Endogenous Oxytocin and Human Social Interactions: A Systematic Review and Meta-Analysis

Abstract

While there has been an increase in studies investigating the relationship between endogenous oxytocin (OXT) concentrations and human social interactions over the past decades, these studies still seem far from converging, both in methodological terms and in terms of their results. This systematic review and meta-analysis were aimed at a comprehensive evaluation and synthesis of empirical evidence on the relationship between endogenous OXT concentrations and human social interactions by reviewing studies published between 1970 and July 2020 and addressing various related methodological and analytical limitations. Sixty-three studies were included in the qualitative synthesis, and results from 51 studies were pooled in a meta-analysis (n = 3,741 participants). The results indicated that social interaction did not lead to an expected hormonal response in causal designs, either in a pre-post design (g = 0.079) or when comparing experimental conditions with and without social interaction (g = 0.256). However, in correlational designs, the overall mean effect size of the correlations between indicators of social interaction and OXT concentrations was significantly different from zero (z = 0.137). In both designs, subgroup analyses revealed that studies involving either parent-child interactions, or the utilization of the ELISA method for OXT analysis, or unrestricted eating, drinking, or exercise before biofluid collection showed significantly higher than zero mean effect sizes. This review exposes the observed inconsistencies and suggests that standardized, replicable, and reliable approaches to assessing social interaction and measuring OXT concentrations need to be developed to study neurochemical mechanisms of sociality in humans.

Over the last decade, the number of human studies on oxytocin (OXT) has grown dramatically from 1,187 in the 1980s to 4,157 in the 2010s¹. In the 2020s, although the decade has just begun, the number of publications has already reached 1,637 studies². Such an increase cannot be explained by the general growth of the field of hormonal research, as another well-studied hormone, cortisol, has received 1.5 times less attention in the last decade (Figure 1). This makes OXT one of the most widely studied hormones of the present time. The increase in the number of articles devoted to OXT may be attributed to its role in prosocial behavior, defined here broadly as social interactions involving trust, cooperation, altruism, and social-emotional responses to an interaction partner (Bartz et al., 2011). Given the current era of rapidly expanding social media and extended social connections and networks, studying these behaviors is more relevant than ever. Furthermore, an interesting insight into the role of mass media in the popularization of OXT as a “social hormone” is presented in the paper of Steinbach and Maasen (2018).

Fig. 1. The Relative Increase in the Number of Articles Devoted to Oxytocin and Cortisol in Humans in the 1990s, 2000s, and 2010s Compared to the 1980s, PubMed.

Note. Source: https://www.ncbi.nlm.nih.gov/pubmed/

OXT is a substance produced in the hypothalamus and released either centrally, acting as a neuropeptide (neurotransmitter or neuromodulator), or secreted into the peripheral blood system, acting as a hormone. Alongside its crucial role in physiological processes such as childbirth, breastfeeding, and reproductive behaviors, OXT has gained significant attention in research aimed at the biological underpinnings of social processes and is often considered a biological indicator of social affiliation, bonding, and attachment (Feldman, 2012, p.527), touch-related affiliative processes (Feldman, Gordon, Schneiderman, et al., 2010), parenting (Gordon et al., 2017), individual sensitivity to social cues (Bartz et al., 2011), and social motivation (Bartz et al., 2011). Still, research on the causal relationship between OXT concentrations and social behavior has yet to generate converging evidence. For example, a positive OXT response (an increase in the concentration of OXT compared to baseline) to social interaction has been shown in some (Brondino et al., 2017; Feldman, Gordon, & Zagoory-Sharon, 2010; Krause et al., 2016; Vittner et al., 2018), but not in other studies (Bellosta-Batalla et al., 2020; Bick et al., 2013; Elmadih et al., 2014; Heinrichs et al., 2001); other studies show a decrease in the concentration of OXT after social interactions (Schladt et al., 2017). Similarly, different OXT outcomes have been registered when comparing experimental conditions with and without social interaction. The presence of social interaction can lead to an increase (Keri & Kiss, 2011; Keri et al., 2009; Kiss et al., 2011), no change (Smith et al., 2013; Yuhi et al., 2018), or a decrease in OXT concentrations (Schladt et al., 2017; Yuhi et al., 2018) when compared to the absence of social interaction. Regarding correlational studies, a full spectrum of relationships between OXT concentrations and social interaction has been observed. Thus, a positive relation (Algoe et al., 2017; Grewen et al., 2005; Schneiderman et al., 2012), a negative relation (Markova & Siposova, 2019; Tse et al., 2017; Vittner et al., 2019), and no relation (Julian et al., 2018; MacKinnon et al., 2014; Smith et al., 2013) between OXT concentrations and prosocial behavior have all been registered. Furthermore, in studies using a correlational design, additional complexity is added by using different social constructs, which can be assessed directly through observation or indirectly through self-reports, and different indicators of OXT concentration, such as baseline levels or changes in OXT concentrations. Thus, the field is now in an active development stage, in which researchers are questioning the simplistic interpretation of the relationships between OXT concentrations and human social interactions. For this reason, it is essential to consider factors that may influence both the direction and magnitude of the relationship between social interaction and OXT.

Overall, there is no gold standard in the assessment of either social behaviors or prosocial experimental settings; the investigation of the OXT released in response to social interactions is muddied by construct and measurement heterogeneity. In comparison, the Trier Social Stress Test is considered the gold standard for assessing human acute psychosocial stress under laboratory conditions: testing in this paradigm reliably induces a two-to-three-fold increase in concentrations of the stress hormone cortisol in approximately 70-80% of study participants (Allen et al., 2017). In contrast, the paradigms employed to study the relationships between OXT concentrations and social interactions are highly variable. For example, speaking and physical stimulation such as touching, kissing, hugging, and handholding are often used in studies involving dyads of romantic partners (Grewen et al., 2005; Light et al., 2005; Smith et al., 2013). Studies evaluating the role of OXT in the feeling of trust typically rely on social dilemma paradigms (Christensen et al., 2014). In parent-child studies, various play paradigms are assessed, and OXT concentrations are used as an indicator of dyadic synchrony and attachment quality. Even still, research on the causal relationship between OXT concentrations and social behavior has yet to generate converging evidence that also involves assessing maternal sensitivity, responsiveness, and interpersonal attunement, which can give additional complexity to the picture. For instance, Markova and Siposova (2019) observed that mothers characterized by low sensitivity to their children demonstrated increased concentrations of endogenous OXT after an interaction with their infants. Markova and Siposova (2019) interpreted these findings in light of the signaling role of OXT: they hypothesized that OXT is a hormone that responds to bonding deficiency; therefore, elevated concentrations were observed in the group of mothers characterized by low attunement. Similar findings were reported by Elmadih et al. (2014), who registered elevations in OXT concentrations in mothers with low sensitivity. The authors argued that OXT was released in response to the elevated stress of caregiving in the risk group of nonresponsive mothers, pointing to the anti-anxiety and anti-stress effects of the hormone. These findings question the linear relationship between the quality of social interaction and change in endogenous OXT.

