Thin Slices of Initial Group Rapport Predict Social Bonding

Abstract

Rapport plays a key role in small group interaction, yet there remain gaps in understanding the construct. The current study examined whether initial group rapport among triads of strangers could predict later social bonding and group emotion. Results indicated that initial group rapport reliably predicted subsequent social bonding and emotional experience across multiple measures. These findings support use of global thin slice rapport measures in small groups. Further, they suggest that rapport can be assessed reliably within the first minute of meeting and that ratings of group experience in just this first minute offer valuable insight into subsequent group formation.

Keywords

rapport thin slice triads FACS group formation

Introduction

Most of us can identify what it feels like to have rapport. We use the term in our daily lives to describe a sense of pleasant accord in all sorts of interactions: between romantic partners, work groups, acquaintances, groups of friends, teachers, and their pupils. Rapport is a nearly universally recognized and often sought-after social group experience, but the concept of rapport itself is difficult to capture with language. Generally, it describes the felt sense of harmony and “clicking” in an interaction, though these words fall short of conveying the full gestalt experience (Bernieri, 2005). The murkiness of rapport as a concept has made it difficult to operationalize and measure. This is unfortunate, given its clear resonance with human experience and its many potential social psychological implications. The relational experience of rapport, for example, was long ago described as part of the development of a collective mind in crowds (Park & Burgess, 1921). More recent work has also demonstrated the import of this construct in our day-to-day social lives: rapport has been associated with satisfaction in interactions with medical providers; motivation and learning in students; and increased feelings of autonomy, competence, and relatedness (Baker et al., 2020; Hall et al., 2009; Wilson et al., 2010). More scientific scrutiny of this potentially powerful source of mutual influence and social behavior is warranted.

In their seminal work, Tickle-Degnen and Rosenthal (1987, 1990) defined rapport as a dynamic, emergent quality of interactions, composed of three essential components: mutual attentiveness, positivity, and coordination. This definition of rapport has served as the basis for the measurement of rapport in a number of subsequent investigations. However, while their definition has been heuristic for the field, measurement of rapport remains a challenge. In the current discussion, we first clarify difficulties in the assessment of rapport. We then present empirical data for a possible alternative measurement approach. As is true of many studies in this research area, our conceptualization of rapport is based on Tickle-Degnen and Rosenthal’s (1987, 1990) work. We aim to capture the dynamic and emergent quality of groups that allows participants to “click” and feel at ease in an interaction.

Assessment of Rapport

Measuring rapport presents numerous difficulties. First, as Tickle-Degnen and Rosenthal (1990) point out, rapport varies across time and context. Rapport between a married couple discussing vacation plans likely looks different than rapport between teachers and their pupils, or rapport among members of a little league team. The challenge in measuring rapport comes down to capturing the underlying universality of the experience of “clicking” in an interaction while recognizing that this experience may present itself differently in different interactions. To add further complexity, because rapport exists uniquely within interactions, a valid measure of rapport ought to capture its gestalt nature and attend to more than just the experience of individuals in an interaction (Bernieri, 2005).

Self-report measures are a common method for assessing rapport. For instance, Bernieri and colleagues developed a questionnaire that assesses emotional tone, mutual focus, physical coordination, and harmony (Bernieri, 1988; Bernieri et al., 1994; Bernieri, 2005). One benefit to this scale is that, unlike many self-report measures, it does not rely solely on individual experience, but instead on perceptions of the whole interaction. This questionnaire has helped to demonstrate the varying importance of interactional synchrony to the perception of rapport across genders and situations (Bernieri et al., 1994). Nevertheless, all self-report measures necessarily rely on the perspective of single interactants. Accordingly, individual self-report scales may struggle to capture emergent properties of the interaction. Indeed, self-report measures of rapport have demonstrated low intergroup agreement—individuals do not always agree on how much rapport exists between them (Bernieri, 2005; Bernieri & Gillis, 2001). And yet, congruence among individuals is central to the definition of rapport. Consider the case when one participant rates an interaction positively while their partner rates it negatively. Surely, neither participant’s individual rating is representative. One could average interactants’ scores, but this method still fails to account for the relationship between the scores. An ideal measure of rapport would extend beyond the sum of individual interactant’s perceptions.

Perhaps motivated by some of the drawbacks to self-report measures, investigators have worked to develop behavioral measures of rapport. Nonverbal behavior is a powerful source of communication and thus a key element in the development of rapport. Tickle-Degnen and Rosenthal (1990) observed that specific behavioral indicators (e.g., partner’s smiling) relate to distinct components of rapport. Similarly, Bernieri et al. (1996) found that participant self-reports of rapport were predicted by a number of non-verbal behaviors (Bernieri, 2005; Bernieri et al., 1996). Other research suggests that groups with high levels of rapport are characterized by behavioral coordination (see Levine & Moreland, 1998). There are, however, lingering questions for researchers looking to apply a behavioral approach.

First, the interpretive meaning of non-verbal behavior is context-dependent, and as a result, the predictive validity of individual behaviors changes across situations (Bernieri, 2005). For instance, being close to one’s partner may signify social comfort or aggression, depending on the circumstances. Second, to our knowledge, there are no clear guidelines on how to combine concrete behaviors into a global measure of rapport. Should researchers choose a single behavioral measure, or instead use a variety of codes that are somehow combined into a composite rapport index? If composite behavioral indices are to be used, how should these behaviors be combined, what weight should be given to each individual behavior, and how do these decisions change based on the specific social context assessed? Finally, most documented behavioral correlates of rapport have been identified in dyadic interactions based upon the strength of their relationship with a self-report rapport criterion measure. As noted above, it is unclear whether self-report measures are well-suited to assess the emergent quality of rapport. Comprehensive cross-validation with other criterion measures is warranted.

An alternative approach to behavioral coding is to integrate discrete behavioral events into a molar measure by considering outside observers’ impressions of an interaction (Tickle-Degnen & Rosenthal, 1990). Humans are skilled social perceivers who likely depended, throughout their evolution, on the ability to quickly perceive and integrate behavioral cues (Frank & Solbu, 2020). Rather than measuring discrete pieces of the interaction and painstakingly determining which of these behaviors are relevant to rapport in various contexts, observers may be able to view an interaction and instantaneously form a gestalt impression that integrates behaviors with the environmental context and interactional sequence in which they occurred.

