work has suggested that affective realism effects are most extreme under conditions of source and target ambiguity (see Payne & Lundberg, 2014). Given this, we hypothesized that those worse in interoceptive ability would be more likely to attribute their affective reactions during the stressful task to the interviewers’ behaviors and dispositions (because participants lacked clear verbal or non-verbal clues regarding the traits and dispositions of their interviewers). In contrast, because the TSST tasks are designed to be overtly stressful, we reasoned that individual differences would not influence perceptions of the task itself (e.g., difficulty) or participants’ self-reported emotions (e.g., unpleasant, high arousal affect). Consistent with prior work (Kirschbaum, 1993), we reasoned that, on average, most people would find the task difficult and would report high levels of stress. To evaluate these hypotheses, we performed two follow-up sensitivity tests to examine whether the effects of interoceptive ability, sympathetic nervous system activity, and their interaction impacted (a) perceptions of the TSST and (b) endorsements of negative high arousal emotion following the TSST. These models were run using the same methods described above.

Power:

This study was originally powered to detect small-moderate main effects and small interactions of heartbeat detection accuracy and interoceptive beliefs on emotional arousal (MacCormack, Bonar, & Lindquist, under review). A post-hoc sensitivity analysis performed in G*Power (Erdfelder et al., 1996) suggested adequate power (>= 0.80) with an alpha = 0.05, sample size = 161, to detect a small to moderate main effect. Interaction effects mathematically multiply noise in measured variables and require much larger sample sizes (Schondbt & Perugini, 2013; Gelman, 2018). As such our sample was likely underpowered to detect small-medium interaction effects in this secondary data analysis.

Results

Descriptive statistics and bivariate correlations for all continuous variables of interest can be found in Table 2. Consistent with prior studies, most participants performed slightly above chance level on the heartbeat detection task, with participants one-standard deviation below the mean performing below chance, and participants one-standard deviation above the mean performing above chance. T-tests revealed significant gender differences in age and both negative high arousal affect and task judgments following the TSST. Specifically, Men in our sample tended to be a bit older (p = 0.04). They also experienced less intense negative high arousal affect (p < 0.001) and reported less negative task judgments (p < 0.001) following the TSST.

Table 2.

Descriptive statistics and bivariate correlations for variables of interest

	Descriptives						Correlations (r)
Parameter	Mean	SD	Min	Max	SNS	Social	Task	NH	HBD Acc	BMI
Age (years)	19.25	1.31	18.00	29.00	0.11	−0.04	0.08	−0.12	0.01	0.14
BMI (kg/m2)	22.86	3.03	16.44	31.61	−0.10	0.12	0.08	−0.03	−0.19	--
HBD Accuracy	0.60	0.12	0.36	1.00	0.05	−0.02	0.11	−0.18	--	--
Neg High Arousal Affect	1.84	1.17	0.07	5.07	0.08	−0.11	−0.70*	--	--	--
Task Judgments	2.33	0.90	0.67	4.50	−0.07	0.04	--	--	--	--
Social Judgments	3.70	1.24	0.33	6.00	−0.08	--	--	--	--	--
SNS Reactivity	10.95	11.60	−24.67	53.33	--	--	--	--	--	--

Note: Heartbeat Detection (HBD) Accuracy reflects a ratio and so is bounded between 0 and 1.00. Negative high arousal affect (derived from the PANAS) and social judgments were rated on a scale ranging from 0 to 6. Task Judgments were rated on a scale ranging from 0-5. Higher values for social judgments indicate more positive ratings. Higher values for task judgments and negative high arousal affect indicate more negative task judgments and more intense negative high arousal affect respectively. Sympathetic Nervous System (SNS) Reactivity and is operationalized as pre-ejection period * −1. Bivariate correlations between continuous covariates and dependent variables are presented. Warmer colors indicate negative correlations, cooler colors indicate positive correlations. Darker colors indicate larger r-statistics.

^*= p<0.05.

Manipulation Check.

