Interpersonal Autonomic Physiology: A Systematic Review of the Literature

PS

PS is typically defined as an interdependence or association between physiological signals from two or more people. However, synchrony is a nonspecific construct, as its identification is dependent on the procedure used to test it. For example, coherence analysis assesses cyclical patterns in the frequency domain (e.g., Henning et al., 2001), whereas bivariate time-series analysis addresses linear relationships in the time domain (e.g., Levenson & Gottman, 1983). As different analyses of synchrony address different components of data, results can differ substantially. It is therefore important for analyses of PS to be well matched to the research question because different approaches can alter interpretations and implications of study results.

In addition to detecting the presence of synchrony, we identified six key parameters that have been used to further define PS: magnitude, sign, direction, lag, timing, and arousal.

Magnitude refers to the strength of synchrony, such as a regression or correlation coefficient. This typically represents the effect size of a given measure of synchrony, with higher magnitudes indicating signals have greater interdependence or association.

Sign, which is typically positive or negative, indicates that peoples’ arousal levels synchronously move in the same or opposite directions (e.g., a positive or negative correlation). This has been referred to as concordant and discordant (e.g., Di Mascio et al., 1955), as well as in-phase and anti-phase synchrony (e.g., Reed et al., 2013). Whereas positive synchrony indicates partners’ physiologies move in the same direction, negative synchrony can indicate moving toward or away from the state of a partner.

Direction refers to the predictability of one person’s physiology from another’s. Results from tests of predictability may indicate a unidirectional or multi-directional relationship. In a unidirectional relationship, some magnitude of one person’s physiology is predicted by another’s, but not vice versa. In a multi-directional relationship, some magnitude of one person’s physiology is predicted by another’s, and vice versa. For example, Ferrer and Helm (2013) used coupled differential equation models to show that different proportions of male and female partners’ respiration were simultaneously predictable from the other’s respiration.

Lag refers to a shift in temporal alignment of data and has been used to show that a pattern in one person is followed by a similar pattern in another (e.g., Feldman, Magori-Cohen, Galili, Singer, & Louzoun, 2011; Müller & Lindenberger, 2011). The difference in time alignment from one data set to another is the specific lag, which may indicate millisecond differences, or much longer time offsets. An example of this can be seen in Müller and Lindenberger (2011), who used Granger causality models—a method of testing whether one time series can predict another—to show that respiration rates (RRs) of choir singers could be predicted from the conductor’s. These results were interpreted as evidence that the conductor was leading the choir. Although lagged PS implies a unidirectional relationship, there is some evidence that the length of a lag (e.g., 1 s vs. 10 s time offset) can reflect psychosocial properties that are independent of direction and worthy of exploration (e.g., Feldman et al., 2011; Messina et al., 2013). For example, in a study of therapist–client dyads, Messina et al. (2013) found that the lag of significant cross-correlations between empathy and PS differed by therapist training. This was interpreted as evidence that therapists with greater training are better able to maintain empathy with clients.

The timing parameter refers to the length of time that a given interaction is assessed or observed. Most extant studies assessed PS for a somewhat arbitrarily defined period of time, such as the length of a condition. Thus, the parameters of synchrony are an aggregate of a given time period. This leads to notable differences in how results can be interpreted across studies, as aggregates of interpersonal interactions over an hour may be substantively different from aggregates of a few minutes. This issue becomes more complex when considering that multiple interpersonal physiological processes can operate at different time scales simultaneously. For instance, synchrony measured in short time scales (e.g., seconds or minutes) may reveal patterns related to momentary interactions, whereas long-term measures (e.g., hours or days) could show interactions that take longer to develop. Both are equally valid approaches, but results could be indicative of different types of processes. An example of this can be seen in Ferrer and Helm (2013), who compared couples’ interpersonal patterns of respiration over minutes, with couples’ affect over days. They found that in some conditions, couples’ patterns of PS in respiration over 3-min trials were the same as those found in daily affect, indicating that, in this case, both short- and long-term interactions were similar. The time scale assessed is therefore a key element of a research question. Another critical issue with timing is choosing an appropriate time scale when the temporal process is unclear. As longer time periods increase potential for relevant changes to be aggregated together, data should be assessed on a time scale that is theorized or known to be consistent with the process of interest. One way to manage this issue is to use time-varying models, which can assess when and how PS changes over time. Although few studies have incorporated time-varying models into IAP research (for exceptions, see Müller & Lindenberger, 2011; Quer, Daftari, & Rao, in press; Waters, West, & Mendes, 2014), this is an important consideration for future work.