Various studies have examined global observer ratings of dyadic rapport, mostly from brief clips of a longer interaction. Largely, studies have focused on testing the validity of these measures by comparing them to a self-report criterion or correlating the global ratings with nonverbal behaviors (e.g., Bernieri & Gillis, 1995; Bernieri et al., 1996; Gillis et al., 1995; Grahe & Bernieri, 1999). This work has informed understanding of how human judges use cues to perceive social information. Few studies, however, have examined the predictive validity of observer-coded rapport by correlating it with the outcomes it ought to predict. These few studies have demonstrated that thin slice measures of rapport can be effective in predicting a range of socially and clinically meaningful outcomes (Hall et al., 2009). Thin slice global ratings of rapport remain a promising direction for future research.

Thin Slice Assessment

For over half a century, researchers interested in expressive behavior, communication, psychopathology, and interpersonal judgments have used brief clips of a longer behavioral stream to draw conclusions and make predictions about the entire interaction, or about the individuals observed (e.g., Milmoe et al., 1967). Ambady and Rosenthal (1992) reviewed the emerging literature and concluded that measures based on “thin slices” of the behavioral stream (e.g., judgments of specific characteristics or coded nonverbal behavior based on as little as 30 seconds of an interaction) can predict important real-world outcomes, such as teacher effectiveness, deception, and even clinical depression. These findings are striking. Apparently, there is sufficient information contained in a 1-min (or briefer) snippet of human behavior to predict phenomena as complex and multifaceted as depression (Ambady & Rosenthal, 1992).

In the decades following Ambady and Rosenthal’s (1992) review, a flourishing thin slice literature has emerged. Thin slice measures have predicted individual-level outcomes as diverse as personality traits, gender, sexual orientation, teacher effectiveness, and even doctor-patient satisfaction (Ambady et al., 2000; Saville & Balas, 2014). Other studies have contributed to the clinical literature. Mounting evidence, for example, suggests that thin slice measures predict the presence of anxiety, depression, and personality disorders, as well as patient clinical outcomes (Slepian et al., 2014).

The Current Study

Compared to its focus on individual traits and experiences, a relatively small portion of the thin slice literature has focused on perception and prediction of social processes (Saville & Balas, 2014), such as judging the relationship status of dyads (Ambady & Gray, 2002). Even less work has examined the ability of observers to perceive dynamic group experiences from a thin slice of the interaction. This limited focus on small group processes is unfortunate given that collective experiences, such as collective emotions, tend to be stronger in magnitude and longer lasting than their individual counterparts (Goldenberg et al., 2020). As discussed above, the rapport literature stands out in this regard, as researchers have leveraged thin slice measures of the dynamic experience of rapport. However, with just a few exceptions (e.g., Hall et al., 2009), these studies tend to rely on self-reported social outcome measures in dyads. Accordingly, the primary aim of this study was to evaluate the predictive utility of a global thin slice measure of initial rapport for capturing (across a dynamic, multimodal set of measures) subsequent social experience during small group formation.

This study offered several features that distinguish it from prior investigations. First, a major characteristic of the thin slice rapport literature has been its reliance on dyads. Dyads, however, represent a unique form of social group which necessarily struggle to address certain social constructs of great interest to small groups researchers (e.g., social exclusion, in-group/out-group behavior). Studies examining groups larger than dyads are therefore needed to generalize our understanding of how initial rapport predicts subsequent social functioning in traditional group settings. The extant literature leaves largely unaddressed how rapport functions and can be assessed in small groups. The present study examined three-person groups, which are less unwieldy than still larger groups, yet include elements of group experience not found with dyads (Levine & Moreland, 1998). We believe that the data presented here will help to extend the rapport construct to small group experience.

Second, the present study focused on the initial moments of social integration, when members are unacquainted with each other and interactions tend to be ambiguous (Leary & Kowalski, 1995). While prior rapport studies have examined thin slice measures among relatively unacquainted individuals, we are unaware of any that have targeted the very first moments of group formation. This is noteworthy, as the adjectives used to describe rapport (e.g., clicking) often emphasize the spontaneous and immediate quality of this experience. Moreover, there is compelling evidence that initial impressions can be formed immediately (Willis & Todorov, 2006) and that first impressions can have important long-term relational consequences (Human et al., 2013).

Third, prior research examining the ability of thin slice rapport judgments to predict dyadic experiences has relied on structured paradigms (e.g., experimental tasks). Such work has yielded interesting data regarding the relationship between non-verbal and self-report measures of rapport, and observer judgment policies (e.g., Bernieri & Gillis, 1995; Bernieri et al., 1996; Grahe & Bernieri, 1999). Thin slice rapport research also is needed, however, to examine unstructured social interaction, in which group members reciprocally and dynamically affect each other without being confined to specific interactional behaviors (Ickes & Gonzalez, 1994).

While unstructured thin slice rapport research using groups larger than dyads would advance understanding of group formation processes, it is not altogether surprising that, to our knowledge, such work has yet to be conducted. Trios of unacquainted persons require larger samples, especially when one statistically accounts for the interdependence of group members (Creswell et al., 2012) and considers the noise inherent in unstructured group paradigms. The large sample size creates particularly labor-intensive challenges if one aims to unobtrusively capture moment-to-moment fluctuations in emotional responses, which is crucial when studying dynamic, coordinated social interaction. For instance, using Paul Ekman’s Facial Action Coding System (Ekman et al., 2002) to assess minutes of socio-emotional experience can quickly lead to millions of frames of coding (e.g., Sayette et al., 2012). Further, capturing the very first moments of group formation presents numerous methodological challenges. One must ensure that participants are not previously acquainted (which is particularly complicated when recruiting in schools or university communities) and prevent participants from accidentally meeting each other, even as they arrive at the lab for their study session.

In summary, the present study leveraged a unique database and conducted new behavioral coding to test the ability of a global observational thin slice measure of initial group rapport among strangers to predict subsequent social experience. This study boasted several novel features, including a large sample of triads, unstructured social interaction, numerous methodological checks to ensure assessment of the very first moments of group formation, and a multimodal assessment of subsequent social experience that required detailed coding of millions of frames of video. We aimed to test the validity of our global observational approach to assessing a thin slice of initial group rapport using multiple criterion (outcome) measures, including a group-level, dynamic, real-time nonverbal criterion measure. By examining the impact of initial group rapport on subsequently recorded measures of bonding and emotion, we hoped to better understand the influence of initial rapport in group formation processes.