Text-based analysis of open-ended feelings reports revealed evidence for increases in subjective negative valence (Figure 1, Panel A) and arousal (Figure 1, Panel B) following the TSST. Confirming the effectiveness of the TSST, participants used significantly more negatively valenced language after (M = 0.44, SD = 0.09) compared to before (M = 0.36, SD = 0.09) the TSST; t(149) = 9.79, p < 0.001, d = 0.80. Participants also used significantly more aroused language after (M = 0.48, SD = 0.06) compared to before (M = 0.44, SD = 0.05) the TSST; t(149) = 7.50, p < 0.001, d = 0.61.

Figure 1.

Average valence (panel A) and arousal (panel B) before and after the Trier Social Stress Test (TSST); SNS activity (−1*PEP) before and during the first minute of the TSST (panel C). Text-based analysis of open-ended feelings reports revealed evidence for increases in negative valence and increases in arousal during the TSST. Analysis of physiological data revealed increases in SNS activity (i.e., decreases in PEP) during the TSST.

Further confirming the effectiveness of the TSST, analysis of physiological data also revealed increases in SNS activity during (M = −109.80, SD = 13.80) compared to before (M = −120.92, SD = 12.19) the TSST; t(149) = 11.73, p < 0.001, d = 0.96 (Figure 1, Panel C).

Hypothesis Tests - Social Judgments.

Neither age, Gender, nor BMI significantly covaried with social judgments (see Table 3). Likewise, neither sympathetic nervous system (SNS) reactivity nor heartbeat detection (HBD) accuracy significantly covaried with social judgments. However, the effect of SNS reactivity on social judgments was significantly moderated by participants’ degree of HBD accuracy; b=. 19, SE=.08, 95% CIs [0.02, 0.35], t(157) = 2.24, p=.03, sr²= .03.

Table 3.

Regression results using HBD accuracy with social judgments as the criterion

Predictor	b	b 95% CI [LL, UL]	sr2	sr2 95% CI [LL, UL]	Fit	Difference
(Intercept)	3.64**	[3.38, 3.89]
Age	−0.06	[−0.21, 0.09]	.00	[−.02, .02]
Gender	0.16	[−0.23, 0.56]	.00	[−.02, .02]
BMI	0.05	[−0.01, 0.11]	.01	[−.02, .05]
					R2 = .020
					95% CI[.00,.07]
(Intercept)	3.63**	[3.37, 3.88]
Age	−0.05	[−0.20, 0.10]	.00	[−.01, .02]
Gender	0.18	[−0.22, 0.59]	.00	[−.02, .03]
BMI	0.05	[−0.02, 0.11]	.01	[−.02, .04]
HBD Accuracy	−0.07	[−1.72, 1.57]	.00	[−.00, .00]
SNS Reactivity	−0.01	[−0.02, 0.01]	.01	[−.02, .03]
					R2 = .026	ΔR2 = .005
					95% CI[.00,.06]	95% CI[−.02, .03]
(Intercept2	3.63**	[3.38, 3.88]
Age	−0.05	[−0.20, 0.10]	.00	[−.01, .02]
Gender	0.15	[−0.25, 0.55]	.00	[−.01, .02]
BMI	0.04	[−0.03, 0.10]	.01	[−.02, .03]
HBD Accuracy	−0.41	[−2.06, 1.24]	.00	[−.01, .01]
SNS Reactivity	−0.00	[−0.02, 0.01]	.00	[−.01, .02]
HBD Accuracy * SNS Reactivity	0.19*	[0.02, 0.35]	.03	[−.02, .08]
					R2 = .056	ΔR2 = .031*
					95% CI[.00,.10]	95% CI[−.02, .08]

Note. Higher criterion values indicate more positive social judgments. All continuous predictors are mean-centered. A significant b-weight indicates the beta-weight and semi-partial correlation are also significant. b represents unstandardized regression weights. beta indicates the standardized regression weights. sr² represents the semi-partial correlation squared. r represents the zero-order correlation. LL and UL indicate the lower and upper limits of a confidence interval, respectively.

^*Indicates p < .05.

^**indicates p < .01.

