Socioeconomic Status Shapes Dyadic Interactions: Examining Behavioral and Physiologic Responses

Abstract

With more opportunities for diverse interactions, little is known about how social interactions involving people of different socioeconomic status (SES) may unfold. We investigated social-attunement patterns in dyadic interactions involving SES. Unacquainted adults recruited from a community in the United States interacted with similar-or-different-SES partners in the lab (N = 130 dyads). Attunement was assessed throughout the interaction by examining physiological linkage—how much a person’s physiological change is predicted by another’s physiological change over time. Overall, low-SES participants showed stronger physiological linkage—indicating greater attunement—to partners across SES. Participants also appeared more comfortable when interacting with low-SES partners. There were no SES differences in dominance during the conversation. After the interaction, participants reported liking similar-SES partners more than different-SES partners. These patterns suggest that during interactions, lower-SES individuals are more other-focused than high-SES individuals, and in-group preference prevails. We note limitations in the racial representation of our sample.

Introduction

Greater efforts at promoting socioeconomic diversity across communities (Cortright, 2018), workplaces (Dittmann, 2021; Ingram, 2021), and schools (RTI International, 2019) have increased opportunities for cross-class interactions. Although such opportunities are important for reducing the socioeconomic divide in societies, there is limited understanding of how actual cross-class interactions may unfold.

The present research examined potential interpersonal and intergroup processes—using physiological, behavioral, and self-report measures—during first-time social interactions involving dyads similar and different in socioeconomic status (SES).

Intergroup interactions are frequently characterized by affiliative goals guided by homophily—the tendency to connect with those of similar group membership (Alves et al., 2016; Bond et al., 1968). Homophily is often studied in racial encounters (e.g., Mendes et al., 2002; Shelton et al., 2005; West et al., 2017), but much less so in SES encounters. In one early demonstration, participants reported stronger liking toward a bogus interaction partner manipulated to be similar, rather than dissimilar, to their reported economic status (Byrne et al., 1966). Similarly, a recent study involving actual interactions of randomly paired dyads found the strongest social connection in participants similarly high or similarly low in SES, based on self-reports and nonverbal behaviors during the interaction (Côté et al., 2017).

Current theory and evidence suggest individual SES differences in social orientation. Lower-SES individuals, who contend with frequent social threats (e.g., Kraus et al., 2011), are predisposed to monitor their environments and adjust to the responses of others (Stephens et al., 2007; Townsend et al., 2014). Illustrating this, lower-SES individuals paid more visual attention to street pedestrians (Dietze & Knowles, 2016) and were more accurate at inferring others’ emotions (Dietze & Knowles, 2021; Kraus et al., 2010) than their higher SES counterparts. Conversely, well-resourced high-SES individuals encounter fewer social threats, allowing them to express themselves uniquely and act independently of others (Kraus & Stephens, 2012). Despite these findings, no studies to our knowledge have examined whether these tendencies may shape distinct interpersonal processes during social interactions.

We propose that SES differences in social orientation may produce interpersonal effects on social attunement—being attentive and responsive to others—during a novel social interaction. In unfamiliar interactions, low-SES individuals may be inclined to monitor for social threats by being other-focused and increasing their attunement to how interaction partners respond. Conversely, high-SES individuals who are less concerned about social threats may be able to afford greater self-focus and be less attuned to their interaction partners’ responses. Furthermore, distinct behavioral patterns may emerge from social-attunement differences because of individuals’ own SES and the SES of their interaction partners. Specifically, less socially attuned high-SES individuals may show greater conversational dominance—tending to take the lead in conversations (Dovidio et al., 1988; Fragale et al., 2011)—than low-SES individuals. Relatedly, low-SES individuals who are socially attuned and less likely to dominate may elicit greater interpersonal comfort (Gregory & Webster, 1996; Holtgraves & Yang, 1992) from those interacting with them.

We also anticipated intergroup effects on social attunement that reflect homophily. Following past research observing stronger signs of affiliation with those similar in SES during social interactions (Côté et al., 2017), we expected the presence of affiliative goals to shape greater attunement to similar-SES interaction partners than to different-SES interaction partners.

The present research tested the interpersonal and intergroup effects on attunement in a social-interaction lab study of dyads paired systematically by SES. Participants were first classified as high or low SES on the basis of a multidimensional SES index and then paired into two types of participant-partner dyads: similar-SES dyads, comprising “low-low” and “high-high” SES dyads, and different-SES dyads, comprising “low-high” and “high-low” SES dyads (different-SES dyads are one type of dyad). In this lab study, participants could reveal their SES through the conversation questions we prompted them to ask each other. We examined the proposed processes by assessing a complementary suite of physiologic response, behaviors, and self-reports during the interaction. Assessing physiologic responses continuously and unobtrusively during social interactions allow people to interact naturally without interruption. Physiologic responses, along with behaviors assessed by trained coders, provide information about psychological processes that people might not be conscious of or willing to report.

We examined social attunement by assessing physiologic linkage—a dyadic process in which a person’s physiological response is influenced by the physiological response of another using a time-lagged design (West & Mendes, 2023). Physiological response here is assessed by preejection period (PEP)—a time-based cardiovascular response measure of the duration from left ventricle contraction to aortic valve opening, with shorter periods representing greater ventricle contractile force. Unlike heart rate, which is influenced by both sympathetic and parasympathetic nervous system branches, PEP reflects pure sympathetic nervous system (SNS) activation. Engaging in active tasks, such as playing a cooperative game or conversing with someone, increases SNS activation, resulting in PEP decreases (shorter duration) from a resting state, or PEP reactivity (Mendes, 2016). Greater PEP reactivity indicates stronger affective response to the task. During an interaction, physiologic linkage is elicited when a person attends to, detects, and then experiences a partner’s affective response to interaction tasks (Thorson et al., 2018). Physiologic linkage to a partner is estimated by how much a person’s current PEP reactivity is predicted by their partner’s PEP reactivity moments earlier (West et al., 2017), with stronger linkage reflecting greater attunement to the partner. Linkage is estimated as a path in a multilevel model, and here, we tested whether the strength of that path is moderated by the SES of the participant and the partner.