Method

This investigation made use of archival data and novel behavioral coding from a larger study examining the effect of alcohol consumption on social bonding (Sayette et al., 2012). In the original study, groups of three strangers were seated around a table and allowed to interact freely for a period of 36 min. The interactions were recorded and participants’ real-time affect-related facial expressions were coded by researchers certified in the Facial Action Coding System (FACS; Ekman et al., 2002). Specific to the current investigation, we generated new data from the existing videos by coding initial group rapport in the first minute of these interactions. We then assessed the relationship between the initial group rapport variable and subsequent real-time measures of positive affect and group bonding. Further, to ensure relevance to the broader small groups literature, our current analyses focus solely on the subsample of groups from the original study who knowingly consumed juice, and neither expected nor consumed alcohol.

Participants

In the original study conducted by Sayette et al. (2012), healthy male and female subjects between the ages of 21 and 28 were recruited to participate in a social drinking study at the University of Pittsburgh. Participants underwent both a phone and an in-person screening and were excluded for medical conditions contraindicating alcohol, having a body weight 15% more or less than the ideal for their height (on the basis that alcohol dosing procedures may be different for body weights outside of this range), as well as for past alcohol abuse or dependence. The current investigation focuses on the 80 groups of three participants who completed study procedures while consuming a juice control beverage. Two groups were excluded from the current analyses due to missing video data. The final sample reported here, therefore, included 78 groups of three (234 individuals). Gender composition was relatively evenly distributed across these groups, with 20 groups of all males, 20 groups with one female and two males, 19 groups with two females and one male, and 19 groups with three females. Participants were on average 22.52 years old (SD = 2.02) and 81.60% of participants identified themselves as white, 9.83% as black, 1.71% as Hispanic, 4.70% as Asian, and 2.14% as other.

Procedure

During an in-person screening session, participants underwent informed consent procedures and were asked to fill out questionnaires pertaining to their medical history. Demographic information was also collected. Eligible participants were invited to revisit the lab on another day for the study session.

A random group of three participants came in for each study session. Participants were verified as strangers and did not interact before the moment they were seated together for the study procedures (see Sayette et al., 2012). Specifically, to ensure that participants were unacquainted at the outset of the study, we overbooked for each session (inviting four or five participants to each session) to account for chance acquaintanceships (always sending at least one person home to ensure three strangers were used) and used both self-report and behavioral observation to detect such familiarity. Participants were also individually greeted and seated in separate waiting rooms to avoid pre-study discussion and then were moved simultaneously to the study room without an opportunity to speak (the experimenter provided instructions and walked them briskly down the corridor as they entered the room) until the very moment the initial rapport measurement began.

Prior to meeting group members, participants completed a battery of assessments, including measures of personality and affect reported in Sayette et al. (2012). The experimenter then poured a glass of juice for each participant and the group was seated together around a circular table (for additional details on beverage administration in conditions other than the juice control described here, see Sayette et al., 2012). Once the experimenter exited the room, the group was allowed to interact freely for 36 min. At 12 and 24 min past the start of the interaction period, the experimenter entered the room to refill their juice glasses. The entire 36-min interaction was recorded at 30 frames per second. Cameras were strategically mounted around the room to allow for a head-on view of each participant as well as an overhead view of the whole group. Participants were made aware of the cameras and told that they were there so that the experimenter could monitor their drink consumption rates. At the end of the experiment, participants learned that we had also recorded these interactions and they were offered without consequence the option to have the video erased. Nobody chose to do so.

After 36 min, the group was separated and participants completed the Perceived Group Reinforcement scale (described below), an eight-item mood measure, and a series of experimental tasks (which are not relevant to the current investigation). Participants were then debriefed and allowed to leave.

Assessment of Initial Group Rapport

In line with Tickle-Degnen and Rosenthal’s (1987, 1990) research, in this investigation rapport was conceptualized as a group experience reflective of a harmonious, pleasant interaction marked by a high degree of comfort, attention, and synchronicity among members. In colloquial terms, this experience is well summed up by the experience of “clicking” within a group.

While a primary aim of the current investigation was to assess the validity of a global thin slice measure of rapport, our choice of assessment was not without an empirical basis. Specifically, to support the initial validity of our approach, we implemented a measure of rapport based on research from Hall et al. (2009). In their study, Hall et al. (2009) measured rapport between medical students and standardized patients using observer impressions. Research assistants watched brief clips of each doctor-patient interaction and coded rapport on a 1 to 9 scale (no rapport to high rapport). Rapport, measured in this way, was significantly correlated with a number of ecologically valid variables, including the emotional skills of the doctor and peer ratings of the doctor’s interpersonal sensitivity (Hall et al., 2009).

Similar to Hall et al. (2009), in the current investigation, initial group rapport was rated by two research assistants trained to code rapport reliably (see below for details). The rapport measure used in the final analysis reflects an average of the two raters’ scores. Each coder viewed a 1-min clip from the beginning of all group interactions. Coders then rated rapport on an 8-point Likert scale ranging from 0 “no rapport” to 7 “high rapport”¹. In line with past research, coders did not listen to audio associated with the video as it is suspected that the verbal content may diminish attention to other aspects of the interaction (see Grahe & Bernieri, 1999). Coders watched each clip one time and were instructed to watch the full minute of interaction before deciding on a rating. Coders were given the following definition of rapport, adopted from Hall et al. (2009, p. 324), “Rapport is defined as: a relationship that is pleasant and engaging, with a high degree of liking or positive affect, mutual attention, harmonious relation, easy/smooth communication, and/or symmetry and synchrony in the interaction.”

Training procedures for the coding of initial group rapport were based on the recommendations of Blanch-Hartigan et al. (2018). The first author worked directly with two undergraduate research assistants to train up to reliability. The training period started with a brief discussion of the definition of group rapport (i.e., the definition provided above) as well as review of a sample video from a pilot dataset. Next, the first author, along with the two research assistants, coded four blocks of 5 to 10 training videos and conducted consensus discussions on each coder’s ratings after each block. Reliability was calculated after each coding block and found to be sufficient after the fourth (consistency and agreement ICC were both >0.93). Then, to confirm that reliability was maintained in the study data, the coders rated nine videos from the full sample and held consensus discussions. Three of these videos were from groups in the juice control condition discussed here. These ratings are retained in the present dataset. Major themes that came up in consensus discussions included: viewing rapport as the average over the entire 1-min video clip, the importance of attending to the group as a whole, and the importance of focusing on gross impressions rather than specific behaviors. Throughout consensus discussions, coders also noted that it was useful to base rapport ratings on how uncomfortable or at ease the coder imagined they would feel as a third-party observer in the room during the interaction.