As predicted, simple slopes analyses revealed that greater SNS reactivity impacted negative evaluations of the interviewers—but only for individuals who were less accurate at detecting their own cardiovascular signals. Specifically, simple slopes analyses revealed that the relationship between SNS reactivity and social judgments was only significant at levels of HBD accuracy one standard deviation below the mean: $\widehat{b}=-0.03$, se = 0.01, t(154) = 2.25, p = 0.03, whereas this relationship was not significant at either mean-level or at one standard deviation above the mean for HBD accuracy. Participants with lesser interoceptive accuracy and greater SNS reactivity rated their interviewers more negatively on average (see Figure 2).

Figure 2. Interaction of interoceptive accuracy and physiological reactivity on social judgments.

Results from final regression models reveal that the effect of SNS reactivity on social judgments is significantly moderated by participants’ degree of HBD accuracy. Simple slopes revealed a significant negative slope of SNS reactivity on social judgment for individuals 1 SD below the mean of interoceptive accuracy (blue, solid), such that individuals with lower interoceptive accuracy were more likely to judge the interviewers more negatively when they were more physiologically reactive. Slopes for mean (green, dashed) and one standard deviation above the mean (yellow, dotted) were non-significant.

Specificity Tests - Task Judgments.

Neither age, nor BMI significantly covaried with social judgments (see supplemental table 2). However, there was a significant effect of gender on task judgments, with Men rating the task less negatively than Women; b = −0.59, SE = 0.16, 95% CI[−0.991, −0.27], t(154) = −3.61, p < 0.001, sr² = 0.08. Neither HBD accuracy, SNS reactivity, nor their interaction significantly covaried with situational judgments in any model tested (all ps > 0.05).

Specificity Tests – Negative High Arousal Affect.

Neither age, nor BMI significantly covaried with reported negative high arousal affect (see supplemental table 3). However, there was a significant effect of gender on negative high arousal affect, with Men endorsing less intense negative high arousal emotions than Women; b = −0.60, SE = 0.18, 95% CI[−0.97, −0.23], t(154) = −3.25, p < 0.01, sr² = 0.06. Neither HBD accuracy nor SNS reactivity significantly covaried with negative high arousal affect when entered at step two. However, when the interaction effect was entered at step 3, the effect of HBD accuracy on negative high arousal affect became significant. Consistent with similar models reported in (MacCormack, Bonar, & Lindquist, under review), individuals with greater interoceptive accuracy endorsed less intense negative high arousal emotions following the TSST; b = −1.67, SE = 0.76, 95% CI[−3.18, −0.16], t(154) = −2.19, p = 0.03, sr² = 0.03. Neither SNS reactivity nor the interaction between HBD accuracy and SNS reactivity significantly covaried with negative high arousal affect in any model tested (all ps > 0.05; see Figure 3).

Figure 3. Size of interaction effect between interoceptive accuracy and physiological reactivity on social judgments, task judgments and subjective affect.

As predicted, interoceptive accuracy interacted with physiological reactivity to impact affective realism, operationalize as judgments of the interviewers. These factors did not impact perceptions of the interview task itself or participants’ subjective affective responses, however. Standardized regression coefficients with 95% CI for interaction terms (HBD Accuracy x SNS Reactivity) from final regression models (controlling for Age, BMI, and Gender) using social judgments (Social; yellow), task judgments (Task; blue), and negative high arousal affect (Affect; gray) as criterion following the Trier Social Stress Test.