Low convergence in the current understanding of the relationships between OXT concentrations and social interactions is further complicated by the issue of the correspondence between central and peripheral OXT concentrations. Endogenous OXT is both a neuropeptide and a peptide hormone produced in the hypothalamus (Carter, 1998). Nonetheless, because OXT is metabolized in the kidneys and liver, only part of the central OXT released into the bloodstream could be registered in the three most utilized in human studies peripheral biological fluids, i.e., blood, saliva, and urine. This raises the question of correspondence between central and peripheral OXT concentrations. The situation is complicated by the fact that, in addition to the brain, various other organs have been reported to produce OXT, such as the uterus, placenta, amnion, corpus luteum, testis, and heart (Gimpl & Fahrenholz, 2001); these OXT sources also could possibly contribute to the OXT concentrations in social interaction settings. A recent meta-analysis (Valstad et al., 2017) demonstrated that indicators of peripheral and central OXT concentration were not correlated under basal conditions (r = .08, p =.31), but significant associations were observed between these indicators after exogenous OXT administration (r = .66, p < 0.0001) and after experimentally induced stress (r = .49, p = .001). Thus, the question concerning the nature of endogenous peripheral OXT concentrations is still open.

There are also methodological debates concerning the accuracy of OXT assays (ELISA, enzyme-linked immunosorbent assay, vs. RIA, radioimmunoassay) and their cross-reproducibility (Christensen et al., 2014). In addition, questions surround the use of extracted versus unextracted samples. Extraction is the first recommended step for many biochemical procedures; it represents the separation of a target analyte from a matrix (components of a sample other than the target analyte). Some authors argued the need for extraction when analyzing the OXT concentration, positing a substantial enrichment and concentration of analytes with an increase in precision and reduction of matrix interference (Algoe et al., 2017). Other authors opted out of using the extraction procedure to minimize the variability introduced by the extra assay steps ostensibly (Hoge et al., 2012) or the unintentional removal of the majority of OXT, including OXT that is bound to other molecules in the plasma (MacKinnon et al., 2014), which could play a significant physiological role (MacLean et al., 2019). Additionally, the lack of extraction of samples may lead to an increase in the detected concentrations of OXT, as the antibodies utilized to perform immunoassay could non-specifically bind to other proteins and peptides. In addition to the presence or absence of extraction per se, numerous methodological features inherent in the extraction procedure can also vary from study to study and hence have a potential contribution to the resulting concentrations of OXT and their variability, such as extraction type (for instance, solid-phase or liquid-liquid), the brand of materials used for extraction, wash and elution buffers used, specific protocols utilized, etc. The resulting correlation between OXT concentrations in extracted and unextracted samples was found to be ambiguous as well: it was estimated to be strong (r = .89) by Michopoulos et al. (2011), whereas Szeto et al. (2011) registered no statistically significant association (Spearman’s rho = −.10, p = .54).

Other methodological discrepancies include the timing of the OXT collection and demographic variables. Justifying the sampling time, authors often cite the research by Amico et al. (1987), which demonstrated that the half-life of blood OXT is estimated at 5-10 min. Yet, Amico and colleagues used a synthetic OXT infusion, whereas there is a need for a more naturalistic study design and an estimation of OXT dynamics not only in blood but also in saliva and other biological fluids. Additionally, in a review focused on the role of OXT and vasopressin in human socio-emotional development, Torres et al. (2018) emphasized the need to consider the age and sex of participants, as both of these demographic variables consistently introduce variability into the pattern of findings on the relationships between endogenous oxytocin concentrations and human social interactions.

In summary, currently, there are numerous unstandardized paradigms for studying the relationships between OXT concentrations and human social interactions, and considerable debates on the methods and procedures for measuring the concentration of OXT have not subsided. As a result, this inconsistency in the paradigms of social interactions and the method of OXT measurement used can potentially lead to inconsistent findings with low reproducibility. In this study, we aimed to scrutinize the methodological and analytical limitations to better understand the relationships between OXT concentrations and human social interactions. To accomplish this aim, we established the following sub-aims: 1) to describe the included studies in terms of characteristics of study design, study sample, social interaction, and procedures for OXT collection and analysis; and 2) to analyze meta-analytically the available data to reveal (a) the relations between various types of social interactions and OXT concentrations; and (b) moderators that could affect this relationship. Given the conflicting nature of the published evidence, we did not have a priori hypotheses regarding the direction of the relations between social interactions and OXT concentrations.

Method

This systematic review and meta-analysis are presented in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (Moher et al., 2009). In addition, the study protocol detailing the review question, search strategy, inclusion criteria, and other information was pre-registered with PROSPERO on 25th December 2020 (CRD: CRD42020210970, https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=210970).

Search Strategy

Figure 2 illustrates the search and selection process.

Fig. 2. Flowchart of the Systematic Search and Study Selection.

Note. m designates the number of studies, k – the number of effect sizes.

A systematic literature search was conducted on July 2, 2020, across six online databases (Embase, PsycINFO, PubMed, Scopus, Web of Science Core Collection, and ProQuest Dissertations & Theses Global), covering the titles, abstracts, and keywords of publications; unpublished research was represented in the search hits with dissertations or theses. Search terms we used to find studies were a combination of terms identifying oxytocin (“Oxytocin,” “OXT”) and a wide range of terms associated with social interaction (“social*,” “synchron*,” “bond*,” “attachment,” “ interaction*,” “ affiliat*,” “dyad*,” “relationship*,” “attunement,” “communicat*,” “engagement,” “joint attention,” “coordinat*,” “empath*,” “gaze,” “mutual*,” “play,” “responsiv*,” “reunion,” “romantic,” “reciprocal,” “touch,” “contact,” “support,” or “emotion*”). The searched terms were combined with the database-specific filter “human,” where this was available.