After this training period, the two undergraduate coders independently rated initial group rapport in the first minute of each of the group interactions. Coder’s ratings were averaged for the final initial group rapport measure.

Outcome Measures

The outcome variables used to evaluate our measure of initial group rapport have been detailed in Sayette et al. (2012). Here, we describe each measure and briefly remark on the validity of these measures.

FACS Coding

Participants’ affect-related facial expressions were coded throughout the interactions by research assistants certified in the Facial Action Coding System (FACS; Ekman et al., 2002). Relevant to the current investigation, this facial data allowed for unobtrusive assessment of real-time emotion and group bonding throughout participants’ interactions. Here, our primary aim was to examine the association of initial group rapport with these theoretically relevant and previously validated criterion measures.

Coders viewed video of each participant’s face throughout the interaction and coded every frame (one-thirtieth of a second) for the presence or absence of relevant facial Action Units (AUs). Here our analyses focus on Duchenne smiles, defined by the combination of AUs 6 and 12 (Ekman, 1989). This combination of AUs is widely thought to represent enjoyment (Ekman et al., 1990, 2002). Because our current hypotheses centered on the relation between initial rapport and subsequent group experience, our analyses focused only on facial affect expressed in the final 18 min of the interactions (excluding the 2 min when the investigator entered the room to refill participants’ drinks). Duchenne smiles in the final 18 min of the interactions were used to index both experiences of real-time positive affect as well as real-time group bonding. These measures are further detailed below.

Expressed Positive Affect

Participants’ experiences of positive affect during the interaction were indexed with the total number of frames in which each participant displayed a Duchenne smile in the final 18 min (excluding 2 min in which investigators entered the room to refill drinks) of the interaction. This outcome measure reflects the amount of time each participant spent expressing happiness in the latter half of their interactions. Duchenne smiles have been found to reflect positive affect (Ekman et al., 1990; Frank et al., 1993) and are associated with multiple indicators of positive social experience including perceptions of altruism, cooperation, and self-reports of love (Brown et al., 2003; Gonzaga et al., 2001; Mehu et al., 2007). These data suggest that Duchenne smiles are a suitable criterion measure for the assessment of rapport.

Expressed Group Bonding

Real-time group bonding was operationalized as the number of frames during which every member of the group displayed a Duchenne smile simultaneously (Kirchner et al., 2006). Henceforth, we refer to these instances as “triadic smiles” and conceptualize them as experiences of shared joy and affiliation among group members. The final expressed group bonding measure reflects the amount of time groups spent expressing a triadic smile in the final 18 min of their interaction (excluding 2 min during which investigators entered the room to refill drinks). This measure has a great deal of face validity as a criterion measure for rapport as it assesses both the positivity and coordination components of rapport as defined by Tickle-Degnen and Rosenthal (1987, 1990).² Moreover, research suggests that synchronized facial affect predicts perceptions of social connection between dyads (Cheong et al., 2020) and that triadic Duchenne smiles in particular are correlated with other measures of social connectedness, such as coordinated speech turns (Kirchner et al., 2006).

Self-Report Social Bonding

In addition to real-time measures of affect and bonding, we examined the relationship between observer-coded initial group rapport and a more traditional self-report index of group bonding (completed after the 36-min interaction), using the Perceived Group Reinforcement Scale (PGRS; Kirchner et al., 2006). The PGRS includes 12 Likert-style items such as “I like this group” and “The members of this group are interested in what I have to say.” Each item is rated on a 9-point Likert scale ranging from 1 “strongly agree” to 9 “strongly disagree” (α = .90). The total score reflects an average of the items, with higher scores reflecting stronger bonding. Items on this scale were adapted from the Group Attitude Scale (Evans & Jarvis, 1986) and the Perceived Cohesion Scale (Bollen & Hoyle, 1990), both of which have been previously validated. Specifically, the Group Attitude Scale predicts numerous measures related to group bonding, such as interpersonal attraction among group members, attendance to group events, and group cohesion (Evans & Jarvis, 1986). The Perceived Cohesion Scale has a strong two-factor structure assessing the constructs of group belonging and morale (Chin et al., 1999). Moreover, past investigations have shown that the PGRS itself is correlated with non-verbal measures of social bonding (Kirchner et al., 2006).

Analysis Plan

The primary aim of our analyses was to test whether initial group rapport during a small group interaction predicts later [i.e., the final half (18 min) of the interaction] measures of group bonding and positive affect. Toward this aim, data were analyzed using a regression framework, with initial group rapport predicting each outcome variable in separate models. Because our data had a nested structure (individual participants in groups of three), outcomes measured at the individual level (i.e., expressed positive affect and PGRS scores) were analyzed in a multilevel modeling framework (MLM). The effects structure of regressions with group and individual outcomes are displayed below:

Group-level outcomes (i.e., Triadic Smiling):

T r i a d i c S m i l e s_{j} = β_{0} + β_{1} R a p p o r t_{j} + ε_{j}

(1)

Where β₀ represents the estimated number of frames spent expressing triadic smiles in groups with average initial rapport, and β₁ represents the predicted increase in triadic smiles for a 1 -unit increase (between groups) in initial group rapport.

Individual-level outcomes (i.e., expressed positive affect; PGRS scores):

\begin{array}{l} I n d i v i d u a l O u t c o m e_{i j} = β_{0 j} + ε_{i j} \\ β_{0 j} = γ_{00} + γ_{01} R a p p o r t_{j} + u_{0 j} \end{array}

(2)

Where γ₀₀ represents the estimated value of the cluster random intercept (β_0j) in groups with average initial rapport, and γ₀₁ represents the predicted increase in the random intercept (β_0j) for a 1 -unit increase (between groups) in initial group rapport.

In addition to the multilevel nature of our data, another challenge is that the operationalization of positive affect and expressed group bonding relied on facial data that were indexed with a count variable—the number of frames during which an individual or a group displayed a facial expression—and deviated significantly from a standard normal distribution (Shapiro-Wilk normality test, both p < .001). To address this challenge, we conducted analyses of facial affect data in a beta regression framework using transformed data (see below). This modeling approach allows for more flexible variance structures than linear regression (Zimprich, 2010), and therefore was better suited to our data. For optimal transparency, we also briefly report on results in a linear regression framework.