Discussion

This study explored the influence of interoceptive ability and sympathetic nervous system reactivity on experiences of affective realism following an ecologically valid social stressor. Consistent with our hypotheses, we found that affective realism occurred when people used their own internal bodily sensations to make inferences about others’ intentions and dispositions following the Trier Social Stress Test. Specifically, we found evidence that individuals who were less adept at identifying the timing of their own heartbeats rated interviewers more negatively when experiencing greater sympathetic nervous system reactivity. While we cannot evaluate precise mechanisms with these data, we speculate that individuals low in interoceptive ability may be more likely to misattribute their own internal feelings (e.g., of physiological arousal) when evaluating the meaning of socially ambiguous behaviors. This inference would be consistent with prior research suggesting that individuals with lesser interoceptive ability are more likely to use their own affective states when inferring the emotions of others (von Mohr et al., 2021) and to have difficulty differentiating their own body from the bodies of others (Babo-Rebelo et al., 2019; Tajadura-Jiménez & Tsakiris, 2014; Tsakiris et al., 2011). In this sense, awareness of interoceptive signals may be “self-stabilizing,” helping to anchor sensory experiences—including those from within the body—in first-person, self-focused subjective experience as opposed to experience focused on people and things in the external world (i.e., world-focused experience; Lambie & Marcel, 2002; Lee et al. 2018; Lindquist & Barrett, 2008). Future studies should more specifically evaluate interoceptive ability as a mechanism of affective realism. If interoceptive ability is a pathway by which affect becomes a source of realism, then intervening on interoceptive ability may help individuals to uncouple negative affective reactions from harmful perceptions and behaviors (Allen & Tsakiris, 2018; Azzalini et al., 2019; Bebo-Rebelo & Tallon-Baudry, 2018).

Notably, we observed affective realism specifically in the context of social ratings. Neither interoceptive ability nor its interaction with sympathetic nervous system reactivity predicted differences in participants’ perceptions about the interview task itself. It is possible that affective realism is most likely to occur in social contexts because social behavior is both affectively relevant (see Atzil et al., 2018; Fotopoulou & Tsakiris, 2017) and inherently ambiguous (i.e., social judgments require inferences about others’ unobservable internal experiences and intentions that draw on both the external context and one’s own affective reactions; Waytz et al., 2010). The interviewers in our study were trained to be neutral in their behavior, meaning they were neither explicitly supportive nor critical of participants. These design principles are implemented during the Trier Social Stress Test to maximize social evaluative threat while minimizing the kinds of affiliative social cues humans rely on to navigate stressful social situations (Dickerson & Kemeny, 2004).

We did find a significant main effect of interoceptive ability on negative high arousal emotion ratings, insofar as those individuals who have better awareness of their internal state endorsed less intense negative high arousal emotions, regardless of their level of sympathetic nervous system reactivity. This finding is consistent with MacCormack et al. (under review), who ran models assessing interoceptive ability and interoceptive beliefs (i.e., that body states are harmful v. helpful) on high arousal affect. Although speculative, these findings may again point to the “self-stabilizing” effect of interoceptive ability. Interoceptive ability may buffer against excessive subjective self-focused distress because individuals can differentiate bodily responses that occur when engaging in a motivated performance task (e.g., a small to moderate increase in heartrate during speech) from an emotional response indicating threat. This ability may in and of itself serve to diminish affective realism because participants are experiencing their physiological state as just that—physiological activation and not as a sign that the world around them is distressing.

Like all else – affective realism can be both adaptive and maladaptive. Future research should explore the boundary conditions determining when or how affective realism may be adaptive across different contexts. It is worth noting that most participants in our sample performed at or slightly above chance accuracy in the heartbeat detection task suggesting that most people are not particularly attuned to their interoceptive sensations. However, it is the participants who performed at below-chance levels who showed heightened affective realism effects. In the present study we evaluated how individuals used affect to make sense of social targets under conditions of social evaluative threat. If relatively greater interoceptive ability can reduce social affective realism and self-focused threat when feelings run high (i.e., during experiences of strong physiological reactivity), then cultivating interoceptive ability might help mitigate these outcomes, specifically in situations where they are most likely to cause harm. For instance, there is evidence that active-duty police officers experience affective realism when using unnecessary lethal force in situations where they are experiencing excessive levels of cardiovascular arousal (Andersen et al., 2018). There have been attempts to increase interoceptive ability in adults using biofeedback or meditation-based techniques albeit with mixed results. For example, some studies have claimed that various meditation-based techniques may increase cardiovascular interoception whereas others have produced null effects (Nielsen & Kaszniak, 2006; Khalsa et al 2008; Khalsa et al 2020). Biofeedback techniques have shown greater promise. Early studies suggested null effects (Meyerholz et al 2019) though more recent studies pairing biofeedback with physical activity manipulations have elicited significant improvements in interoceptive ability (Quadt et al 2021). Interestingly, one recent study suggested a link between resonance breathing (respiration paced at 6 times per minute, compared to natural breathing at 12-20 times per minute) and interoceptive ability (Leganes-Fonteneau et al., 2021). We find this example particularly compelling given that, preliminary evidence has suggested that heart-rate variability biofeedback training via resonance breathing (Lehrer & Gevirtz, 2014) during scenario-based threat exposures can generate long-term reductions in lethal force decision errors amongst active-duty police officers (Andersen et al., 2018). It is thus possible that these methods could be used to both decrease cardiovascular arousal and induce transient increases in interoceptive ability.