Following our theorized interpersonal effect on attunement, we predicted that participant SES would moderate the linkage path. Specifically, low-SES participants would show stronger positive physiological linkage to their partners—indicating greater attunement to partners—than high-SES participants. Analytically, the linkage path (i.e., the effect of the partner’s prior PEP-reactivity score on the participant’s current PEP-reactivity score) would be moderated by participant SES, so that the linkage path would be stronger for low-SES participants than for high-SES participants. In contrast, if goals to affiliate with similar others (i.e., homophily) are present, we expected a different pattern of effects: The linkage path would be moderated by a Participant SES × Partner SES interaction. This interaction would indicate that participants in similar-SES dyads show stronger linkage to their partners than those in different-SES dyads, suggesting greater attunement within similar-SES dyads.

We also examined behavioral patterns involving conversational dominance and interpersonal comfort. We operationalized conversational dominance as behaviors that characterize leading a conversation—namely interrupting, asking for follow-ups, and topic control. These behaviors are well-established in contexts involving status differences: Interruptions are more frequently made by higher-status people (e.g., men in contexts where gender is status-based; Hall, 1984), and interrupters are perceived as higher in status than noninterrupters (Farley, 2008; Robinson & Reis, 1989). Asking follow-up questions is more common among high-SES patients than low-SES patients in medical encounters (Street et al., 2005) and more common among high-SES students than low-SES students in classrooms (Calarco, 2011). Taking “topic control” during an interaction (e.g., answering for a partner) is also associated with being higher in status (P. Brown & Levinson, 1978). Given these past effects, we predicted that high-SES participants would exhibit greater conversational dominance across these behaviors than low-SES participants.

We operationalized interpersonal comfort as how much each person appeared nervous relative to relaxed while interacting, indicated by two well-established behavioral markers—speech clarity and fidgeting. Low speech clarity and high fidgeting indicate greater nervousness, whereas high speech clarity and low fidgeting indicate greater relaxation. Clear and fluent speech is associated with social comfort (Baker & Edelmann, 2002), and fidgeting (e.g., moving about restlessly, playing with objects nonessential to the task) is associated with state anxiety (Creamer et al., 1995) and distress (Fairbanks et al., 1982). Fidgeting also occurs more frequently in socially threatening settings, such as cross-race interactions during which European Americans are motivated to not appear prejudiced (West et al., 2017). If social attunement elicits comfort in others, we predicted a partner SES effect: We anticipated that participants would show greater interpersonal comfort when paired with low-SES partners than when paired with high-SES partners. A Participant SES × Partner SES interaction should be found—similar to the linkage hypotheses—if participants have strong affiliative goals with similar-SES partners, indicating that participants in similar-SES dyads express greater interpersonal comfort than those in different-SES dyads.

Finally, we examined how the proposed differences in attunement impact self-reports of partner liking (Templeton et al., 2022). Liking someone may reflect one’s interpersonal comfort with the person (Reis et al., 2011). Thus, we expected effects on partner liking that are parallel to effects on interpersonal comfort: Participants would report liking low-SES partners more than high-SES partners. In contrast, if homophily exerts a stronger influence, we expected a Participant SES × Partner SES interaction effect, with participants in similar-SES dyads reporting liking their partners more than those in different-SES dyads. We summarize all predictions in Table 1.

Table 1.

Summary of Processes and Hypothesized Patterns With Dependent Measures

Measure	Participant SES	Partner SES	Participant SES × Partner SES
Physiological linkage	Low-SES participants will show stronger linkage to their partners than high-SES participants, indicating that low-SES individuals are more attuned to their partners.		Participants in similar-SES dyads will show stronger linkage to their partners than those in different- SES dyads, indicating that individuals are more attuned to partners similar in SES.
Behavior	High-SES participants will appear more dominant in conversation than low-SES participants.	Participants will appear more comfortable when paired with low-SES partners than when paired with high-SES partners.	Participants in similar-SES dyads will appear more comfortable than those in different-SES dyads.
Partner liking		Participants will like low-SES partners more than high-SES partners.	Participants in similar-SES dyads will like their partners more than those in different- SES dyads.

Research Transparency Statement

General disclosures

Conflicts of interest: All authors declare no conflicts of interest. Funding: This research was supported by a National Institute of Aging grant (No. R24AG048024) to W. B. Mendes and a National Science Foundation grant (No. BCS 1430799) to W. B. Mendes and T. V. West. Artificial intelligence: No artificial-intelligence-assisted technologies were used in this research or the creation of this article. Ethics: This research received approval from the Ethics Review Board at the University of California, San Francisco.

Study disclosures

Preregistration: The primary hypothesis, method, and analysis plan for physiologic linkage was preregistered (https://aspredicted.org/D6S_X4C) on January 8, 2018. The hypotheses and analysis plan for behaviors and liking were not preregistered. Data collection began on July 19, 2017, and no data were analyzed until data collection ended on August 14, 2018. Materials: All study materials are publicly available (https://osf.io/tzv2b/). Data: All primary data are publicly available (https://osf.io/tzv2b/). Analysis scripts: All analysis scripts are publicly available (https://osf.io/tzv2b/). Computational reproducibility: The computational reproducibility of the results has not been independently confirmed by the journal’s STAR team, because this research was reviewed under the previous Editorial Board.

Method

Study overview

We conducted a social-interaction lab study that involved stranger dyads paired by their reported SES. Prior to the lab study, participants completed a prestudy survey that assessed both objective and subjective aspects of SES. A multidimensional SES index score for each participant was derived using a principal components analysis of responses and used to systematically pair them into dyads. Participants who were both lower or both higher in SES relative to the sample and whose SES scores were no more than 0.5 standard deviations apart were paired as similar-SES dyads. Participants who were lower or higher in SES and paired with someone higher or lower in SES, respectively, and with an SES score at least 1.5 standard deviations apart, were paired as different-SES dyads. All dyads were matched on age, sex, and race.