Finally, the pattern of arousal can be assessed as a covariate of PS. At its simplest, mean arousal level can be used as a covariate of synchrony. This can be seen in Creaven, Skowron, Hughes, Howard, and Loken (2014), who found that higher maternal HR was associated with lower magnitudes of mother–child PS. More complex models of arousal patterns can also be tested as covariates or moderators of PS, such as assessing whether increasing PS correlates with decreasing arousal. Furthermore, arousal patterns themselves may be assessed as an interpersonal moderator. For instance, Butler and Randall (2013) termed one type of moderation stress buffering, defined as a period when an individual moderates the stress level of another, but the partners are not necessarily experiencing the same physiology (i.e., magnitude may equal zero). Such a pattern may lead to muted arousal responses and faster return to homeostatic states, whether or not PS occurs. For example, the presence of a relaxed partner may lead an individual to maintain a degree of calmness during a stressful event, although the partners’ arousal patterns are dissimilar. Whereas stress buffering specifically refers to interpersonal moderation of a stress response, a wide range of interpersonal physiological moderators can also be tested. Specific physiological states in one person may contribute to reduced or amplified arousal levels in others, or the likelihood that a given type of PS will develop. In addition, there may be interaction effects that contribute to the potential for moderation, such as someone with very high arousal levels being less likely to experience stress buffering or PS than someone with lower arousal. This general approach to IAP research is therefore not dependent on PS, yet has significant potential to reveal other types of interpersonal physiological interactions.

Another construct in the literature is asynchrony, used to describe a lack of observable PS (Reed et al., 2013). Although difficult to substantiate without the use of multiple models to test for PS, the concept of asynchrony is an important one, as it describes periods of dissimilarity between people (i.e., magnitude = 0). Asynchrony has been found to be predictive of specific relationship types (e.g., Reed et al., 2013), suggesting that identification of periods that lack PS can also be informative of an interaction. For example, periods of asynchrony may occur during an interaction if an individual is ignoring the state of a partner (e.g., Marci & Orr, 2006), or if one member does not correspond with others in a group.

Psychosocial terms

A number of psychosocial constructs identified in the literature refer to specific combinations of PS parameters. However, unique properties as well as inconsistent operational definitions of psychosocial terms can make direct comparisons with measures of PS difficult. Poorly defined terms can blur important distinctions between processes, and inconsistencies can hinder cross-study comparisons. The construct coregulation has been identified as a prime example of this issue (Butler & Randall, 2013). It has been defined as a bidirectional emotional relationship leading to emotional and physiological stability around an optimal state for both partners in a close relationship (Butler & Randall, 2013). However, whereas this definition specifically refers to a bidirectional interaction, others have conceptualized coregulation as including unidirectional interactions, such as a parent calming a child (e.g., Feldman, 2003). Unidirectional coregulation was observed in Field, Healy, and LeBlanc (1989), who found that infants’ HR followed their mothers’ at a short (<1 s) lag. Similarly, specifying the state (i.e., optimal) and relationship type (i.e., close) limits the context of coregulation, as the same physiological pattern could occur in other relationships or around a suboptimal state, such as coworkers maintaining a bidirectional relationship around a stressful state.

Due to terminological variations, it is important for IAP research to specifically and consistently define and differentiate between psychosocial terms (e.g., coregulation) and physiological measures (e.g., PS parameters) under study. An example of this level of terminological clarity can be seen in Butler (2011), who presents a set of psychosocial terms along with statistical components of synchrony (therein termed covariation) they represent. This approach helps clarify and distinguish between the measurable properties of a given physiological interaction and associated psychosocial terms.

Well-defined terms also help make two distinct research approaches possible. The most common is to test whether a given type or magnitude of PS is present. This is typically applied when the primary interest is to determine differences in PS across groups or conditions. For example, Guastello, Pincus, and Gunderson (2006) compared PS between groups engaged in different conversations, and found that competitive, cooperative, and control groups displayed similar magnitudes of PS. Alternatively, when the research question regards the conditions associated with a given type of PS, data can be searched to locate instances of specific parameters of synchrony. For example, using this approach, Helm, Sbarra, and Ferrer (2014) tested couples’ RSA for morphostatic coregulation as defined by Butler (2011; i.e., coregulation around a stable level) and found that it was present across different conversation tasks. Similarly, Stratford, Lal, and Meara (2012) searched data from therapy sessions for periods with the highest magnitude of positive PS, then assessed neurological activity during those periods.

Results by measure

We assessed results by specific measures of cardiac activity (Appendix B), EDA (Appendix C), respiratory activity (Appendix D), thermal changes (Appendix E), and indexes of multiple measures (Appendix F). Results are summarized in Appendices B to F, including the specific physiological measure used, the context in which synchrony was tested, whether findings were statistically significant, and summaries of relevant results. Each result was further explored according to six categories: general evidence of PS, differences in PS between groups, differences in PS between conditions within groups, moderators of PS (e.g., resting HR moderating PS levels), psychosocial correlates of PS (e.g., PS correlated with empathy), and specific typologies of PS (e.g., significant findings of positive or negative PS). We could not identify any patterns in findings by measure.