The search process is reflected in Figure 2. The database search identified, in total, 17,900 citations: Embase (2,596 records), PsycINFO (1,836 records), PubMed (2,240 records), Scopus (4,568 records), Web of Science Core Collection (5,906 records), and ProQuest Dissertations & Theses Global (754 records). In addition, the reference lists of relevant reviews and included studies were screened for any additional studies (snowball search), which yielded 124 additional studies. After duplicates were removed, 9,810 records remained.

Inclusion and Exclusion Criteria

Due to the employment of many forms of social interaction and the lack of validated protocols for reliable OXT response in these settings, we aimed to include a broad spectrum of studies for a comprehensive perspective on the relationships between OXT concentrations and social interactions in humans rather than limit the study selection to specific types of experimental designs. Because the scope of this research was to investigate the relationships between OXT concentrations and social interactions, we excluded studies involving additional confounders, such as stressors exposure (e.g., Trier Social Stress Test, Still face paradigm, conflict interaction) or physiological processes (delivery, breastfeeding, and sexual stimulation), or membership in a sample representing a clinical/at-risk population. Due to the effects of stress on OXT concentrations illustrate a complex physiological mechanism, some studies have observed that stressors initiate an OXT release (Jong et al., 2015), whereas other research points to inconsistent patterns of OXT concentrations in response to stressors with both an increase and decrease, depending on blood pressure and breastfeeding status (Light et al., 2000). Nevertheless, other studies revealed no change in OXT in response to stressors (Altemus et al., 2001). Thus, there is a clear need to undertake a synthesis of the literature on the role of OXT in the stress response, but this task is outside the scope of this systematic review and meta-analysis.

The following inclusion criteria were applied: (1) published in a peer-reviewed journal, including articles in press or dissertations or theses between 1970 and July 2, 2020; (2) empirical and with a quantitative indicator of the concentration of OXT in any biological fluid; and (3) inclusive of any social interaction of two or more participants. We put no restrictions on the language of publication.

The following criteria were applied to exclude studies from the systematic review: (1) case series and case reports, reviews and meta-analyses, conference abstracts, opinions, commentaries, letters to editors; (2) studies conducted on nonhuman animals; (3) studies utilizing the administration of any substances, including exogenous OXT and invasive procedures that could potentially affect OXT concentration; (4) studies including a presentation of social stimuli (e.g., video, pictures) without live interactions; (5) clinical population or datasets with combined observations from clinical and non-clinical participants; (6) no relation between OXT and social interaction was studied (i.e., a) OXT was not collected before or within 1-2 hours after social interaction; instead, it was collected on a separate day, or sometimes several months before or after the social interaction; b) OXT was collected only after social interaction, without analyzing its correlations with social constructs, and without collecting OXT from a control group that did not undergo social interaction; as a result, no data can be obtained for the purposes of our study; these cases should not be confused with cases where OXT was explicitly collected and the data were analyzed but not presented in the article – in those instances, we contacted the authors and requested that they provide the missing data); (7) studies pooling of OXT samples from different participants into one before analysis, or combining groups with different types of interaction into one analysis; (8) studies utilizing social interactions involving breastfeeding, sex, massage, and physical exercise (due to the higher physiological component in these kinds of interactions compared to social encounters), stress exposure (due to complex and sometimes conflicting data on the effects of stress exposure on OXT concentrations), as well as artificial interaction techniques (i.e., mirroring); (9) study participants were from high-risk populations (for example, survivors of natural disasters); (10) data from the study duplicated data from another study; and (11) OXT donors were women during the early post-delivery period.

Selection of Eligible Studies

After removing duplicates, 9,810 records remained and were independently double-screened by authors who examined all titles and abstracts to identify eligible studies using Abstrackr software (Wallace et al., 2012). The list of records was formed randomly for each author, and pairs of authors were also formed randomly for each record. Twelve raters took part in the abstract screening and eight in the full-text screening. The disagreements between the raters over the eligibility of studies on the screening stage of both abstracts and full texts were resolved at weekly team meetings of the entire research group.

In cases where entries did not provide reviewers with enough information to decide whether they should be included, the studies were selected for further evaluation. In total, 195 records were retained after the abstract screening. Their full texts were retrieved and independently double-screened by authors, again utilizing Abstrackr software (Wallace et al., 2012). Ultimately, 63 out of 195 studies were included in the systematic review and further data extraction.

Data Extraction

Data from each study were extracted into spreadsheets that included information on: (1) reference information (e.g., title, authors, publication year); (2) study design (associative (correlational and regression), within-group comparison, between-group comparison); (3) characteristics of the study sample (age, sex, size, race/ethnicity, type of relations between interacting participants, phase of the menstrual cycle, using of hormonal drugs); (4) characteristics of social interaction (type of social interaction, the place where the experiment was conducted, duration of social interaction, type of assessment of behavioral constructs – questionnaire on the current interaction, general questionnaire, behavioral analysis); (5) characteristics of the OXT collection and analysis (type of biological fluid, type of assay for analysis, time of day of the collection, the presence of a procedure for the separation of participants preceding the collection of baseline OXT in order to avoid their interaction, OXT sampling relative to the beginning of social interaction, the presence of extraction procedure, extraction type, sample collection method, special precautions before OXT collection, such as fasting); and (6) statistical information utilized to derive an effect size. Two authors performed data extraction independently, and were checked by a third author.