After the initial introduction, in which the dyad met each other and exchanged background information about themselves and completed a structured conversation, the dyad engaged in a cooperative game and finally in a modified-speech task from the Trier Social Stress Test (TSST; Kirschbaum et al., 1993). After each task, participants answered questions about the interaction. Throughout the interaction, participants’ cardiovascular responses were measured noninvasively to derive changes in SNS activation (i.e., PEP). Because the tasks were designed to progressively increase affective and cognitive demands over time, we expected PEP reactivity to increase as tasks progressed, reflecting the dynamic changes throughout the interaction. The modified-speech task was positioned at the end of the study because this task evokes the greatest affect intensity. Moreover, the speech task is a coexperience for the dyad during which they could use one another as sources of information (i.e., be other-focused) or completely ignore their partner and focus on themselves instead (i.e., be self-focused). The self-versus-other nature of the task was critical to testing our hypotheses related to attunement.

All interactions were videotaped to assess nonverbal behaviors during the interaction. Figure 1 presents an overview of the study procedure and tasks. These structured tasks follow the general approach taken over the years in lab social-interaction studies that mimic new-acquaintance interactions in a fast-tracked manner (e.g., Blascovich et al., 2001; C. L. Brown et al., 2021; Mendes et al., 2007, 2008; West et al., 2017). Specifically, dyads first engaged in a self-introduction and a structured interaction to get to know each other, followed by cooperative tasks that acquainted them more. Then they completed cooperative tasks that were progressively more demanding. These tasks are designed to be challenging, which is necessary for assessing physiologic reactivity and patterns of physiological linkage (Thorson et al., 2018). The tasks were also ordered to increase the intensity of psychological and affective demands, capturing how a new social relationship may progress as people become more familiar with each other.

Fig. 1.

Overview of the study procedure. Dashed lines indicate the procedure that occurred before the lab session. Bold outlines indicate tasks that assessed liking and behaviors. Cardiovascular responses were recorded throughout the lab-session tasks.

Participants

We recruited 264 participants (48.5% female) from the San Francisco Bay area, using online ads (posted on Craigslist.com) and flyers. The mean age of the sample was 29.47 years (SD = 7.7, range = 18–53). The racial composition was 42.9% European American, 21.5% Asian American, 20.6% Hispanic American, 9.2% African American, and 4.8% mixed race. Participants were paired into 132 similar-SES or different-SES dyads, matched on age, gender, and race. Two dyads quit the study midway, which resulted in a final sample of 130 dyads (N_similar = 48 dyads; N_different = 82 dyads). Missing data were caused by physiologic equipment malfunction, experimenter error, or a participant not responding to a survey item. Participants were paid $50. The study was reviewed and approved by the Ethics Review Board at the University of California, San Francisco.

Power analysis and sampling strategy

Our target sample size was a minimum of 120 dyads (240 participants). The target minimum sample size of 120 dyads was determined on the basis of a power analysis using a simulation method conducted in prior research for detecting physiological linkage effects within dyadic interactions (Bolger et al., 2012; Lane & Hennes, 2018; Thorson et al., 2018). We focused on power to detect effects for linkage (rather than behavior or self-report), given that linkage effects were the primary and novel outcomes of this research. This method simulated 1,000 hypothetical study data with 50 to 70 dyads each using a typical range of linkage estimates. Each simulated study data was then analyzed individually with the number of times the key linkage effect was statistically significant at p < .05. The analyses revealed a statistical power of between 50% to 94% to detect linkage effects for sample sizes between 50 and 70 dyads. Therefore, our target sample size of 120 dyads with an SES pairing manipulation should be sufficiently powered to detect a small effect of η_p² = .010 (Cohen et al., 2003).

As is our strategy in psychophysiology lab studies, we oversampled by 10% in anticipation of missing data, resulting in the recruitment of 132 dyads (264 participants). Relying on our hypotheses about participant and partner SES (effects independent of their interaction), we aimed to have a roughly equal number of high- and low-SES participants in the study (resulting in 131 high-SES participants and 129 low-SES participants, after two dyads dropped out). Given our interest in understanding cross-group dyads, we created more different-SES dyads than similar-SES dyads (see West et al., 2017). Moreover, given the heterogeneity in variables such as age, gender, and race, a larger sample was required.

Procedure

Prelab session: manipulating SES dyads.

To pair participants by SES, in a prestudy survey, we asked participants to report on a range of objective and subjective SES measures, namely annual household income (1 = less than $15,000, 10 = more than $300,000; M = 5.18, SD = 2.18, range = 1–10), annual personal income (1 = less than $15,000, 10 = more than $300,000; M = 3.47, SD = 2.03, range = 1–10), MacArthur Scale of Subjective Status (SSS) in the United States (Adler et al., 2000; 1 = lowest, 10 = highest; M = 5.08, SD = 1.80, range = 1–10), MacArthur SSS in the community (1 = lowest, 10 = highest; M = 5.40, SD = 1.89, range = 1–10), own education (1 = less than high school diploma, 10 = MD/PhD/JD or other doctorate level degree; M = 5.89, SD = 2.13, range = 2–10), father’s education (1 = less than high school diploma, 10 = MD/PhD/JD or other doctorate level degree; M = 4.88, SD = 2.97, range = 1–10), mother’s education (1 = less than high school diploma, 10 = MD/PhD/JD or other doctorate level degree; M = 4.45, SD = 2.61, range = 1–10), and class identification (Jackman, 1979; 1 = poor, 2 = working class, 3 = middle-class, 4 = upper-middle class, 5 = upper-class; M = 2.55, SD = 0.70, range = 1–5). We derived a multidimensional SES index by applying a principal components analysis to the SES measures, following a social-epidemiological approach to provide a holistic index of a person’s SES (Osborn & Morris, 1979; Savage et al., 2013; Vyas & Kumaranayake, 2006). All measures loaded onto one factor. To calculate our composite score, we used the weights from the PCA to calculate participants’ SES index score within the sample (α = .78, M = 7.50, SD = 2.21, range = 2.13–12.86). We then converted SES scores into standard deviation scores for pairing. Detailed results from the PCA, including the factor loadings and weights, are provided in the Supplemental Material available online.