An example of the ambiguity of results can be seen in findings of PS through EDA. Whereas studies using most metrics have found significant PS in dyads (e.g., Chanel Kivikangas, & Ravaja, 2012, using skin conductance; Järvelä et al., 2013, using skin conductance response—the phasic component of skin conductance; Marci & Orr, 2006 using skin conductance level), others do not (e.g., Reed et al., 2013, using skin conductance). Cross-study comparisons of between-group differences lead to similar ambiguities, such as one result indicating significant PS in skin conductance between dyads of friends but not strangers (Shearn, Spellman, Straley, Meirick, & Stryker, 1999), whereas another found significant PS in the skin conductance of strangers (Silver & Parente, 2004). Results are also unclear when considering the relationship between psychosocial constructs and PS. For example, most studies comparing PS in EDA with empathy found significant correlations when using skin conductance level (e.g., Marci et al., 2007; Marci & Orr, 2006; Messina et al., 2013; in therapists–clients), although one found no association using skin conductance (Chatel-Goldman, Congedo, Jutten, & Schwartz, 2014, in couples), and another found both significant and non-significant results using multiple novel metrics of skin conductance responses (Robinson, Herman, & Kaplan, 1982, in therapists–clients).

Although no definitive patterns have yet emerged, relatively consistent findings of PS across physiological measures suggest that synchrony is a robust phenomenon identifiable through numerous methods. Studies such as Levenson and Gottman’s (1983), which used an index combining multiple physiological measures, may therefore be reasonable approaches for capturing a general autonomic pattern. Still, as different measures reflect distinct autonomic processes, and it is unknown whether different types or outcomes of PS are system-specific, it remains important to test each system independently.

Idiographic analyses

Both idiographic and nomothetic methodologies have been used in IAP research (see Appendix A for the methodology and analysis used in each study). At the idiographic level, one of the most commonly used approaches to measure the relationship between two variables is a Pearson correlation. For example, recent research has used correlations to assess PS between mothers and children (Ebisch et al., 2012), strangers (Silver & Parente, 2004), and teammates (Chanel et al., 2012). However, correlations are not well-suited for continuously measured physiology, because data typically show sequential dependency (i.e., autocorrelations) and non-stationarity over time (i.e., changing mean and variance). An example of issues using correlations to measure PS can be seen in Liu, Zhou, Palumbo, and Wang (in press). There, they show that results from correlations of both simulated data, and electrodermal data from couples, can show spurious results. Similarly, both autocorrelation and non-stationarity violate general linear model (GLM) assumptions of data independence and stationarity, so they need to be accounted for if applying GLM analyses.

Time-series analysis is one way to account for sequential dependency in repeated-measures data, and hence provides a more valid representation of the association between variables than Pearson correlation. It can be carried out in either the time or frequency domain. Time domain models used to capture PS typically come from the autoregressive-moving-average (ARMA) model family. Once dependency in the data is modeled and thus statistically removed, standard analyses based on the GLM can then be performed. Feldman et al. (2011) used these procedures with cross-correlations to assess PS in the interbeat intervals of mothers and children, and found significant levels of PS during face-to-face interactions. Frequency-based techniques (e.g., coherence, phase difference, directed coherence) decompose data into cyclical components, and typically test whether similar frequency bands occur across data sets. These analyses can quantify similarities in the cycles of multiple measures, their lead–lag relations, and directed influences between variables (Liu & Molenaar, in press). Another example of a frequency-based technique can be seen in Henning et al. (2001), who used weighted cross-coherence in skin conductance variability, HR, and respiration and found that PS predicted team errors.

Most time-series models, however, are based on the assumption of stationarity, meaning that statistical properties (i.e., means, covariance matrices) of data are constant over time. This implies that the interpersonal dynamics remain constant during an interaction, which may not be the case. As models aggregate data, if there are multiple processes within a period, results may not accurately represent any of the individual patterns. To model time-varying effects in idiographic data, new methods are emerging, including wavelet analysis, threshold autoregressive models (Hamaker, Zhang, & van der Maas, 2009), and regime-switching models (Yang & Chow, 2010). These methods have rarely been applied to IAP data. In one example, Quer et al. (in press) used wavelet analysis to show that the dynamics of PS between groups of people meditating changed over time. Moving window techniques have also been adapted to detect dynamics in PS, most commonly as part of an analysis known as a “linkage index” (e.g., Marci et al., 2007). Here, an analysis is run on a segment of data, the window is stepped forward in time, and the analysis rerun on the next segment. Continuing this procedure through the data yields a time series of synchrony scores, illustrating how synchrony dynamics change over time. Another example of this can be seen in Stratford et al. (2012), where they calculated “linkage indexes” in windows to determine when synchrony in the skin conductance of therapists and clients peaked during sessions. However, as issues of stationarity and autocorrelation apply to any given window, the potential for Type I error is compounded each time the analysis is calculated on a window.