If the data reported in the article were insufficient to compute effect sizes (ESs), the publication authors were contacted twice during a given month to provide the data. If the authors did not respond to the request, but the data were present in figures, we digitized the data points using the open-source tool WebPlotDigitizer (Rohatgi, 2020). Otherwise, the article in question was excluded from the meta-analysis. We extracted four types of statistics during the extraction phase: mean differences between OXT concentrations, Pearson’s and Spearman’s correlation coefficients between OXT concentrations and indicators of social interactions, and regression coefficients for associations between OXT concentrations and indicators of social interactions. The ESs were estimated only for mean differences and correlation coefficients; only these indices were used for further meta-analytic synthesis. We did not include regression coefficients due to the high variability of the regression models used and the variables included in them, which would have added even greater heterogeneity to the analyses. Also, no data on mean differences were presented in one of the included studies (Grewen et al., 2005), and the authors did not provide them to us, although t-statistics data were presented. Despite the existence of methods for conversion of t-statistics into Cohen’s d, they cannot be used for a dependent or paired t-test without correlation coefficient r (Cooper et al., 2019, p.216; Wilson, 2016, p.4).

We used the following formulas to calculate Hedges’ g ESs and variances for the results presented as mean differences. Hedges’ g is Cohen’s d corrected for small samples (Cumming, 2012, p.309). We calculated Hedges’ g_av using the following equation for dependent pairs of means (for pre-post design and between-conditions (control vs. interaction) design), with SD_av value as a standardizer (Cumming, 2012, p.291), taking into account the lack of correlation coefficients r and other data from which they can be extracted in the vast majority of studies:

$${\text{Hedges}}^{\prime }{g}_{av}=\frac{M_{diff}}{S{D}_{av}}\times \left(1-\frac{3}{4df-1}\right),$$

where $\frac{M_{diff}}{{SD}_{av}}$ = Cohen’s $d_{av}$, $M_{diff}$ is the mean difference, ${SD}_{av}=\sqrt{\left(\frac{S{D}_1^2+S{D}_2^2}{2}\right)}$, ${SD}_1$ and ${SD}_2$ are standard deviations of the first and second measure, $\left(1-\frac{3}{4df-1}\right)$ = correction factor $J$ for small sample sizes, $df$ is the degrees of freedom, which for dependent groups is $n\text{-}1$, where $n$ is the number of participants in the sample. For the calculation of variance of Hedges’ $g_{av}$, we used the following equation (Cumming, 2012, p.313):

$$v{g}_{av}=J^2\times v{d}_{av},$$

where ${vd}_{av}=\frac{n_1+n_2}{n_1\times n_2}+\frac{d_{av}^2}{2(n_1+n_2)}$; equations for $d_{av}$ and $J$ are presented above; $n_1$ and $n_2$ are the numbers of participants in the sample. In the absence of correlation coefficients $r$ and other data from which they can be extracted in the vast majority of studies, a conservative method was utilized for variance calculation, adapted from the calculation of variance for Cohen’s d for independent samples (Hirst et al., 2018).

Fisher’s $r$ to $z$ transformation was used for Pearson’s and Spearman’s correlation coefficients (Fisher, 1921) in order to normalize their distribution. The following formulas were used for the calculation of $z$ and variance values (Cooper et al., 2019, pp.220-221):

$$\begin{gathered} z=\frac{1}{2}\times \text{ln}\left(\frac{1+r}{1-r}\right) \\ {v}_z=\frac{1}{n-3} \end{gathered}$$

where $r$ is correlation coefficient, $n$ is the number of participants in the sample.

Despite the availability of formulas for converting correlation coefficients into Hedges’ $g$ values, we decided not to follow this approach. Consequently, we did not combine the ESs derived from mean values and correlation coefficients due to the conceptual differences in our research questions.

The criteria for inclusion in the meta-analysis were: 1) the presence of data sufficient to compute ESs (sample sizes, mean values, and standard deviations or errors for causal designs, and correlation coefficients and sample sizes for studies using a correlational design); 2) the presence of at least two studies in each of the investigated experimental designs in order to perform separate meta-analyses, specifically, a) within-group pre-post interaction; b) within-group between-conditions (control condition vs. interaction); c) between-group (control condition vs. interaction); and d) correlational. Yet, data from between-group design did not meet these inclusion criteria; there were only two ESs from the same study in this category of designs, so these data were excluded (Figure 2).

We used the following convention to interpret the ESs (Cohen, 1988): 1) for d/g ESs, 0.2, 0.5, and 0.8 are considered small, medium, and large ESs, respectively; and 2) for z/r ESs, 0.1, 0.3, and 0.5 are considered small, medium, and large ESs, respectively.

Data Analyses

Prior to the analyses, we checked ESs for outliers by screening for standardized z-values larger than 3.29 or smaller than −3.29 (Tabachnik & Fidell, 2013). As a result, only one ES (from Brondino et al. (2017)) was deemed as an outlier. An extremely high value of this outlier (Hedges’ g = 1.5) could be explained by the nature of social interaction in this particular case, namely, gossiping about a scandalous event in the life of a friend that could be overly emotional compared with other cases of moderately emotional conversations. Furthermore, the results were affected by it; specifically, the analysis without the outlier led to results non-significantly different from zero (p = .101), whereas including it led to approaching the borderline of statistical significance (p = .065). For these reasons, we excluded the outlier from the analysis.

The analyses were performed using the metafor package (Viechtbauer & Viechtbauer, 2017) for the R environment (Version 4.1.2; R Development Core Team (2021)) following the guidelines formulated by Assink and Wibbelink (2016) for modeling a three-level random effects model as described by Van den Noortgate et al. (2013). Because traditional approaches to meta-analyses require all included ESs to be independent, we used a method by which the (possible) dependence of ESs can be modeled by combining ESs in a three-level random effects model (Assink & Wibbelink, 2016). This model accounts for three sources of variance: sampling variance (level 1), the variance between ESs from the same study (level 2), and the variance between studies (level 3). The advantage of the three-level approach is that all ESs derived from the same study can be included, resulting in maximum information and statistical power. To address potential correlated sampling variance (i.e., correlations of level 1 sampling errors), we performed cluster-robust variance estimation on all three-level multilevel meta-analytic models produced in metafor, augmented with clubSandwich package with small-sample correction (Pustejovsky, 2022). In addition, we used restricted maximum likelihood (REML) approaches to test whether ES estimates differed significantly from zero.