We chose to pair participants using a multidimensional SES index instead of a single metric of household income or subjective social status for two reasons. First, there is emerging recognition in the psychological study of social class that people’s social-class identity includes multiple relevant dimensions beyond objective indicators of current income and education, such as their past economic circumstance and standing (Destin et al., 2017), as well as current and past subjective standing within their most immediate and relevant socioeconomic context (Tan et al., 2020). The single-metric approach does not account for these other relevant aspects of social-class identity that can also shape social perception and behavior. Second, because the current sample is drawn from the San Francisco Bay Area, a relatively expensive area in the United States, objective income alone may not reflect the full range of factors that shape SES perceptions in this context (Tan & Tai, 2025). A multidimensional index of SES considers both objective and subjective factors as well as past and present economic circumstances that capture important variation in SES within the sample and context.

Participants were randomly assigned to a similar-SES or different-SES dyad pairing before the lab study session. We matched dyad members on age, gender/sex, and race. For age, we paired participants below 25 with partners no more than 3 years apart in age, and we paired participants 25 and above with partners no more than 5 years apart in age (see West et al., 2017, for a similar strategy). The racial and ethnic composition in the San Francisco Bay Area includes majority White and Asian American groups, whereas the minority groups are the Black and Latino populations. Thus, we created dyads that were based on the majority compared to minority composition. The final sample included 68 male and 64 female dyads and 82 majority-race and 48 minority-race dyads. A breakdown of dyads by SES and race is provided in the Supplemental Material.

We applied the following rules in pairing dyads by SES: For similar-SES dyads, participants’ SES scores were no more than 0.5 standard deviations apart. For instance, a pair with standard deviation scores of −1.00 and −1.40 and a pair with standard-deviation scores of 1.00 and 1.40 were classified as similar-SES dyads. Dyads with negative standard-deviation scores (e.g., −1.00 and −1.40) were classified as low-low-SES dyads; dyads with positive standard-deviation scores (e.g., 1.00 and 1.40) were classified as high-high-SES dyads. We did not pair similar-SES dyads in the middle range that had a mix of positive and negative scores (e.g., −0.30 and 0.10). This pairing strategy resulted in 47 similar-SES dyads (M_{SES-difference} = 0.21, SD = 0.14, range = 0.001– 0.49), with 24 low-low pairs (M_{SES-difference} = −0.52, SD = 0.36) and 23 high-high pairs (M_{SES-difference} = 0.52, SD = 0.41). The M_{SES-difference} values reported indicate that the SES scores of all participants in similar-SES dyads differed by less than 0.5 SDs from their partners. For different-SES dyads, participants’ SES scores were at least 1.5 SD apart. For instance, a pair with standard-deviation scores of −2.50 and −1.00 and a pair with standard-deviation scores of 1.00 and 2.50 were classified as different-SES dyads. This resulted in 83 different-SES dyads (M_difference = 2.26, SD = 0.62, range = 1.51–4.37). Here, the M_{SES-difference} values indicate that the SES scores of all participants in different-SES dyads differed by more than 1.5 SDs from their partners’.

Our decision to pair similar-SES dyads and different-SES dyads at the 0.5-SD and 1.5-SD cutoffs was determined by several considerations: We wanted to minimize or maximize differences for similar-SES or different-SES dyads, respectively, on the basis of the observed SES index range in our sample, and we wanted to do what was feasible for us to match sufficient dyads, given our sample-size target. Regarding maximizing differences, a threshold of 1.5 SD units is consistent with a very large effect size (Cohen, 1988), and that difference translates, on average, to participants drawn from nonoverlapping distributions of more than 71% (Sullivan & Feinn, 2012).

Lab session.

Dyad members were scheduled to arrive at the same time but to different locations, so they did not meet until after the baseline period. Once they were in separate experimental rooms, an experimenter explained the physiological sensors and obtained informed consent. Next, a female experimenter attached the sensors to participants to track their cardiovascular responses. Electrocardiography and impedance cardiography signals were recorded at 1000 Hz and integrated using Biopac MP150 hardware (Biopac, Inc., Goleta, CA). A 5-min baseline was then recorded for participants while they were alone in the room. After this, participants completed measures of their current affective state. The experimenter then revealed to participants that they would be working with another participant for the rest of the study. Following this, the experimenter verbally requested participants’ consent for the remainder of the study and then took them to a larger room for the dyadic interaction. Participants were seated facing each other approximately 5 feet away. Before the interaction began, the experimenter asked participants whether they already knew one another to ensure they were unacquainted. None indicated that they knew each other.

Initial interaction.

Participants first engaged in a “getting-to-know-you task”—described to them as a self-introduction, but designed to subtly indicate class background. SES can be conveyed rapidly and accurately to others through behavioral cues (e.g., physical appearance, mannerisms, and linguistic choices), and leisure activities and preferences (Bourdieu, 1979; Kraus & Keltner, 2009; Kraus et al., 2017). Accordingly, we provided participants with a list of 12 questions to ask each other during the introductions—questions designed to give off SES cues. The first four questions asked about their hometowns, the last school they attended, their current occupation, and their parents’ occupations. The remaining questions asked about lifestyle choices and preferences (e.g., favorite restaurants, favorite clothing store, sports they played, and hobbies; see the complete list in the Supplemental Material). Participants took turns asking their partners the questions. One partner was randomly assigned to ask all 12 questions first, and then the other asked the same 12 questions.

Structured interaction.