Another approach that has been used to analyze idiographic data is nonlinear modeling. Nonlinear modeling is a broad term that includes analyses that can detect patterns unobservable through linear regression. One example of nonlinear modeling used in interpersonal research is cross-recurrence analysis. This phase space analysis is based on graphing multiple time series, where one axis represents one person and the other axis represents the other person. This plotting method shows periods when both individuals’ states are simultaneously the same. Patterns in the plot can then be used to derive measures of the predictability, stability, complexity, and smoothness of the dynamic, bivariate system (Konvalinka et al., 2011). Strang, Funke, Russell, Dukes, and Middendorf (2014) and Konvalinka et al. (2011) used cross-recurrence analysis to detect PS in interbeat intervals. Konvalinka et al. (2011) found that PS developed between fire walkers and related spectators during ritual fire walking, but not between fire walkers and unrelated spectators. Strang et al. (2014) found that PS between previously unacquainted teammates was no greater than with randomly paired data from different team members. In another nonlinear approach, Guastello et al. (2006) compared results from linear and nonlinear regressions of skin conductance between participants during competitive, cooperative, or neutral discussions. Their results indicated that linear analyses were outperformed by nonlinear tests, which indicated significant PS in all dyads, but no differences across conditions.

Nomothetic analyses

Again, nomothetic analyses involve either aggregating individual data or conducting analyses in a multilevel modeling framework. Aggregation of individual data provides a simple summary of overall trends at the group level, but ignores individual differences. In contrast, multilevel models allow researchers to simultaneously assess group-level tendencies while accounting for individual differences. In addition, multilevel models are very flexible in terms of model specification. In some studies, multilevel growth curve models are used to represent response variables with functions of time (e.g., a linear curve), and PS is represented by covariation in function parameters (e.g., random slopes) between partners. In other studies, PS is represented by covariation between partners in the fluctuations around a deterministic trend (e.g., Reed et al., 2013). In the latter case, an individual-specific index of PS can be estimated, which makes it possible to examine factors or covariates that explain individual differences in strength of synchrony. However, multilevel models typically require a large sample size and normal distributions of parameters. Heterogeneity and small sample sizes common in physiological studies complicate such analyses. A new nomothetic approach applied to PS, dynamical correlation, may be better suited to these data issues (Liu et al., in press). The technique is appropriate for small sample sizes, heterogeneous data, and missing data, and is unaffected by autocorrelation or non-stationarity. In addition, no modeling is required, making it a simpler approach to implement. More importantly, it uses a population-centering step that accounts for group-based similarities, so resulting measures of synchrony are due to dyad-specific relationships. Still, the analysis returns a single descriptive statistic of synchrony for a group, which may not be indicative of mechanisms or processes involved, and does not estimate an individual-specific index of PS.

Dynamic system modeling

A unique approach for studying synchrony is dynamic systems modeling. A dynamic system is often described by a set of equations that expresses how the state of a system changes as a function of its previous state, either in a linear or nonlinear fashion. Dynamic system modeling can be carried out as an idiographic approach by using a set of equations (typically differential equations) to explain variation in each participant’s data. Alternatively, it can be carried out at the group level by simply aggregating data across participants or using a multilevel model. Helm, Sbarra, and Ferrer (2012) and Ferrer and Helm (2013) are credited with bringing this approach to the study of PS in their work assessing couples. A significant advantage of this technique is that statistical parameters of synchrony including magnitude (i.e., the strength of synchrony), sign (i.e., positive or negative), direction (i.e., the degree to which each person can be predicted by the other), and lag (i.e., time differences between peoples responses) can be tested simultaneously. This makes dynamic systems models uniquely adaptable to the study of PS.