On a longer scale, socialization during childhood may also influence interoceptive ability. For example, one study found that children whose mothers used more body-focused language exhibited greater socio-affective skills (e.g., emotion regulation, social initiation, cooperation, self-control) (MacCormack et al., 2020). Individuals high in alexithymia—an inability to describe and understand one’s own emotional experiences—also have poorer interoceptive ability across multiple sensory domains (Brewer et al. 2016; Murphy et al. 2018) and alexithymia appears to be transmitted, at least in part, via parental socialization (Le et al. 2002). Early interventions during childhood may thus help to increase interoceptive ability.

Limitations and Constraints on Generality

While our findings help to advance research on affective realism, they are not without limitations. Firstly, as in most laboratory studies, our sample was not representative of all adults: it was limited to healthy young adults and over-represented some ethnic and racial identities and socio-economic statuses. Future research should thus replicate these findings in larger, more representative samples. Several social identity variables (e.g., age, gender, and socioeconomic status) as well as myriad clinical conditions (e.g., anxiety, depression, functional motor disorders) have been linked to interoceptive ability and affective experience, making more inclusive sampling in future research both an ethical and theoretical imperative (Grabauskaitė et al., 2017; Khalsa et al., 2018; Moeini-Jazani et al., 2017; Murphy et al., 2019; Paulus et al., 2019; Proffitt Leyva et al., 2020; Proffitt Leyva & Hill, 2018). Secondly, while the heartbeat detection task has been well-validated, it is a trait-level proxy for how individuals represent and utilize interoceptive sensations from a single interoceptive modality (i.e., heartbeats). Furthermore, research has suggested that if anything, the heartbeat detection task may under-estimate interoceptive sensitivity among those who naturally perceive their heartbeat at different times relative to ventricular depolarization (Ring & Brener, 2018). Despite this, the heartbeat detection task has strong predictive validity and correlates with cortical activity in regions known to integrate interoceptive information into conscious experience (e.g., right anterior insular/opercular cortex, Critchley et al., 2004). Thirdly, in the present analyses we focused on experiences of negative high arousal affect. Future work should assess the specificity or generalizability of these effects for experiences distributed across the affective circumplex. Finally, our study did not determine the ultimate mechanism by which affective realism occurs. Ultimately, we employed a correlational design which introduces the possibility of third variables. It would be interesting in future research to examine how SNS reactivity, interoception, and social affective realism are mediated by socialization factors (Fotopolou & Tsakiris 2017, Atzil et al 2018, MacCormack et al 2020) or by interactions amongst brain networks associated with affect, vision, and mentalizing (Provenzano et al., 2019).

Conclusion

Experiences of affective realism are both mundane and consequential, particularly when they bias social judgments that lead to real-world consequences. These data suggest that interoceptive ability may be an important individual difference that mitigates when and to what extent experiences of negative high arousal affect might bias social judgments.

Open Practices Statement.

The analyses reported in this article were not formally pre-registered. Materials have been made available in the SOM. Requests for the data can be sent via email to the lead author. Full code for analyses, diagnostics, and plots are available at https://osf.io/8ys5a/.

Author Bios

Mallory J. Feldman is a doctoral student in Psychology and Neuroscience at the University of North Carolina at Chapel Hill. Jennifer K. MacCormack is an Assistant Professor in Psychology at the University of Virginia. Adrienne S. Bonar is a doctoral student in Psychology and Neuroscience at the University of North Carolina at Chapel Hill. Kristen A. Lindquist is a Professor of Psychology and Neuroscience at the University of North Carolina at Chapel Hill.