Next, participants engaged in a 4-min structured interaction, during which they asked each other another set of questions. Each partner received different questions. Although both sets of questions contained a range of general conversation topics, some were designed to elicit information reflecting SES and likely reinforced SES signals during the interaction. Questions included, “What is your greatest regret?”, “What is the best gift you’ve ever received?”, and “What would you do with one million dollars?” (see the complete list in the Supplemental Material). The interaction ended once 4 min were over, regardless of the number of questions asked. After this, participants completed a questionnaire in which they rated their impressions of each other.

Cooperative task.

Participants then played a cooperative word-guessing game similar to Taboo (West et al., 2017) for 6 min. In this task, participants took turns trying to get their partner to guess words without using any “taboo” words listed on their prompt cards. Participants were told they would receive $0.25 for every word guessed correctly and lose $0.25 for each taboo word used. Each participant took three turns as the clue giver and three turns as the guesser. The task was designed to be highly engaging, with the monetary bonus providing additional motivation. Additionally, the game required participants to pay close attention to their partners to guess the answers successfully. At the end of the game, participants rated how they felt about their interaction during the game.

Speech task.

For the final task, participants were instructed to deliver an impromptu speech on an assigned topic in front of a male evaluator (a modified version of the speech portion of the TSST; Kirschbaum et al., 1993). Participants were told that they would be judged on the quality and clarity of both speeches and that their joint performance could earn their team an additional bonus. Participants were provided with two topics (“Is social media harmful for society?” and “Do video games cause bad behavior?”) and were given 3 min to prepare written arguments on both topics. Next, they were randomly assigned to deliver a speech on one of the two topics. Participants then proceeded to the speech-preparation phase, where they prepared their speeches for 1 min using the arguments prepared earlier. Next, an evaluator entered the room and sat between both participants. During the speech-delivery phase, participants took turns delivering their speeches for 3 min each. The evaluator remained silent and maintained a stoic expression throughout. After both speeches were delivered, participants were then taken back to their individual rooms to complete the final questionnaires. Finally, participants were paid and debriefed individually by the experimenter.

Dependent measures.

PEP reactivity.

We assessed sympathetic nervous system activation by assessing changes in PEP throughout the interaction across all tasks. The electrocardiogram and impedance cardiography waveform data were processed off-line by visually examining and editing them; the data were then ensemble-averaged using Mindware software (IMP, Version 2.6, Mindware Technologies, Gahanna, OH) to estimate PEP. PEP represents the time between ventricular contraction and the opening of the aortic valve of the heart and is considered a pure index of sympathetic nervous-system activity (Brownley et al., 2000). We scored PEP in 30-s bins, and PEP reactivity scores were obtained by subtracting the last 30 s of baseline from all 30-s bins recorded during the interaction. As described earlier, greater PEP reactivity (larger decreases in PEP from baseline) indicates greater sympathetic nervous system activation. Consistent with similar dyadic studies, we used PEP reactivity as the primary measure of physiological linkage (see C. L. Brown et al., 2021; Thorson et al., 2018; West et al., 2017; see West & Mendes, 2023, for a review).

Behaviors.

Four trained research assistants unaware of the participants’ SES coded videos of participants’ behaviors. Behaviors were coded for the first 2 min of each structured interaction, following the recommendation of Murphy and Hall (2021), who demonstrated that judgments made from thin slices (i.e., the first few minutes of a longer interaction) are accurate, representative, and just as valid (as predictors of outcomes) as judgments based on the full interaction.

To train coders, we followed the training procedures of Thorson and West (2024). One pair coded interpersonal comfort, and one pair coded conversational dominance. Each pair coded 10 training dyads together to establish a shared understanding of the coding scheme, followed by five independent dyads to test that shared understanding. Disparities in the scheme were discussed with the help of a separate member of the team (the lead coder) who oversaw the coding process. Coders were blind to hypotheses. Separate cameras recorded each participant with each dyad member facing the camera, and with the two dyad members occupying separate frames. Coders coded each dyad member separately, except for conversational dominance, which was coded by watching both videos side by side.

Interpersonal comfort was operationalized as the degree to which each person appeared nervous (struggling to get a word out, fidgeting, moving around) to relaxed (speaking clearly and comfortably, not moving around), scored for each member of the dyad on a scale ranging from −3 (nervous) to 3 (comfortable), with zero defined as “speaking clearly but not confidently, reserved, adjusting the body for comfort but not fidgeting” (ICC = .75, in the very good to excellent range; Cicchetti, 1994).

Conversational dominance was operationalized as the degree to which one person in the dyad took the lead during the conversation (interrupted the partner, asked for follow-ups, helped the partner answer questions), scored for each member of the dyad on a scale ranging from −3 (took no control over the conversation) to 3 (took control over the conversation), with zero defined as “shared control with the partner.” Scores for the two members summed to zero, making this a within-dyad variable (ICC = .83, in the excellent range).

Ratings of partner liking.

We measured participants’ judgments of their interaction partners twice during the study—following the structured interaction and following cooperative game. We asked participants to report their general liking toward their partner on six items: “How much do you and your partner have in common?”, “How much do you like your partner?”, “How much did you enjoy interacting with your partner?”, “How much would you like to be friends with your partner?”, “How much would you want to interact with your partner in the future?”, and “How comfortable do you feel with your partner?” These were obtained twice, after the structured interaction and the cooperative game similar to Taboo. All items were rated on a 7-point scale (1 = not at all, 4 = somewhat, 7 = a great deal) and averaged to form a composite of liking after both tasks (structured interaction, α = .90, M = 5.21, SD = 1.14; Taboo, α = .93, M = 5.22, SD = 1.18).

Manipulation checks.

To determine the success of our SES pairing manipulation, we asked participants after the entire interaction to rate their own SES, as well as their partner’s, on the MacArthur ladder scale (1 = very bottom, 10 = very top). We also assessed judgments of similarity by asking participants after the entire interaction to rate their agreement with two statements: “I perceive myself and my interaction partner to be part of the same group” and “Overall, I think my interaction partner and I are very similar” (1 = strongly disagree, 7 = strongly agree; r = .72, p < .001; M = 4.40, SD = 1.36).