Q and I² statistics were utilized to assess heterogeneity among studies. A significant Q value indicates a lack of homogeneity of findings across studies. Still, its results reflected variance between all ESs in the data set rather than the within-study variance (level 2) and the between-study variance (level 3), which are of particular interest. Unlike Q values, I² values represent the proportion of the observed variance resulting from variance in true ESs (within-study variance, level 2, and between-study variance, level 3) rather than sampling error (sampling variance, level 1). I² values were obtained using formulas of Cheung translated into R syntax (Assink & Wibbelink, 2016). I² values of 25%, 50%, and 75% correspond to small, medium, and large amounts of heterogeneity (Higgins et al., 2003). Furthermore, we analyzed the between-study variance using the tau-squared (τ²) index. And finally, we used a likelihood ratio test to test for between-study and within-study heterogeneity (Raudenbush & Bryk, 2002).

If there was evidence for heterogeneity in ESs, then moderator analyses were completed. For moderator analyses, categorical variables were transformed to k – 1 dummy variables through binary coding, and continuous variables were centered around their means (Assink & Wibbelink, 2016). When analyzing categorical moderators, subgroups with any number of studies and ESs were taken into consideration. Because there was no accepted threshold for the minimum number of studies and ESs for a meta-analysis, we decided to provide all the data available, leaving judgment of the minimum threshold to the discretion of the reader.

Positive values of all ESs, whether they are Hedges’ g or correlation coefficients, indicate that 1) social interaction leads to an increase in OXT concentrations compared with baseline levels (pre-post design); and/or 2) higher OXT concentrations in the presence of social interaction compared with the absence of social interaction (between-groups comparison and between-conditions (within-group) comparison design); and/or 3) positive associations between OXT concentrations and the measures of social interaction.

Study Quality and Reporting Bias Assessment

Study quality was independently evaluated by two raters using the a priori developed criteria aligned with the Critical Appraisal Tools of Joanna Briggs Institute (Joanna Briggs Institute, 2020) and also included additional relevant items. Because we included studies with heterogeneous designs and the most validated instruments for the assessment of study quality were developed for specific study designs (e.g., cross-sectional, case-control, cohort), no specific tool was appropriate for all the studies in our pool. Study quality was deemed high if a study reported the following information for OXT donors: 1) sample size; 2) age; 3) race/ethnicity; 4) sex composition; 5) duration of social interaction; 6) time of day of OXT collection; 7) OXT sampling time relative to the beginning of social interaction; 8) restrictions before OXT collection (e.g., eating or breastfeeding); 9) health status of non-clinical participants; and 10) compared groups have similar characteristics and treatment conditions except for the presence of social interaction (for between-groups design only). We did not include study design, type of relations between interacting participants, type of social interaction, type of biological fluid, and type of assay for analysis in this assessment because this information was provided for all included studies.

Disagreements between the raters were resolved through discussion. The quality score was computed as the proportion of items for which information was provided out of the total number of items. Study quality was not used as an exclusion criterion; the quality scores were utilized as a moderator to control for the risk of bias.

Reporting bias represents a tendency for studies with significant findings to have a higher probability of being published (both by journals and authors) compared to studies with non-significant results. First, to analyze the reporting bias, the data were visualized using a funnel plot of ES against standard error. The funnel plot is asymmetrical if publication bias is present (Torgerson, 2006). Second, Egger’s regression test (Egger et al., 1997) was applied to analyze funnel plots asymmetry, where publication bias was considered present when the intercepts significantly deviated from zero. In addition, Egger multi-level meta-analysis (MLMA) test (Rodgers & Pustejovsky, 2021) was completed for evaluation of the slopes significantly deviated from zero.

The data and analysis codes are available on the Open Science Framework (Burenkova et al., 2023, June 23).

Results

Qualitative Synthesis

Study Characteristics

Sixty-three studies (m) were included in the qualitative synthesis (Table 1). Description of each study, including ESs, baseline OXT concentrations, demographic characteristics, social interaction parameters, and data on OXT collection and analysis, are provided in Appendix.

Table 1.

Characteristics of Studies Included in the Qualitative and Quantitative Synthesis

		Qualitative Synthesis	Quantitative Synthesis (meta-analysis)
Variable	Category	Number of studies (m)	Number of studies (m)	Number of effect sizes (k)
Report date	1970s	0	0	0
	1980s	0	0	0
	1990s	1	0	0
	2000s	10	8	15
	2010s	48	40	205
	2020	4	3	3
Study design*	Within-group pre-post interaction	40	27	72
	Within-group between-conditions (control vs. interaction)	5	5	8
	Between-group (control vs. interaction)	2	0	0
	Associative (correlational and regression)	33	31	143
Types of interactions	Playing	36	29	133
	Speaking	9	6	23
	Tactile	5	5	16
	Trust-related interaction	4	3	7
	Other	3	3	6
	Speaking and Tactile	3	3	14
	Singing	3	2	8
	Hypnosis	2	2	16
Relations between interacting participants	Parent and child	40	32	147
	Romantic couple	8	6	33
	Dyad of strangers	8	6	11
	Group of 3+ (3 or more) people except for parents with children	6	5	14
	Couple of familiar individuals	2	1	1
	Hypnotist and a participant	2	2	16
	Psychotherapist and a participant	1	1	1
Age category	Adults	58	48	203
	Children	10	8	20
Sex	Females only	40	34	112
	Males only	15	12	40
	Both females and males	19	13	45
Race/ethnicity	White	19	16	73
	Black	9	7	29
	Hispanic	7	5	20
	Asian	7	6	26
	Non-Hispanic	2	2	10
Experiment location	Laboratory condition	36	29	138
	Home	15	12	44
	Other	10	8	33
Time of day of OXT collection	Afternoon and evening (after 12:00 p.m.)	32	27	138
	Morning and afternoon	14	9	40
Restrictions before an experiment	Eating	21	19	70
	Smoking	17	12	41
	Caffeine consumption	16	12	37
	Breastfeeding	14	14	60
	Alcohol consumption	10	8	29
	Drinking (except for water)	8	8	32
	Doing exercise	5	4	13
	Drinking any liquids	4	4	11
Biological fluid collected	Blood plasma	33	26	99
	Saliva	25	22	102
	Urine	7	6	20
	Blood serum	2	1	2
Techniques for samples’ collection	Venipuncture	20	18	65
	Absorbent device	18	16	83
	Intravenous catheter/cannula	13	7	31
	Passive drool method	8	7	19
	Urine miscellaneous collection	7	6	20
Methods for OXT analysis	ELISA, enzyme-linked immunosorbent assay	48	42	188
	RIA, radioimmunoassay	14	8	34
	HPLC, high-performance liquid chromatography	1	1	1

Note. *hereinafter, the total sum of studies could exceed 63 because one study could include multiple designs, types of interaction, and other methodological variations.