Null hypothesis testing

Another critical consideration, regardless of the statistical analysis used, is how to validate the detection of PS because synchrony may appear spuriously due to a variety of data conditions (e.g., autocorrelation, linear trends, random chance). For example, due to spurious co-occurrences of unrelated respiratory kinematics, McFarland (2001) observed high correlations in randomly paired RRs from people having conversations. Furthermore, the probability of chance findings can vary across contexts. As data structures can differ by condition (e.g., more or less likely to have a trend over time), the probability of spurious synchrony being detected in one condition may differ from the probability of spurious synchrony detected in another. For example, Codrons, Bernardi, Vandoni, and Bernardi (2014) randomly paired data from participants who completed a baseline and an arm-swinging task in isolation. Using well-suited analyses (Generalized Partial Directed Coherence), they found that PS detected in RRs between random individuals during the arm-swinging task was significantly greater than during baseline. Similarly, McFarland (2001) found different distributions of correlations from randomly paired data collected from dyads engaged in spontaneous versus scripted conversations. Due to the inconsistent probability of chance findings, standard null hypothesis testing may not be sufficient to determine whether findings of PS are statistically significantly different from chance. Therefore, it is important to determine the probability of spurious findings in each context to validate results. A number of studies have dealt with this issue by generating a null hypothesis based on data randomly paired from individuals who participated in the same task with a different partner, then testing whether PS in random dyads is significantly different from actual dyads (e.g., Helm et al., 2012; Marci et al., 2007; Reed et al., 2013). Additional methods used to determine null hypotheses have also been used, including comparing PS from real dyads (or groups) with simulated data, and with data from individuals who participated in the same task while alone (Strang et al., 2014). These customized methods of null hypothesis testing can help determine whether findings of PS are due to data structures, conditional demands, common interpersonal processes, or interpersonal interdependence, although methodological designs are still needed to test causality (see Appendix A for studies that used customized null hypotheses).

A related method designed to isolate interpersonal interdependence uses a baseline condition wherein participants do not interact, and tests whether PS is significantly greater when they do interact (e.g., Helm et al., 2012; Levenson & Gottman, 1983). Significant increases in synchrony from baseline then suggest that interpersonal dynamics, rather than dyadic or context-specific conditions, generated the increase in PS. Still, comparisons with randomized dyads are needed to determine the probability of chance results. Helm et al. (2012) used this approach, comparing periods when participants interacted to periods when they did not, as well as to data from unmatched dyads. They found that couples’ baseline PS was significantly greater than in random data, and PS was significantly greater during interaction periods than baseline. Overall, these types of strategies are needed for null hypothesis testing, without which it may be unclear whether findings of PS are significantly different from chance.

Therapist–client

Research on IAP began more than half a century ago, when a series of studies found significant positive and negative correlations in the EDA and HR of therapists and clients during therapy, interpreted as evidence of therapeutic rapport and empathy (Coleman, Greenblatt, & Solomon, 1956; Di Mascio, Boyd, & Greenblatt, 1957; Di Mascio et al., 1955). These studies found that therapist notes from sessions with high positive correlations in HR had fewer references to being distracted from therapy than sessions with low correlations in HR (Coleman et al., 1956). In addition, the authors noted that all clients showed reduced HR with one particular therapist, potentially an early example of stress buffering. Contemporary research utilizing similar methods supports these results, including significant associations between PS and empathy, as well as attention (e.g., Marci et al., 2007; Marci & Orr, 2006; Robinson et al., 1982).

Couples

Much of the existing interpersonal physiological research has focused on couples (for a review, see Timmons, Margolin, & Saxbe, 2015). In their seminal work in this area, Levenson and Gottman (1983) used a bivariate time-series analysis of multiple physiological measures to show that couples’ PS during arguments could account for 60% of variance in marital satisfaction. As they found no statistically significant results when couples were discussing neutral topics, they concluded that PS only developed during negative interactions. They postulated that dissatisfied couples could not disengage from the arousal of a conflict, whereas satisfied couples were able to “step back” and listen. Aspects of Levenson and Gottman’s (1983) findings have been supported a number of times, including another early study that found PS in couples’ skin conductance and HR during conflict conversations (Thomsen & Gilbert, 1998). Although differences across couples were noted, overall, husbands’ physiology was a better predictor of wives’ than vice versa, suggesting that husbands tended to lead the interactions. A more recent study found that marital conflict was positively correlated with PS in RSA (Gates, Gatzke-Kopp, Sandsten, & Blandon, 2015). Similarly, using longitudinal multilevel dyadic models, Reed et al. (2013) found that negative interactions between couples (e.g., demand/withdraw behaviors) moderated PS in blood pressure. The presence of negative interactions coincided with positive PS, and their absence with negative PS. The authors concluded that negative PS results from turn taking during dialog and could be a key component in any conversation. Although Reed et al.’s (2013) conclusions may be correct, contradictory findings suggest that positive PS also develops between couples in neutral and positive contexts. For example, Helm et al. (2014) found significantly greater magnitudes of positive PS in couples with higher quality relationships across positive, neutral, and negative conversations, whether partners were listening or talking. However, as Helm et al. (2014) used cross-lagged panel models to test for PS in RSA, differences from Reed et al.’s (2013) findings may be due to the use of different physiological measures and statistical methods.