Analytic Strategy

Analytic strategy for physiological linkage

To examine whether participants showed physiological linkage from one 30-s interval to the next to similar-SES or different-SES partners, we estimated a stability and influence model (Thorson et al., 2018) with PROC MIXED in SAS (West et al., 2017). Participants’ PEP reactivity at one point was treated as a function of their own reactivity at the prior time point (30 s prior, the stability path) and their partner’s reactivity at that prior time point (the linkage path). Degrees of freedom were estimated using the Satterthwaite method, which involves a weighted average of the between-subjects and within-subjects degrees of freedom (see Kenny et al., 2006). Degrees of freedom in this method, which can be fractional, are based on the total number of data points and adjusted for the nonindependence of observations. Because the nonindependence of observations is considered, the degrees of freedom in these analyses vary across different tests.

We tested whether the strength of linkage varied as a function of participants’ SES, their partner’s SES, and the Participant SES × Partner SES interaction (which compares similar-SES dyads to different-SES dyads). Following the recommendations of Ledermann et al. (2011), this model included the stability path (i.e., the effect of the participant’s time x PEP reactivity score on their time x + 1 PEP reactivity score) and all interactions between the SES variables and the stability path. Thus, linkage is estimated, accounting for stability.

The interactions of theoretical interest were the two-way Partner Prior PEP Reactivity Score × Participant SES interaction, the two-way Partner Prior PEP Reactivity Score × Partner SES interaction, and the three-way Partner Prior PEP Reactivity Score × Participant SES × Partner SES interaction. Together, these interactions test whether the strength of the linkage path varies as a function of one’s own SES (the Partner Prior PEP Reactivity Score × Participant SES), one’s partner’s SES (the Partner Prior PEP Reactivity Score × Partner SES), and the dyadic combination of both people’s SES (the Partner Prior PEP Reactivity Score × Participant SES × Partner SES). When estimated simultaneously, these analyses reveal which of the four types of participants (low-SES participants with high-SES partners, high-SES participants with low-SES partners, high-SES participants with high-SES partners, and low-SES participants with low-SES partners) show the strongest linkage to their partners during the study.

Analytic strategy for behaviors and self-reports

Our two time points of self-reported partner liking were analyzed using a repeated-measures dyadic analysis to adjust for nonindependence between the two measures within person (each participant’s Time 1 and Time 2 response), between the two dyad members’ within time point (i.e., Person 1’s and Person 2’s Time 1 data; Person 1 and Person 2’s Time 2 data), and between dyad members across time points (Person 1 Time 1 with Person 2 Time 2; Person 2 Time 1 with Person 1 Time 2). Given that the dyad members were indistinguishable, we set parameter constraints on the variance/covariance matrix to set the variances of the two people to be equal (and all corresponding covariances). For behaviors, which were measured once, we estimated a dyadic model adjusting for nonindependence in partner’s behaviors.

In each model (except for conversational dominance), we included the main effects of participant SES and partner SES, and the Participant SES × Partner SES interaction. For liking, we also included the main effect of time point (structured interaction at Time 1 vs. cooperative game at Time 2), and all interactions between the SES variables and time point. For conversational dominance, we examined the absolute value of the dyad-level score as a function of three-level dyadic composition variable (similar-SES high-high dyads, similar-SES low-low dyads, and different-SES dyads) to test whether the likelihood that one partner led the conversation more than the other partner differed by dyad type. We further examined whether within different-SES dyads, one partner was more likely to dominate the conversation than in similar-SES dyads, and if so, which partner (the high- or low-SES one).

We report effect sizes for each parameter as rs for multilevel models involving the t statistic and partial R²s for multilevel models involving the F statistic (Edwards et al., 2008). Supplemental analyses for each dependent measure are also reported in the Supplemental Material.

Results

Supplemental data

Before presenting findings from the main study, we summarize results from a separate sample testing the possible demand characteristics of the initial interaction—specifically, whether the questions participants asked each other revealed the intent of the study to examine SES. An online sample (N = 407) read scenarios that described the main study from the participant’s perspective (the full study details are presented in the Supplemental Material). We asked them to imagine that they were interacting with a partner of the same age, same sex/gender, and similar race/ethnicity as themselves and were able to ask them these 12 questions. We then showed them the same list of 12 questions from the initial conversation task and asked them how likely they would be to glean specific group-category information about their partner from meeting and asking these questions. Specifically, we presented eight social categories—personality, religion, relationship status, ideology, sexual orientation, socioeconomic status, cultural/geographic background, and astrological sign—and asked participants how useful the questions would be to learn about the partner’s category membership. Each category was rated on a 7-point scale (ranging from −3 to +3). These ratings were followed by the question, “Which ONE category do you think the information from the questions would provide the most insight [on]?”

We first examined whether the ratings were above the midpoint on any of the categories and observed that participants thought the questions would reveal category membership for five out of the eight categories—personality, culture/geographic, SES, relationship status, and sexual orientation. Thus, participants believed that the 12 questions would reveal SES information, but their ratings indicated the belief that they would learn more about a person’s personality and culture than about their SES. Consistent with the ratings, participants chose personality (49%) as the one category that the questions would reveal, followed by culture/geography (19%). Just 9% of the participants ranked SES as the category they would learn most about from the questions. These data demonstrate that participants perceived that they would learn about someone’s SES from these questions, but that SES would not be the primary category gleaned from these questions—suggesting that the study’s intent to examine SES was not readily discerned.

Main study: manipulation checks and perceived partner SES and partner similarity

We assessed whether participants’ judgment of their own and their partner’s SES mapped onto our classification of them as low or high SES, respectively. In judging their own SES, the expected significant main effect of participant SES was observed; high-SES participants rated themselves higher than low-SES participants, b = 0.86, t(204.22) = 7.74, p < .001, R²_β = .23. There was also a significant main effect of partner SES; participants paired with low-SES partners rated themselves higher than participants paired with high-SES partners, b = −0.23, t(204.22) = −2.06, p = .040, R²_β = .020. In judging their partner’s SES, there was no main effect of participant SES, b = 0.16, t(219.03) = 1.51, p = .13, R²_β = .010; that is, participants’ own SES did not shape how they perceived their partners’ SES. Critically, the expected main effect of partner SES emerged, which indicated that participants accurately assessed their partner’s SES: low-SES versus high-SES partners, b = 0.28, t(219.03) = 2.66, p = .008, R²_β = .031.