These studies were published between 1991 and 2020, with a substantial increase in recent years: 83% of the studies (m = 52) were published after 2010. Although the lower search limit was 1970, nothing was published until 1991. The most widely used study designs were associative (correlational and regression) and pre-post designs (Table 1). Studies were conducted in 15 countries: USA (m =22), Israel (m =17), Hungary (m =5), Canada (m =3), Germany (m =3), Japan (m =3), China (m =2), Czech Republic (m =2), Sweden (m =2), Italy (m =1), Portugal (m =1), Spain (m =1), Switzerland (m =1), UK (m =1), and Jamaica (m =1).

Sample and Interaction Characteristics

The total number of participants who provided biological fluid samples for OXT analysis (hereinafter, all descriptions will only be applied to participants who provided OXT samples and not to all participants who took part in the studies) was 4,386. Sample sizes of studies ranged from 4 to 354 participants. The average sample size was 71.90±66.91 (M±SD), and the Mdn value was 53.00.

Parents and children were the most common participants, followed by romantic couples and dyads of strangers; in most studies, OXT donors were adult participants (Table 1). In parent-child dyads, the majority of participants who donated biofluids for OXT analyses were parents (m = 35), and in 10 studies, children were donors. Among these 45 studies, 5 used samples from both parents and children.

The number of studies with available average ages of participants who provided biological fluid samples for OXT analysis was 54. The age of these participants was 28.24±9.73 y (M±SD), and the Mdn value was 29.28 y, range 0.38-54.20 y. In a subsample of children, the age of 394 participants was 4.93±3.65 y (M±SD), and the Mdn value was 4.49 y, range 0.38-11.55 y. In a subsample of adults (n = 3,417), the age of participants was 31.07±5.47 y (M±SD), and the Mdn value was 29.53 y, range 20.64-54.20 y.

The number of studies with available participants’ sex was 58. In these studies, 69.06% of the participants were female; most of the studies included only females (Table 1). Of 63 studies, females after menarche (older than ten years) participated in 56 studies. In these 56 studies, only 17 mentioned the menstrual cycle phase, either directly or that the relevant data were collected; the cycle phase was stated directly only in 5 studies. This reporting bias makes this factor unsuitable for further quantitative analysis. In the remaining studies (m = 39), either description of the menstrual cycle phase was not applicable (due to gestation or early postpartum period, up to 6 weeks), or it was not provided. Of the 56 studies with females after menarche, breastfeeding status was mentioned in 29 studies, in 18 of which at least some women were breastfeeding at the time of the study. Race/ethnicity was disclosed in 20 of 63 studies; in the vast majority, only White participants were involved (Table 1).

The number of studies with available data on the duration of interactions was 53. This duration was 18.76±19.16 min (M±SD), and the Mdn value was 10.00 min, range 2-90 min (see also Figure 4A). Among the types of interactions, the most common was playing, which included, in the case of between-adult interactions, any forms of playing (such as verbal creativity games or board games) and, in the case of parent-child interactions, any type of interaction, except for exclusively speaking or tactile. The latter two were, respectively, in second and third place after playing (Table 1).

Fig. 4. A Histogram of the Number of Observations Available for (A) Interaction Duration, min; (B) Time After the Start of Interaction, min; (C) Time After the End of Interaction, min.

Note. The number of observations refers to the number of individual values, which could exceed the number of studies if there are several values in each study.

For pre-post design, we were interested in the dynamics of participants’ characteristics of behavior or experience in response to social interaction. These objective or subjective (self-reported) characteristics could potentially reflect the valence of the social interactions that may be relevant for explaining the patterns of the OXT response to social interaction. Of the 40 studies with pre-post design included in the review, 13 had data on the dynamics of participants’ characteristics of behavior or experience during social interaction. In all these cases, questionnaires such as the PANAS (Positive and Negative Affect Schedule), STAI (State-Trait Anxiety Inventory), and others were used. Among these 13 studies, six had shown positive dynamics of behavioral indicators following the interaction (increase in positive characteristics or decrease in negative ones), one had shown negative dynamics, three had shown no change, and the rest did not report their results.

In correlational design, social interactions were classified as positive in 28 studies, negative in 8, synchronizing in 4, and had unclear valence in one study (touch and eye gaze that could have both positive and negative valence). We separately identified the category of synchronizing because, even though in numerous studies, interpersonal synchrony enhanced positive social interaction and strengthened social bonds, a few studies have also revealed its potential adverse effects, inducing hostile or non-collaborative attitudes towards out-group members (Wiltermuth, 2012a, 2012b). In correlational studies, primarily behavioral analyses were utilized (m = 25); also, in 5 studies, authors used questionnaires or interviews about ongoing social interaction, and in 7 studies, they used questionnaires or interviews about psychological traits associated with social behavior, like relationship quality, support, engagement, attachment avoidance, and others.

OXT Concentrations Assessment

Sampling

In most studies, samples were collected in the afternoon and evening (Table 1). For 17 studies, the time of day of OXT collection was not specified. In 36 studies, an experiment was performed in a laboratory (Table 1), while 15 studies were conducted at home, and 10 studies were conducted in other settings (like clinics, art classes, university campuses, and public places). For three studies, conditions for sample collection were not clearly described. The presence of the procedure of separation of partners before interaction was mentioned in approximately half of the studies (m = 26), and it was not mentioned in the rest of the studies. Information regarding the use of hormonal contraception, steroid medications, or other medications that would likely influence OXT concentrations was mentioned in 18 studies. Among 6 of them, participants did not use any of these medications. Before starting the experiment, participants were asked to follow several procedures, of which not eating was the most common (Table 1). For 33 studies, limitations on eating, drinking, or other relevant behaviors were not clearly described.

Among biological fluids, blood plasma (m = 33), saliva (m = 25), urine (m = 7), and blood serum (m = 2) were collected (Table 1). All studies include a description of the type of biological fluid collected. In Figure 3, the annual number of studies according to the biological fluids collected is reflected. Saliva collection, which began in the early 2010s due to the availability of the relevant assays, has become dominant over blood collection in recent years.