A recent series of studies assessed PS between couples who were still and quiet, and either blindfolded, gazing at each other, or attempting to silently synchronize with each other (Ferrer & Helm, 2013; Helm et al., 2012, 2014; McAssey et al., 2013). Multiple novel analytical approaches were used in these studies, although the greatest advancements came from dynamic systems analyses capable of detecting multiple parameters of PS within a trial. Results from these studies suggest that sitting next to a partner while blindfolded is sufficient for PS in HR to develop (Ferrer & Helm, 2013), and that instructions can influence patterns of PS to change when behaviors remain constant. Using similar procedures, Liu et al. (in press) used dynamical correlations, a well-suited nomothetic analysis, to show that PS did not develop between couples seated back-to-back, but did when they turned and sat quietly face-to-face. Results were interpreted as indicating that PS resulted from interpersonal processes, as during the back-to-back phase, couples were exposed to the same metabolic demands and environmental conditions, but displayed significantly less PS than when they turned and faced each other.

Taken together, findings from couples research suggest that PS is not dependent on environmental or metabolic demands, that PS may be influenced by intention, and that dynamic interpersonal physiological patterns change according to context. Results from other populations largely support findings from couples, although more work is needed before generalizations can be made.

Parent–child

Parent–child dyads have received the most research attention to date, and relatively consistent findings of PS have been documented (see Hill-Soderlund et al., 2008; Moore, 2009, for exceptions). Although there is some evidence that PS between parents and children is dependent on behavioral synchrony (Feldman et al., 2011), other results indicate that PS can occur in the absence of behavioral synchrony (Suveg, Shaffer, & Davis, 2016; Woltering, Lishak, Elliott, Ferraro, & Granic, 2015). Such discrepancies between behavioral synchrony and PS may be due to context, as one study found that when PS was higher between mothers and children, behavioral synchrony and self-regulation were lower in high-risk children (Suveg et al., 2016). These results suggest that PS between parents and children in a negative context may disrupt behavioral synchrony, as well as the child’s self-regulation.

Overall, PS between mothers and children has been found to develop across groups and conditions using a variety of measures and analyses. Significant magnitudes of PS appear equally across many categorizations of mother–child dyads, including depressed and non-depressed mothers (Field et al., 1989), child maltreating and non-maltreating mothers (Creaven et al., 2014), mothers and their child with typical and atypical self-regulation (Woltering et al., 2015), and both high- and low-conflict mother–daughter relationships (Manini et al., 2013) and high- and low-risk families (Suveg et al., 2016). However, there is some evidence that PS is affected in children with autism, as magnitudes of PS between parents and their children with autism were negatively correlated with severity of the child’s diagnosis (Baker et al., 2015). This groundbreaking finding suggests that PS could be a useful diagnostic and clinical tool when supporting individuals with autism spectrum disorders, although systematic research is still needed.

Evidence from multiple studies suggests that PS magnitudes increase when mothers are under stress, such that individual physiological profiles moderate the development of PS (Ebisch et al., 2012; Ghafar-Tabrizi, 2008; Manini et al., 2013; Waters et al., 2014). In addition, children have been found to be more likely to avoid others when mothers undergo a stressful-negative event, suggesting that infant behavior is influenced by mothers’ psychophysiological state (Waters et al., 2014). Mothers’ influence on infants was further supported in a study by Van Puyvelde et al. (2015), which showed that by intentionally adjusting themselves, mothers could change their infants’ physiology. In this study, mothers paced their breathing while holding their infants at different ages. While controlling for motor and respiratory confounds, results indicated that mother–infant RSA synchronized across different breathing paces and infant ages until infants were 2 months old, but RSA did not synchronize when infants reached 3 months of age. This is the first study to investigate a potential mother–infant physiological relationship independent of a social interaction, and suggests that early infants rely on their mother for self-regulation. Findings by Ham and Tronick (2009) suggest that mothers capitalize on this autonomic contagion to regulate the arousal levels of their children, as they found that mothers calm themselves in an effort to calm their infants.

Although emphasis has been placed on directed PS from mother to child, synchrony has also been shown to be influenced by the child’s actions. One recent study found that positive PS occurred in the absence of child behavior problems, and negative PS occurred when behavior problems were present (Lunkenheimer et al., 2015). Similarly, Manini et al. (2013) found that the direction of influence between mothers and their daughters switched across discussions. These results suggest that during parent–child interactions, the type and direction of PS are contextually dependent.

Teammates

Assessment of PS in teams has gained significant interest in recent years (for a focused review, see Ekman et al., 2012). Henning et al. (2001) were the first to explore PS between team members through a series of studies examining teamwork during video games. Using cross-coherence and cross-correlations in respiration, HRV, and skin conductance, they found that PS was a significant predictor of team performance. Although their findings have not been supported consistently (e.g., Walker, Muth, Switzer, & Rosopa, 2013), recent work has also shown a significant correlation between team performance and PS (Montague, Xu, & Chiou, 2014). Henning et al.’s (2001) results also indicated that PS was not associated with team coordination, suggesting that PS is not dependent on coordinated behaviors. In a follow-up study, Henning and Korbelak (2005) used cross-correlations of teammates’ HRV to show that PS is predictive of future team performance during video game play.