We also examined whether participants’ ratings of their own and partner’s SES correlated with the SES index score. Participants’ SES rating correlated significantly with their SES index score, r(245) = .67, p < .001. Additionally, ratings of the partner’s SES also correlated significantly with the partner’s SES index score, r(244) = .23, p < .001.

Finally, we examined whether participants paired in similar-SES and different-SES dyads differed on perceptions of similarity with their partners. There was a main effect of participant SES: High-SES participants perceived less similarity with their partner than low-SES participants did, b = −0.23, t(199.50) = −2.32, p = .021, R²_β = .026. There was no main effect of partner SES, b = −0.07, t(199.50) = −0.70, p = .48, R²_β = .002. It is important to note that there was a significant Participant × Partner SES interaction: Similar-SES dyads reported greater similarity than different-SES dyads, b = 0.26, t(120.36) = 2.56, p = .012, R²_β = .051 (see the Supplemental Material for the other simple-effect comparison). Overall, these results suggest that (a) we successfully manipulated SES pairings, (b) participants accurately perceived the SES of their partner, and (c) similar-SES dyads perceived themselves to be more similar to each other than different-SES dyads did.

Physiologic PEP linkage

We examined the hypothesis that low-SES participants would show stronger physiological linkage to participants than high-SES participants. First, we included the main effects of stability (i.e., the effect of the respondent’s PEP reactivity at the prior time point on their PEP reactivity at the next time point) and influence (i.e., the effect of the partner’s PEP reactivity score at the prior time point on the respondent’s PEP reactivity score at the next time point), and all potential interactions between the stability and influence paths and the SES variables. The stability path was significant, F(1, 114) = 1742.75, p < .001, R²_β = .94, indicating that participants’ PEP reactivity was stable from one moment to the next. The main effect for influence was nonsignificant, F(1, 79.1) = 3.85, p = .053, R²_β = 0.046. Central to the hypothesis, we observed a significant Participant SES × Partner Prior PEP Reactivity interaction, F(1, 79.2) = 4.56, p = .036, R²_β = .054 (Fig. 2). For low-SES participants, the linkage path was positive and significant, t(66.8) = 2.96, p = .004, R²_β = .11, indicating that low-SES participants had positive linkage to their partners, irrespective of partner SES. Notably, for high-SES participants, the linkage path was not significantly different from zero, t(93.6) = −0.12, p = .905. Consistent with our hypothesis, these patterns suggest that low-SES participants are more socially attuned, and therefore more physiologically linked, to their partners than high-SES participants.

Fig. 2.

PEP linkage as a function of manipulated participant and partner SES across the entire interaction. The y-axis is the B path from the partner prior PEP reactivity score to the participant PEP reactivity score. Error bars represent standard errors of the estimate. PEP = preejection period; SES = socioeconomic status.

We also tested the hypothesis that participants would show stronger linkage to similar-SES than different-SES partners. However, the analysis revealed a nonsignificant Participant SES × Partner SES × Partner Prior PEP Reactivity interaction, F(1, 79.3) = 3.09, p = .083, R²_β = .03. This suggests that participants were not more socially attuned to similar-SES than to different-SES partners, contrary to what homophily would predict.

Behaviors

We tested the hypothesis that high-SES participants would show greater conversational dominance than low-SES participants. First, the analysis revealed no main effect of dyad type (a three-level variable that compared similar-SES high, similar-SES low, and different-SES dyads) on the absolute value of the conversational dominance for the dyad, F(2, 78) = .258, p = .773 (M_high-high = .571, SE = .874; M_low-low = .364, SE = .595). Within different-SES dyads, we further examined whether high- or low-SES participants were more likely to lead. The mean was positive for high-SES participants, although not significantly different from zero (M = .188, SE = .818, p = .09). Overall, these patterns did not support the hypothesis.

For interpersonal comfort, we hypothesized a partner SES effect in which participants would show stronger signs of interpersonal comfort with low-SES partners than high-SES partners. Consistent with the hypothesis, we observed a main effect of partner SES, F(1, 129.572) = 4.611, p = .034, R²_β = .03. As seen in Figure 3, people paired with low-SES partners appeared more comfortable interpersonally (M = .807, SE = .113) than those paired with high-SES partners (M = .471, SE = .017). We also tested the hypothesis that participants would show greater interpersonal comfort with similar-SES partners than with different-SES partners. However, no effects of participant SES or Participant SES × Partner SES were found (ps > .424), indicating that the effect of partner SES on participants’ displays of interpersonal comfort was consistent across high- and low-SES participants. In sum, these results suggest that low-SES partners overall elicited greater interpersonal comfort than high-SES partners, whereas similarity in SES did not elicit greater interpersonal comfort.

Fig. 3.

Interpersonal comfort as a function of participant and partner SES averaged across tasks. Error bars represent standard errors of the estimate. SES = socioeconomic status.

Partner liking

Here, we examined two hypotheses parallel to the hypotheses for interpersonal comfort. First, we hypothesized a partner SES effect in which participants would report liking low-SES partners more than high-SES partners. We examined the main effects of participant and partner SES, the main effect of time (structured interaction vs. cooperative game), and all interactions between SES and time. The analysis revealed no main effect of participant SES, F(1, 238) = .69, p = .40. Unexpectedly, there was no main effect for partner SES, F(1, 238) = .16, p = .686, contrary to hypotheses. We did find a significant Participant SES × Partner SES interaction, F(1, 206) = 5.60, p = .019, R²_β = .03. As seen in Figure 4, participants in different-SES dyads reported liking their partners less than those in similar-SES dyads. There was no main effect of time, t(121) = 0.01, p = .935, or interactions between participant SES, partner SES, and time (ps > .573), indicating that the observed participant-SES and partner-SES interaction effect was consistent across the two time points. Partner liking did not appear to be driven by interacting with socially attuned low-SES partners, but instead, was largely driven by interacting with similar-SES partners (i.e., homophily).