Fig. 3.

The Annual Number of Studies Based on the Type of the Biological Fluids Used

The following techniques were utilized for sample collection (Table 1): 1) for blood: venipuncture (m = 20) and intravenous catheter/cannula placement (m = 13) methods, and for two studies, blood collection technique was not specified; 2) for saliva: absorbent device method (use of absorbent tampons for saliva collection, m = 18), passive drool method (passive release of saliva accumulating in the mouth into a test tube, m = 8); 3) for urine collection, there was no standardized collection method, and in some cases, the type of container used for the collection was not reported.

Among 36 studies that utilized blood collection protocols, heparin (m = 16) or EDTA (m = 9) was added as an anticoagulant to samples; 10 studies had no information on whether and which anticoagulants were added. The protease inhibitor Trasylol (aprotinin) was added to samples in 16 studies. Of these, 14 studies used the above-mentioned anticoagulants, and 2 did not.

In correlational design, most studies used baseline OXT concentration (m = 21). Also, the following measurements were used: post-interaction OXT concentration (m = 7), OXT concentration change (m = 5), log(post-interaction OXT) minus log(baseline OXT) (m = 3), 24-h cumulative OXT concentration (m = 1), the area under the curve (AUC, m = 2), and composite OXT concentration averaged from baseline and post-interaction assessments (m = 1).

Samples’ preparation and OXT measurement

The details of samples’ preparation and OXT measurement were explicitly presented not in all studies; sometimes, authors referred to other studies, and in these cases, we took information from the references provided by authors, if it was available. OXT extraction procedure was performed in 20 studies, among which 12 studies used solid phase extraction, 3 liquid-liquid (acetone-ether) extraction, and 5 did not specify the type of extraction. When utilizing solid phase extraction, three main devices were used: Strata-X (3 studies with plate version, 1 study with cartridge version), Oasis Hydrophilic-Lipophilic-Balanced (HLB) device (1 study with plate version, 2 studies with cartridge version), and Sep-Pak C18 (4 studies with cartridge version); furthermore, 1 study mentioned the use of all types of devices. Some studies (m = 13) utilized a concentration procedure without extraction (all samples were saliva). In 14 studies, the authors diluted the samples (all were blood).

In all studies, the method of the analysis of OXT concentrations was stated. In most studies, OXT concentrations were examined using ELISA (m = 48). Other methods included RIA (m = 14) and high-performance liquid chromatography, HPLC (m = 1) (Table 1). The only explicitly stated type of ELISA was competitive enzyme-linked immunosorbent assay (m = 44), and in 4 studies, ELISA type was not specified. Enzo (m = 20) and Assay Design (m = 22) kits were among the most used. Among RIA, liquid phase radioimmunoassay was used in 3 studies, and in the rest of them, the type of RIA was not specified, as well as the names of kits used. Relatively few studies using ELISA employed an extraction procedure (m = 11 of 48 studies), whereas extraction was much more common in studies using RIA (m = 8 of 14 studies).

The Resulting Values

Baseline OXT values

Of 63 studies, baseline OXT values were presented in 45 studies, and 39 of those studies provided measurements in standard units (pg/mL) or in units that could be converted to standard ones. In other studies, baseline OXT values were either not provided or were expressed in different units of measurement (pg of OXT/mg protein, μg of OXT/mg of creatinine, ng of OXT, pM of OXT), which were not compatible with the main body of data (pg/mL).

In blood, baseline OXT concentration ranged from 0.47 to 509.83 pg/mL, with a mean of 192.2 pg/mL (SD: 179.0 pg/mL, 37 values). Thirty-two values were obtained from samples without extraction and five with extraction. Extracted samples, one analyzed by ELISA and four by RIA, produced a mean of 3.9 pg/mL (SD: 3.5 pg/mL, range 1.53-9.6 pg/mL). Among non-extracted values, the method of analysis had a great influence on the size of the values: ELISA data were higher (M = 307.6 pg/mL, SD = 124.9 pg/mL, range 0.47-509.83 pg/mL, 23 values) than RIA data (M = 1.7 pg/mL, SD = 0.4 pg/mL, range 0.8-2.1 pg/mL, nine values).

In saliva, OXT concentration ranged from 2.0 to 536.8 pg/mL, with a mean of 72.5 pg/mL (SD: 104.3 pg/mL, 37 values). Twenty-seven values were represented by samples without extraction and ten with extraction. Extracted samples, all analyzed using the ELISA method, produced a mean of 100.2 pg/mL (SD: 171.7 pg/mL, range 2.0-536.8 pg/mL). Two of the values (263.0 and 536.8 pg/mL) seemed too high for the presence of an extraction procedure, so we contacted the authors (Kasos et al., 2018), and they confirmed that the extraction procedure took place. When these two values were removed as outliers, the following values were obtained for the remaining seven studies: mean of 25.3 pg/mL, SD 21.4 pg/mL, range 2.0-54.4 pg/mL. Among non-extracted values, the method of analysis had a significant influence on the size of the values: ELISA generated higher (M = 64.5 pg/mL, SD = 67.1 pg/mL, range 5.7-193.9 pg/mL, 26 values) than RIA (4.1 pg/mL, one value) values, similar to the case of blood.

Urine samples were represented by values that have different unmatched units of measurement; therefore, the corresponding statistics are not provided here.

The dynamics of OXT concentrations in response to social interaction

The collection times of biological samples relative to the beginning (Figure 4B) and the end of social interaction (Figure 4C) varied. The number of studies with available data on time passed after the beginning of social interaction by the moment of OXT collection was 42 (Figure 4B). This time was 22.19±24.86 min (M±SD), the Mdn value was 14.00 min, range 0-105 min. The number of studies with available data on time passed after the end of social interaction by the moment of OXT collection was 35 (Figure 4C). This time was 7.16±10.11 min (M±SD), the Mdn value was 3.00 min, range 0-45 min.