Synchrony has also been found to correlate with psychosocial aspects of teamwork. Using a variety of measures and analyses, teammates’ social interactions (Chanel et al., 2012), as well as empathy and understanding between players (Järvelä et al., 2013), were found to be significantly associated with the magnitude of PS. Interestingly, one study found coherence in high-frequency HRV to be significantly higher during competitive versus cooperative games, as well as evidence that multiple metrics of PS correlated with empathy and positive affect (Chanel et al., 2012). The authors concluded that PS is most likely due to the intensity of an interaction rather than the context under which it occurs (Chanel et al., 2012).

In-vivo teamwork has also been assessed, although results have been inconsistent. Henning et al.’s (2009) study of teamwork between members of a research group observed that cross-correlations in HRV negatively predicted team ratings of their ability to work together. Conversely, Elkins et al. (2009), using correlations between teammates’ RSAs, found that better team performance was associated with significantly higher positive PS during military training tasks.

Strang et al. (2014) assessed PS in interbeat intervals of teammates playing a video game using analyses including cross-correlation and cross-recurrence quantification. In addition, they tested whether PS was significantly greater than simulated data, data paired from individuals who played alone, and data randomly paired from other teams. Results indicated that teammates’ PS was not significantly greater than data randomly paired from other teams, suggesting that PS was due to conditional similarities, rather than interpersonal dynamics.

Other relationships

This subgroup domain is a general categorization of participant relationships that do not fall under the aforementioned group types. The first such study assessed PS in casual relationships (Kaplan et al., 1963). They analyzed conversations of medical students in a group setting, and found significantly greater magnitudes of PS in skin conductance responses when dyads reported strong affective ties (i.e., liked or disliked each other), than when they reported a neutral relationship. Recent work found that when status is manipulated in dyads of strangers, high-status partners lead low-status partners’ pre-ejection period (a measure of SNS) during mock negotiations (Kraus & Mendes, 2014). PS may therefore be more dependent on social context rather than the type of relationship or the given affect. This could explain why studies comparing specific relationship types have been contradictory. Whereas some indicate PS only occurs between friends (McFarland, 2001, using respiration; Shearn et al., 1999, using skin conductance and facial blushing) or family members (Konvalinka et al., 2011, using HR), others found PS between strangers (Silver & Parente, 2004, using skin conductance), and still others found no difference in PS between friends and acquaintances (Field, 1992, using HR).

In a study comparing friends during different conversations, Guastello and colleagues (2006) compared linear and nonlinear regression models capable of detecting PS in skin conductance. PS was detected during high-conflict and neutral conversation topics, with no statistically significant difference between conditions observed. Nonlinear regressions detected significantly greater magnitudes of PS between partners than the linear regressions, prompting the authors to conclude that physiological interdependencies are multilevel processes with both linear and nonlinear characteristics. Mitkidis, McGraw, Roepstorff, and Wallot (2015) used cross-recurrence quantification analysis, but tested whether PS in dyads’ HR during an economic trust game (the public goods game) contributed to the outcome. They found that PS was positively associated with expectations for financial returns from partners, leading the authors to conclude that PS held potential as a proxy measure of trust.

Müller and Lindenberger (2011) applied a series of advanced statistical procedures to assess group interactions in a conductor-led choir, including wavelet analysis, multivariate autoregressive (MVAR) models, network analysis, and Granger causality. Results showed that the magnitude of PS in HRV and RRs was greater when singing in unison compared with singing in parts or rest periods. When the choir was singing in parts, network analyses detected subgroups with greater PS that corresponded to the musical parts subgroups sang. In addition, Granger causality models indicated that changes in the conductor’s HRV and RR predicated similar changes in choir members. These general findings were supported in later work that found PS in singers’ HRV when they sang, but not when they hummed or during a baseline (Vickhoff et al., 2013). Conversely, a study on musicians was unable to detect synchrony in the HR of dyads drumming together, although their analysis was more descriptive than quantitative (Morgan, Gunes, & Bryan-Kinns, 2015).

Expanding on Müller and Lindenberger’s (2011) approach, Quer et al. (in press) used a novel time-varying wavelet analyses to assess group-level PS in people meditating. They found that PS could be detected across measures of HRV and respiration, and that the magnitude of the groups’ synchrony changed over time. Their analysis offers a time-frequency strategy to test changes in the magnitude of synchrony in a group across a range of frequencies, which may prove to be an important advancement.