Fig. 4.

Self-reported liking as a function of participant and partner SES averaged across tasks. Error bars represent standard errors of the estimate. SES = socioeconomic status.

Correlations

The correlations between self-report and behaviors are presented in Table 2. Partner liking was only negatively associated with conversational dominance but not significantly associated with interpersonal comfort.

Table 2.

Correlations Among Self-Reports and Behavioral Measures

Measure	Partner liking	Interpersonal comfort	Conversational dominance
Partner liking	1
Interpersonal comfort	.039	1
Conversional dominance	−.172*	.039	1

^*p<.05.

Discussion

This research examined the interpersonal and intergroup processes that occur during dyadic social interactions with partners who are relatively low and high in SES, using a combination of physiological, behavioral and self-report measures. We observed that low-SES participants overall showed stronger physiological linkage than high-SES participants—indicating greater attunement to partners—regardless of their partners’ SES. Participants also appeared more comfortable when interacting with low-SES partners than with high-SES partners. These patterns align partially with the proposed interpersonal-attunement process shaped by stronger social orientation among lower SES individuals. They also add to the existing literature demonstrating the consequences of SES differences in social orientation (i.e., greater social monitoring, empathic accuracy, and perspective taking in lower-SES individuals; Dietze & Knowles, 2016; Dietze & Knowles, 2021; Kraus et al., 2010).

Contrary to prediction, high-SES participants did not show greater conversational dominance than low-SES participants. This may be due to the structured turn-taking nature of the interaction tasks, which was intended to ensure that both participants in the dyad had some opportunity to speak, but it may have unintentionally reduced the likelihood of conversational dominance emerging. Indeed, scholars have suggested that structured interactions may reduce status gaps in behavior (Mao & Feldman, 2019), and past research documenting status differences in dominance were in less structured contexts, such as doctor-patient interactions and classroom interactions (e.g., Palmer, 1989).

We also found that for liking, participants reported liking similar-SES partners more than different-SES partners, consistent with the homophily hypothesis. However, we did not find that participants liked low-SES partners more than high-SES partners, despite appearing more comfortable around them. In other words, engaging in a cross-class interaction, even with socially attuned low-SES partners, did not result in more liking of those partners. In a recent and related finding, although meaningful (i.e., substantive, involved) cross-class interactions were found to improve lower-SES students’ academic performance, such interactions were still rated as less satisfying than same-class interactions (Carey et al., 2022). This may imply that overriding in-group liking requires frequent or repeated cross-class interactions, whereas engaging in a few positive interactions, or a one-off positive interaction (as in our study), is insufficient (Byrne et al., 1966; Côté et al., 2017). Nonetheless, we note that our speculations relating to the impact of interaction structure on dominance and the impact of interaction frequency on liking are post hoc. Future research should examine how variation in interaction structure (e.g., using behavioral scripts versus not using them; Avery et al. 2009) and frequency (e.g., having or expecting a future interaction) may impact attunement, behaviors, and liking in interactions involving SES.

An important limitation of this research is that we did not examine possible influences of other intersecting social identities often correlated with SES, like race. Because of the demographics of the San Francisco Bay area, the ethnic composition of our dyads is skewed, with twice as many majority-racial-group dyads (i.e., European and Asian Americans) as minority-racial-group dyads (i.e., African Americans, Latino Americans). Therefore, we lacked sufficient power to examine interaction effects between SES and race or ethnicity. A critical unanswered question is whether our observed processes would hold when other intersectional identities are considered; might identities like race override or interact with SES? We also note that our categorization of Asian Americans as a majority race is unique to the demographic profile of San Francisco. This may not generalize more broadly, because Asian Americans are typically racial and ethnic minorities and receive distinct treatment compared with White Americans (Zou & Cheryan, 2017), who are traditionally categorized as the majority group in the United States.

The multidimensional SES index we used to pair dyads is also imperfect. One could argue that the higher weight given to income and subjective rank in the index may not generalize to contexts in which cultural measures of SES (e.g., education) are more central or relevant. Further, all our participants were recruited from San Francisco—an expensive U.S. city—so the socioeconomic distribution of our participants may not be entirely representative of the United States.

To conclude, our research demonstrated the interpersonal and intergroup processes that underlie a first-time encounter involving SES. We found that in such encounters, low-SES individuals exhibited greater attunement to others and elicited greater comfort in others than high-SES individuals, but impressions were formed in line with homophily effects. These findings reveal inherent asymmetries in attending to or eliciting comfort in others that may undermine initial and future cross-class interactions. As opportunities for diverse encounters increase, the interaction dynamics revealed by our research should be noted and addressed to facilitate cross-class interactions. This may go a long way toward improving intergroup attitudes and mitigating the socioeconomic divide.

Supplementary Material

Additional supporting information can be found at http://journals.sagepub.com/doi/suppl/10.1177/09567976251350970

Statement of Relevance.

With greater attention toward increasing socioeconomic diversity across communities, workplaces, and schools, it is unclear how socioeconomic status (SES) differences between individuals shape dyadic social interactions. This research investigated how two strangers, from either similar- or different-SES backgrounds, responded to each other during a casual social interaction. Throughout the interaction, low-SES individuals were found to be more attuned to their interaction partners in general—indicated by their physiological changes that followed or linked to their partners’ physiological changes more strongly—than high-SES individuals. Low-SES individuals’ greater attunement also appeared to elicit greater comfort among the individuals interacting with them. Nonetheless, individuals still preferred and liked interactions with someone from a similar-SES background rather than a different-SES background. These patterns provide initial insights into what happens during casual social interactions involving people of different-SES backgrounds that may differ from what is captured by self-reports.