Beta Regressions

Facial affect data were skewed, kurtotic, and bounded between zero and the total number of frames reviewed for facial expressions (28,800) across the final 18 min of the interaction (excluding 2 min when experimenters entered the room). Beta regression is commonly used to model proportions falling on the unit interval (0–1) and is a recommended approach to deal with the combination of skewness, kurtosis, and bounded data (Smithson & Verkuilen, 2006). Thus, facial affect variables were rescaled as proportions ranging from 0 to 1 and transformed to deal with the extreme value of zero (Cribari-Neto & Zeileis, 2010; Smithson & Verkuilen, 2006). The formulas used to rescale and transform facial affect variables are displayed below:

\tilde{Y} = \frac{Y (n - 1) + 0.5}{n},

(3)

where

Y = \frac{c o u n t}{n},

(4)

Here, count represents the facial affect response variable (i.e., positive affect, expressed group bonding) and n represents the total number of frames examined (i.e., 28,800).

Beta regressions, fitted with a logit link function, were then conducted on the transformed facial affect variable ( $\tilde{Y}$ ). Because of the logit link function, parameter estimates in beta regression can be interpreted as they are in logistic regression, where the odds of the outcome variable increase by a factor of exp(β₁) per 1 -unit change in the predictor variable.

Results

Data processing and summary statistics were conducted with R Statistical Computing Software (v4.0.5; The R Foundation, n.d). Table 1 displays the summary statistics for each outcome measure and the rapport predictor. Beta regressions were conducted with SAS v9.4 software (SAS Institute, 2020) using the GLIMMIX procedure for beta regressions. Both beta regression models relied on default estimation techniques; for the multilevel beta regression examining positive affect, this was residual pseudo-likelihood estimation, for the beta regression examining expressed group bonding, this was maximum likelihood estimation. All linear multilevel models were conducted using the lme4 package version 1.1-29 in R Statistical Computing Software (Bates et al., 2015) and employed restricted maximum likelihood estimation as the default estimation technique. For every linear multilevel model we report the corresponding ICC value to examine the level of clustering within groups.³

Table 1.

Summary Statistics for Variables in Confirmatory Models.

Variable	Mean (SD)
Variable	Juice Control (n = 234)	All beverage conditions (n = 705)
Initial Group Rapport (0–7 scale)	3.35 (1.30)	3.56 (1.37)
Duchenne Smiles	2,738.18 (2,019.01)	3,238.03 (2,485.74)
PGRS (1–9 scale)	7.06 (1.30)	7.02 (1.35)
Triadic Smiles	496.05 (570.64)	666.80 (773.86)

Note. Facial affect data are reported as the total number of frames during which the expression was exhibited, frames were approximately 1/30th of a second.

Rapport Measurement

Average initial group rapport on the 0 to 7 scale was 3.35 (SD = 1.30). An ICC estimate for the coders’ ratings of initial group rapport was calculated using the irr package in R based on a mean-rating (k = 2), two-way mixed-effects model, and consistency agreement ⁴. The ICC consistency value was 0.85 (95% CI [0.77, 0.91]) indicating good reliability.

Tests of Hypotheses

Expressed Positive Affect

To examine the relationship between initial group rapport and participants’ positive affect, we ran a multilevel beta regression predicting positive affect with initial group rapport and a random intercept for the group of three participants. Results indicated that initial group rapport significantly predicted positive affect, such that individuals in groups with higher initial rapport had greater odds of expressing Duchenne smiles in the last 18 min of their interactions (B = 0.12, t(156) = 2.23, p = .03). Specifically, this was a small effect, with the odds of expressing a Duchenne smile increasing by 12.55% (95% CI [1.38%, 24.96%]) for a 1 -unit increase in initial group rapport. An analogous linear regression model similarly indicated that individuals in groups with higher initial rapport spent more time expressing Duchenne smiles during the final 18-min of their interactions, with a small to medium effect size (β = .19, t(76) = 2.26, p = .03).⁵ The ICC value for this model was 0.33, suggesting that 33% of the variance in Duchenne smiles is explained by the clustering of participants within groups.

Expressed Group Bonding

To examine the relationship between initial group rapport and expressed group bonding, we conducted a beta regression predicting triadic smiles with initial group rapport. Results again indicated that initial group rapport predicted group-level bonding, such that groups with higher levels of initial rapport had greater odds of expressing a triadic smile in the last 18 min of their interactions (B = 0.26, t(76) = 3.71, p < .001). Specifically, this was a small effect; the odds of expressing a triadic smile were predicted to increase by 30.32% (95% CI [13.03%, 50.25%]) for a 1 -unit increase in initial group rapport. An analogous linear regression model also demonstrated a positive relationship between initial group rapport and time in triadic smiles, however, this relationship did not reach significance (β = .16, t(76) = 1.43, p = .16). Notably, model fit statistics indicate that the beta regression was a better fit for these data; beta regression Akaike information criterion (AIC) and Bayesian information criterion (BIC) were −502.71 and −495.64 respectively, while linear regression AIC and BIC were 1214.37 and 1221.44 respectively.

Self-Report Group Bonding

To assess the relationship between initial group rapport and self-report group bonding, we conducted a multilevel linear regression predicting PGRS scores with initial group rapport and a random intercept for the group of three participants. Results revealed a significant relationship between initial group rapport and self-report group bonding, such that individuals in groups with higher rapport tended to report more group bonding with a small to medium effect size (β = .19, t(76) = 2.58, p = .012).⁵ The ICC value for this model was 0.18, suggesting that 18% of the variance in self-reported group bonding is explained by the clustering of participants within groups.⁶

Confirmatory Analyses and Gender

To examine the robustness of our measure of rapport, and to rule out the possibility of a gender confound, we reran our primary confirmatory analyses (beta regressions for facial affect variables) while controlling for the gender composition of each three-person group. Overall, the pattern of results observed was consistent with those reported in our initial analyses: rapport, independent of group gender composition, significantly predicted positive affect, expressed group bonding, and self-report group bonding.⁷

Discussion

The construct of rapport during a small group interaction holds considerable appeal as a subject of scientific scrutiny. The experience of “clicking” in an interaction is common, invigorating, and often desired. It is not hard to imagine many experiences that rapport may influence: doctor-patient outcomes, negotiations, dating, interviews, group behavior in sports bars, fraternities, at concerts. . .the list goes on. Yet, despite the potential importance of rapport for a range of social interactions, many questions remain unanswered.