Two extant studies have directly assessed the interaction between synchronized behavior and PS. Codrons et al. (2014) had participants swing their arms while listening to music, a metronome, or in silence. Half the participants completed the exercise in isolation, and the other half completed the procedures in groups of 10. Results indicated significant PS in RR, but not heart period, occurred in groups during the baseline and music conditions. Interestingly, synchrony in arm movement did not coincide with PS, suggesting that behavioral synchrony and PS are independent processes. Noy, Levit-Binun, and Golland (2015) failed to find differences between behavioral synchrony and PS. They explored whether joint movement was associated with PS by having participants play the “mirror game.” Experienced participants put their hands together and moved in synchrony. Results indicated that PS in HR was significantly correlated with synchronized movement and subjectively reported togetherness. However, as PS was also correlated with high HR, the authors cautioned that PS could be due to metabolic demands of movements, rather than interpersonal processes.

Two studies have also used PS to test the influence of a third variable, rather than an interpersonal interaction. Marci (2006) assessed PS in groups watching television commercials to determine whether context influenced group response to advertisements. He used an index of physiological measures similar to Levenson and Gottman’s (1983) to test for PS, and incorporated arousal level in an attempt to capture “emotional engagement.” Using this strategy, he tested whether commercials would produce different results in different contexts (i.e., during shows with high and low ratings). Although their results are difficult to evaluate due to an undisclosed proprietary analysis, this appears to be the first study that aimed to use PS to test group engagement, rather than an interpersonal interaction. More recently, Bachrach, Fontbonne, Joufflineau, and Ulloa (2015) found that audience members’ breathing rates synchronized with dancers during a recital. Synchrony was positively correlated with audience members’ self-reported attention to their own and the dancers’ breathing, but not with their appreciation for the performance. Together, these two studies exemplify the potential to use PS to explore variables beyond interpersonal relationships.

Terminology

Our systematic review of the IAP literature identified terminological variation across the field, including inconsistent operational definitions. This issue is more than mere semantics, as the methodological and statistical approaches used in a study are dependent on the definition of the phenomenon to be identified. A number of authors have highlighted this issue and made attempts to resolve terminological ambiguities by operationally defining specific types of physiological relationships and their inherent statistical properties, as well as their relationships with psychophysiological constructs (e.g., Butler, 2011; Butler & Randall, 2013; Field, 2012; Helm et al., 2014). Quantitatively assessable operational definitions are needed for valid cross-study comparisons.

In line with the need for more specific terminology, as well as growing research interest in interpersonal processes, we propose the use of the term physiological entanglement to specify a measureable interdependence between peoples’ physiological activities. The term, adopted from physics, denotes periods when particles are best understood as a system, rather than a collection of individual units (Horodecki, Horodecki, Horodecki, & Horodecki, 2009). This definition is directly transferable to psychology, meaning that the processes of each individual are better understood when the processes of all individuals are considered together.

To test for physiological entanglement, analyses are needed to isolate PS due to interpersonal interactions. This has been done through methodological checks and statistical modeling (e.g., dynamic systems analysis, Granger causality). The proportion of a dyad’s PS due to entanglement has been methodologically tested by comparing a dyad’s PS when they are interacting with each other to their PS under the same conditions when they are not interacting with each other. For example, Ferrer and Helm (2013) compared the PS in dyads who were quietly seated together while blindfolded with their PS in the same condition without blindfolds. If a dyad’s PS when they were interacting is significantly greater than when they were not interacting, the difference can be considered evidence of entanglement. In addition to these procedures, the authors used dynamic systems analyses designed to determine the degree to which a dyad is interdependent. For instance, their dynamic systems model tests the degree to which changes in each partner are predictable from the other (see Ferrer & Helm, 2013, for details). More complex models can also be used to test for entanglement, such as Muller and Lindenberger’s (2011) use of wavelet analysis paired with network analysis and Granger causality models, although methodological checks are still necessary.

Physiological variables

Although no clear patterns of results by ANS measure were identified in this review, different physiological measures have led to differential findings. For example, positive PS was found to occur in both RR and HR under some conditions, whereas in another condition, HR, but not RR, was synchronized (Ferrer & Helm, 2013; Helm et al., 2012). Such findings suggest that each measure reflects unique processes. Collecting multiple autonomic measures can lead to greater specificity of processes related to each physiological system. Furthermore, cross-measure analyses may be indicative of interactions between systems, such as one individual’s PNS contributing to variance in another’s SNS. An example of successful use of multiple measures can be seen in Creaven et al. (2014). They found significant magnitudes of PS between mothers’ HR and their children’s RSA that differed by group assignment. In addition to specific autonomic measures used, the collection, analysis, and interpretation of physiological data require a substantial knowledge base. The interested reader is referred to other resources (e.g., Cacioppo et al., 2007; Goodwin, 2012) for more details.

Methodological approach