This study examined the predictive validity of a global, thin slice, measure of group rapport in the very first moments of triadic group formation. We are unaware of other small group investigations that have tested how thin slice measures of rapport reveal subsequent dynamic group experience across the interaction. Our study boasted several notable features for this field of research. Namely, it included multimodal coding of more than 230 participants engaging in three-person interactions. Further, our study went to considerable lengths to ensure that each trio was comprised of previously unacquainted participants who were not permitted to meet until the very moment that the interaction of interest began (and was recorded). Such efforts permitted observation of the very first moment of initial group formation—overcoming the typical challenges facing investigators when subjects happen to be acquainted or begin to chat in the waiting room prior to the study outset. These procedures provided a unique opportunity to examine initial rapport during small group formation in an unscripted exchange, with sufficient power to consider group-level indicators of social experience.

Results reveal that, in the first minute of group formation, trained observers were able to reliably judge initial group rapport, and further, that these judgments successfully predicted subsequent group experience. This latter finding emerged not only using traditional self-report measures of social bonding but also dynamic, nonverbal measures of group experience. These results are nontrivial, as display rules of initial engagement (e.g., smiling during a greeting) that may not necessarily reflect true emotional experience (see Ekman & Friesen, 1969) could hinder the ability of the initial thin slice rapport measure to predict subsequent social-emotional responding.

These results represent a notable contribution to the thin slice literature, and more specifically, to the rapport literature. First, our study is fairly unique in its focus on thin slice assessment and prediction of dynamic social experience. While much thin slice research has focused on individual difference variables (e.g., personality), our work indicates that thin slice methods can reliably assess and predict dynamic group processes. Assessment of such group processes often otherwise requires laborious coding and we hope that the present findings encourage future research to explore thin slice methodologies as an efficient and fruitful alternative.

In terms of the rapport literature, the current study provides evidence, within the context of a low-stakes social interaction among strangers, to support the use of a global measure of rapport. Moreover, our study differs from most studies, which have focused on structured dyadic interactions (e.g., Bernieri et al., 1996). To our knowledge, the current study is the first to demonstrate the utility of rapport measurement in unstructured triadic interactions. As noted above, dyadic interactions differ in important ways from larger group experiences, and we believe that the focus on trios in the present study adds important data to this research literature.

Finally, our findings reveal that, during an unstructured small group interaction, the rapport coded during the initial minute of group formation predicted subsequent socio-emotional experience as assessed across a multimodal set of measures. This adds to the body of research suggesting that first impressions and the initial moments of an interaction have impacts on subsequent relational experiences.

While this study provides important initial data to support the global assessment of rapport in group contexts, this work is not without limitations. First, our data only permitted the assessment of rapport within a very specific and controlled interactional context: randomly assigned groups of three strangers in a laboratory environment. However, as others have suggested, the predictive utility of observer-coded rapport may vary across contexts (e.g., cooperative and adversarial, Bernieri et al., 1996). Consequently, it will be important for future research to examine the generalizability of global measures of rapport across a diverse array of social contexts, ranging from group negotiations and debate to moments of humor and celebration. Second, unlike many natural interactions, participants in our database were required to remain seated together for a period of 36 min. This constraint could have limited the strength of the relationship between initial rapport and subsequent experience in our dataset, as “captive” participants in our paradigm may have been motivated to overcome any initial awkwardness or feelings of low rapport because they are aware that they will have to endure interacting with each other for a prescribed amount of time. Future research would do well to employ paradigms that permit interactants to exit the group if desired, as the effect of rapport on subsequent experience is likely to be even stronger in such interactions. Finally, our dataset was composed of young, largely white adults, and as we have noted elsewhere, it would be important to investigate initial rapport in more diverse groups (Fairbairn et al., 2013). Such work could have implications for both basic theory and critical real-world outcomes (e.g., racial disparities in healthcare and policing).

In summary, data suggest that global, thin slice measures of rapport provide a useful, reliable, efficient, and valid means for researchers to examine this important group construct. In particular, we have demonstrated that rapport, assessed during the very first minute of group formation, offers a reliable signal of future group experience. We hope that these findings reinvigorate the study of group rapport in small groups research and encourage future examinations regarding the development, maintenance, and influence of rapport across a diverse array of social groups. More broadly, our study supports the use of thin slice methods to predict dynamic group experience. We, therefore, hope that this work stimulates a host of thin slice investigations on emergent social processes.

Footnotes

Acknowledgements

We thank Drs. Aidan Wright and Amanda Forest for their helpful comments on a prior draft of the manuscript. We are indebted to Justin Moritz and Aliyah Moline-Freeman for their many hours of behavioral coding of initial group rapport. We additionally thank Dr. Scott Fraundorf and Manuel Garcia, for expert statistical consultation. We also thank Drs. Judith Hall and Danielle Blanch Hartigan for their guidance on training and coding procedures. We also acknowledge the students and staff at the Alcohol and Smoking Research Laboratory for their assistance.

Author Contributions

Both authors contributed to the study conception and design. Material preparation and collection of the original dataset was performed by Michael Sayette. Material preparation, coding, and analyses of newly generated data was performed by Madeline Goodwin. The first draft of the manuscript was written by Madeline Goodwin and edited and reviewed by Michael Sayette. All authors read and approved the final manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The data set used in the present study was generated with support from the National Institutes of Health (NIH) Grant (R01 AA015773) to Michael Sayette. NIH had no role, however, in the design of the study, or in the analysis and interpretation of data.

Ethics Approval

Approval for this study was obtained from the IRB of the University of Pittsburgh. The procedures used in this study adhere to the tenets of the Declaration of Helsinki.

Consent to Participate

Informed consent was obtained from all individual participants included in the study.

ORCID iD

Madeline E. Goodwin

Data Availability

The datasets analyzed during the current study are available from the corresponding author upon reasonable request.

Notes

Author Biographies

Madeline E. Goodwin, MS, is a clinical psychology graduate student at the University of Pittsburgh. Her research focuses on small group interactions and the social psychological processes that perpetuate alcohol use.

Michael A. Sayette, PhD, is a Distinguished Professor of Psychology at the University of Pittsburgh. His research interests include psychological theories of alcohol use and abuse, cigarette craving, and social psychological processes in addiction.

References

Ambady

Bernieri

F. J.

Richeson

J. A.

(2000). Toward a histology of social behavior: Judgmental accuracy from thin slices of the behavioral stream. In Zanna

M. P.

(Ed.), Advances in experimental social psychology (Vol. 32, pp. 201–271). Academic Press.

Ambady

Gray

H. M.

(2002). On being sad and mistaken: Mood effects on the accuracy of thin-slice judgments. Journal of Personality and Social Psychology, 83(4), 947–961. https://doi.org/10.1037/0022-3514.83.4.947

Ambady

Rosenthal

(1992). Thin slices of expressive behavior as predictors of interpersonal consequences: A meta-analysis. Psychological Bulletin, 111(2), 256–274. https://doi.org/10.1037/0033-2909.111.2.256

Baker

Z. G.

Watlington

E. M.

Knee

C. R.

(2020). The role of rapport in satisfying one’s basic psychological needs. Motivation and Emotion, 44(2), 329–343. https://doi.org/10.1007/s11031-020-09819-5

Bates

Mächler

Bolker

Walker

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01

Bernieri

Gillis

J. S.

(1995). The judgment of rapport: A cross-cultural comparison between Americans and Greeks. Journal of Nonverbal Behavior, 19(2), 115–130. https://doi.org/10.1007/bf02173170

Bernieri

F. J.

(1988). Coordinated movement in human interaction: Synchrony, posture similarity, and rapport (Publication No. 8901540) [Doctoral dissertation, Harvard University]. ProQuest Dissertations Publishing.

Bernieri

F. J.

(2005). The sourcebook of nonverbal measures: Going beyond words. In Manusov

(Ed.), The expression of rapport (pp. 347–359). Lawrence Erlbaum Associates Publishers.

Bernieri

F. J.

Davis

J. M.

Rosenthal

Knee

C. R.

(1994). Interactional synchrony and rapport: Measuring synchrony in displays devoid of sound and facial affect. Personality and Social Psychology Bulletin, 20(3), 303–311. https://doi.org/10.1177/0146167294203008

10.

Bernieri

F. J.

Gillis

J. S.

(2001). Judging rapport: Employing Brunswik’s lens model to study interpersonal sensitivity. In Hall

J. A.

Bernieri

F. J.

(Eds.), Interpersonal sensitivity: Theory and measurement (pp. 67–88). Lawrence Erlbaum Associates Publishers.

11.

Bernieri

F. J.

Gillis

J. S.

Davis

J. M.

Grahe

J. E.

(1996). Dyad rapport and the accuracy of its judgment across situations: A lens model analysis. Journal of Personality and Social Psychology, 71(1), 110–129. https://doi.org/10.1037/0022-3514.71.1.110

12.

Blanch-Hartigan

Ruben

M. A.

Hall

J. A.

Schmid Mast

(2018). Measuring nonverbal behavior in clinical interactions: A pragmatic guide. Patient Education and Counseling, 101(12), 2209–2218. https://doi.org/10.1016/j.pec.2018.08.013

13.

Bollen

K. A.

Hoyle

R. H.

(1990). Perceived cohesion: A conceptual and empirical examination. Social Forces, 69(2), 479–504. https://doi.org/10.2307/2579670

14.

Brown

W. M.

Palameta

Moore

(2003). Are there nonverbal cues to commitment? An exploratory study using the zero-acquaintance video presentation paradigm. Evolutionary Psychology, 1, 42–69. https://doi.org/10.1177/147470490300100104

15.

Cheong

J. H.

Molani

Sadhukha

Chang

L. J.

(2020). Synchronized affect in shared experiences strengthens social connection. PsyArXiv. https://doi.org/10.31234/osf.io/bd9wn

16.

Chin

W. W.

Salisbury

W. D.

Pearson

A. W.

Stollak

M. J.

(1999). Perceived cohesion in small groups: Adapting and testing the perceived cohesion scale in a small-group setting. Small Group Research, 30(6), 751–766. https://doi.org/10.1177/104649649903000605

17.

Creswell

K. G.

Sayette

M. A.

Manuck

S. B.

Ferrell

R. E.

Hill

S. Y.

Dimoff

J. D.

(2012). DRD4 polymorphism moderates the effect of alcohol consumption on social bonding. PLoS One, 7(2), e28914. https://doi.org/10.1371/journal.pone.0028914 Article e28914.

18.

Cribari-Neto

Zeileis

(2010). Beta regression inR. Journal of Statistical Software, 34(2), 1–24. https://doi.org/10.18637/jss.v034.i02

19.

Ekman

(1989). The argument and evidence about universals in facial expressions of emotion. In Wagner

Manstead

(Eds.), Handbook of social psychophysiology (pp. 143–164). John Wiley & Sons.

20.

Ekman

Davidson

R. J.

Friesen

W. V.

(1990). The Duchenne smile: Emotional expression and brain physiology: II. Journal of Personality and Social Psychology, 58(2), 342–353. citation1990/02/01 00:00Ekman P

21.

Ekman

Friesen

W. V.

(1969). The repertoire of nonverbal behavior: Categories, origins, usage, and coding. Semiotica, 1(1), 49–98. https://doi.org/10.1515/semi.1969.1.1.49

22.

Ekman

Friesen

W. V.

Hager

J. C.

(2002). Facial action coding system on CD-ROM. Network Information Research.

23.

Evans

N. J.

Jarvis

P. A.

(1986). The group attitude scale: A measure of attraction to group. Small Group Behavior, 17(2), 203–216. https://doi.org/10.1177/104649648601700205

24.

Fairbairn

C. E.

Sayette

M. A.

Levine

J. M.

Cohn

J. F.

Creswell

K. G.

(2013). The effects of alcohol on the emotional displays of whites in interracial groups. Emotion, 13(3), 468–477. https://doi.org/10.1037/a0030934

25.

Frank

M. G.

Ekman

Friesen

W. V.

(1993). Behavioral markers and recognizability of the smile of enjoyment. Journal of Personality and Social Psychology, 64(1), 83–93. https://doi.org/10.1037//0022-3514.64.1.83

26.

Frank

M. G.

Solbu

(2020). Social intelligence and nonverbal communication. In Sternberg

R. J.

Kostić

(Eds.), Nonverbal communication: Evolution and today (pp. 119–162). Palgrave Macmillan.

27.

Gignac

G. E.

Szodorai

E. T.

(2016). Effect size guidelines for individual differences researchers. Personality and Individual Differences, 102, 74–78. https://doi.org/10.1016/j.paid.2016.06.069

28.

Gillis

J. S.

Bernieri

F. J.

Wooten

(1995). The effects of stimulus medium and feedback on the judgment of rapport. Organizational Behavior and Human Decision Processes, 63(1), 33–45. https://doi.org/10.1006/obhd.1995.1059

29.

Goldenberg

Garcia

Halperin

Gross

J. J.

(2020). Collective emotions. Current Directions in Psychological Science, 29(2), 154–160. https://doi.org/10.1177/0963721420901574

30.

Gonzaga

G. C.

Keltner

Londahl

E. A.

Smith

M. D.

(2001). Love and the commitment problem in romantic relations and friendship. Journal of Personality and Social Psychology, 81(2), 247–262. https://doi.org/10.1037//0022-3514.81.2.247

31.

Grahe

J. E.

Bernieri

F. J.

(1999). The importance of nonverbal cues in judging rapport. Journal of Nonverbal Behavior, 23(4), 253–269. https://doi.org/10.1023/a:1021698725361

32.

Hall

J. A.

Roter

D. L.

Blanch

D. C.

Frankel

R. M.

(2009). Observer-rated rapport in interactions between medical students and standardized patients. Patient Education and Counseling, 76(3), 323–327. https://doi.org/10.1016/j.pec.2009.05.009

33.

Human

L. J.

Sandstrom

G. M.

Biesanz

J. C.

Dunn

E. W.

(2013). Accurate first impressions leave a lasting impression: The long-term effects of distinctive self-other agreement on relationship development. Social Psychological and Personality Science, 4(4), 395–402. https://doi.org/10.1177/1948550612463735

34.

Ickes

Gonzalez

(1994). Social” cognition and social cognition: From the subjective to the intersubjective. Small Group Research, 25(2), 294–315. https://doi.org/10.1177/1046496494252008

35.

Kiernan

(2018). Insights into using the GLIMMIX procedure to model categorical outcomes with random effects. SAS Institute. https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2018/2179-2018.pdf

36.

Kirchner

T. R.

Sayette

M. A.

Cohn

J. F.

Moreland

R. L.

Levine

J. M.

(2006). Effects of alcohol on group formation among male social drinkers. Journal of Studies on Alcohol, 67(5), 785–793. https://doi.org/10.15288/jsa.2006.67.785

37.

Koo

T. K.

M. Y.

(2016). A guideline of selecting and reporting intraclass correlation coefficients for reliability research. Journal of Chiropractic Medicine, 15(2), 155–163. https://doi.org/10.1016/j.jcm.2016.02.012

38.

Leary

M. R.

Kowalski

R. M.

(1995). Social anxiety. Guilford Press.

39.

Levine

J. M.

Moreland

R. L.

(1998). Small groups. In Gilbert

D. T.

Fiske

S. T.

Lindzey

(Eds.), The handbook of social psychology (Vol. 2, 4th ed., pp. 415–469). McGraw-Hill.

40.

Mehu

Grammer

Dunbar

R. I. M.

(2007). Smiles when sharing. Evolution and Human Behavior, 28(6), 415–422. https://doi.org/10.1016/j.evolhumbehav.2007.05.010

41.

Milmoe

Rosenthal

Blane

H. T.

Chafetz

M. E.

Wolf

(1967). The doctor's voice: Postdictor of successful referral of alcoholic patients. Journal of Abnormal Psychology, 72(1), 78–84. https://doi.org/10.1037/h0024219

42.

Park

R. E.

Burgess

E. W.

(1921). Introduction to the science of sociology. University of Chicago Press.

43.

SAS Institute. (2020). SAS software (Version 9.4). Computer software.

44.

Saville

Balas

(2014). Detecting personal familiarity depends on static frames in “thin slices” of behavior. Psychonomic Bulletin & Review, 21(6), 1537–1543. https://doi.org/10.3758/s13423-014-0629-y

45.

Sayette

M. A.

Creswell

K. G.

Dimoff

J. D.

Fairbairn

C. E.

Cohn

J. F.

Heckman

B. W.

Kirchner

T. R.

Levine

J. M.

Moreland

R. L.

(2012). Alcohol and group formation: A multimodal investigation of the effects of alcohol on emotion and social bonding. Psychological Science, 23(8), 869–878. https://doi.org/10.1177/0956797611435134

46.

Slepian

M. L.

Bogart

K. R.

Ambady

(2014). Thin-slice judgments in the clinical context. Annual Review of Clinical Psychology, 10(1), 131–153. https://doi.org/10.1146/annurev-clinpsy-090413-123522

47.

Smithson

Verkuilen

(2006). A better lemon squeezer? Maximum-likelihood regression with beta-distributed dependent variables. Psychological Methods, 11(1), 54–71. https://doi.org/10.1037/1082-989X.11.1.54

48.

The R Foundation. (n.d). The R project for statistical computing. https://www.R-project.org/.

49.

Tickle-Degnen

Rosenthal

(1987). Group rapport and nonverbal behavior. In Hendrick

(Ed.), Group processes and intergroup relations (pp. 113–136). SAGE Publications.

50.

Tickle-Degnen

Rosenthal

(1990). The nature of rapport and its nonverbal correlates. Psychological Inquiry, 1(4), 285–293. https://doi.org/10.1207/s15327965pli0104_1

51.

Verkuilen

Smithson

(2012). Mixed and mixture regression models for continuous bounded responses using the beta distribution. Journal of Educational and Behavioral Statistics, 37(1), 82–113. https://doi.org/10.3102/1076998610396895

52.

Willis

Todorov

(2006). First impressions: Making up your mind after a 100-ms exposure to a face. Psychological Science, 17(7), 592–598. https://doi.org/10.1111/j.1467-9280.2006.01750.x

53.

Wilson

J. H.

Ryan

R. G.

Pugh

J. L.

(2010). Professor–student rapport scale predicts student outcomes. Teaching of Psychology, 37(4), 246–251. https://doi.org/10.1080/00986283.2010.510976

54.

X. S.

Samtani

M. N.

Dunne

Nandy

Vermeulen

De Ridder

(2013). Mixed-effects beta regression for modeling continuous bounded outcome scores using NONMEM when data are not on the boundaries. Journal of Pharmacokinetics and Pharmacodynamics, 40(4), 537–544. https://doi.org/10.1007/s10928-013-9318-0

55.

Zimprich

(2010). Modeling change in skewed variables using mixed beta regression models. Research in Human Development, 7(1), 9–26. https://doi.org/10.f1080/15427600903578136