As the dynamics of changes in OXT concentrations in response to social interactions still need to be sufficiently established, we attempted to piece together data available from all included studies. To do this, we considered all available pairs of pre-post interaction values and calculated the rate of change of OXT concentrations post-interaction to baseline. For the parameter “time after the start of the interaction,” the number of values of pre-to-post change in OXT concentrations was 35 for saliva, 23 for blood, and 7 for urine. For the parameter “time after the end of the interaction,” the number of values of pre-to-post change in OXT concentrations was 31 for saliva, 23 for blood, and 5 for urine. The resulting dynamics are shown in Figure 5, for both after the start (Figure 5A) and the end (Figure 5B) of interaction. Blood collection was limited to 25 minutes after the start of the initial interaction and 10 minutes after its end, and during all these time ranges, the blood OXT concentrations remained, on average, equal to the baseline. Saliva collection provides a broader range of collection times, and during all these time ranges, the salivary OXT concentrations increased compared to the baseline. According to the Spearman rank-order tests, the only statistically significant association between time and the ratio of post-interaction to baseline OXT concentration was revealed for salivary OXT in the case of the time after the start of interaction (ρ = .34, p = .049).

Fig. 5. Ratio of post-interaction to baseline (pre-interaction) OXT concentration for the Time after the Start of Interaction (A) and Time after the End of Interaction (B).

Note. A "1" value on the Y axis indicates that there is no change from pre- to post-interaction. The lines represent a linear regression line; shaded areas represent standard error; the scaling of the X-axis is different between Figures A and B for more effective data visualization. Please refer to the electronic version of the manuscript for color figures.

Meta-Analysis

Fifty-one studies (m) with a total number of 223 ESs (k) were included in the meta-analysis. Table 1 presents an overview of all included studies.

The total number of participants who provided biological fluid samples for OXT analysis was 3,741. Sample sizes of studies ranged from 4 to 354 participants. The average sample size was 76.35±72.99 (M±SD), and the Mdn value was 50.00. The number of studies with available average ages of participants who provided biological fluid samples for OXT analysis was 41. The age of these participants was 27.65±10.25 y (M±SD), and the Mdn value was 29.34 y, range 0.38-54.20 y. In a subsample of children, the age of participants was 3.81±3.12 y (M±SD), and the Mdn value was 3.16 y, range 0.38-8.95 y. In a subsample of adults, the age of participants was 30.80±5.61 y (M±SD), and the Mdn value was 29.60 y, range 20.64-54.20 y. The number of studies with available participants’ sex was 46. In these studies, 68.02% of the participants were female. The number of studies with available data on the duration of interactions was 42. This duration was 19.23±19.80 min (M±SD), and the Mdn value was 11.65 min, range 2-90 min. Table 2 shows how the types of interactions correspond to the relationships between the participants who are interacting.

Table 2.

Types of Interactions and Relationships Between Interacting Participants

Types of interactions	Relations between interacting participants	Number of studies (m)	Number of effect sizes (k)
Pre-post design
Parent and child	Playing	11	27
	Speaking	1	1
	Tactile	4	14
Romantic couple	Playing	1	1
	Speaking	1	8
	Speaking and Tactile	1	4
Couple of familiar individuals	Speaking	1	1
Dyad of strangers	Playing	1	1
	Speaking	1	1
Group of 3+ people except for parents with children	Other	2	3
	Playing	1	1
	Singing	2	5
Hypnotist and a participant	Hypnosis	2	4
Psychotherapist and a participant	Other	1	1
Within-group between-conditions design
Dyad of strangers	Trust-related	3	3
Group of 3+ people except for parents with children	Other	1	2
	Singing	1	3
Correlational design
Parent and child	Playing	21	103
	Tactile	1	2
Romantic couple	Speaking	2	10
	Speaking and Tactile	3	10
Dyad of strangers	Speaking	1	2
	Trust-related	1	4
Hypnotist and a participant	Hypnosis	2	12

Studies with Pre-Post Design

Overall effect

The overall mean ES of social interaction on OXT concentrations was g = 0.079 (95% CI −0.017 to 0.175) and was not significantly different from zero (p = .101, for m = 27 studies, k = 72 ESs), indicating that social interaction does not lead to either increase or decrease in the OXT concentrations (Table 3 and Figure 6).

Table 3.

Mean Effect Sizes for Each Design Type

Design type	Number of studies (m)	Number of effect sizes (k)	Mean g or z-score (SE)	95% CI	t-statistic	p-value	τ2
Within-group pre-post interaction (g)	27	72	0.079 (0.046)	−0.017; 0.175	1.710	.101	0.025
Within-group between-conditions: control vs. interaction (g)	5	8	0.256 (0.222)	−0.360; 0.872	1.157	.312	0.216
Correlational (z)	31	143	0.137 (0.039)	0.057; 0.218	3.503	.002	0.035

Fig. 6. Forest Plot for Pre-Post Design.

Note. The overall mean ES of social interaction on OXT concentrations was not significantly different from zero (g = 0.079, p = .101). If there was more than one ES in an article, decimal points after the article year indicate different ESs originating from the same article.

Publication bias

We did not observe the asymmetry of the funnel plot (Figure 7). The intercepts from Egger’s regression test did not significantly deviate from zero (b = 0.011, t(70) = 0.72, p = .474). The MLMA Egger test yielded nonsignificant slopes (b = −0.282, p = .642). This suggests that there was no indication of publication bias for studies with pre-post design.

Fig. 7.

Funnel Plot for Pre-Post Design

Heterogeneity in effect sizes

The results of the test for heterogeneity revealed no significant variation between the ESs in the data set (Q(df = 71) = 82.441, p = .167). Nevertheless, the I² index was 32.16%, indicating moderate heterogeneity. According to the approach described by Hunter and Schmidt (2004), heterogeneity can be regarded as substantial if less than 75% of the total amount of variance can be attributed to variance at level 1 (sampling variance). In our dataset, 67.84% of the total amount of variance was attributable to sampling variance; this necessitated moderator analyses. Variance component estimates showed significant variation between studies (σ² = 0.025, χ² = 18.45, p < .0001) but not between ESs within studies (σ² = 0.0, χ² = 0.0, p = 1.0).

Moderator analyses

Moderator analyses were conducted to identify possible characteristics of the study sample, social interaction, or OXT collection and analysis that could moderate the effect of social interaction on OXT concentrations. Table 4 presents the results of the moderators and the omnibus test statistic (p < .05). As evident from Table 4, only the method for OXT analysis moderated the effect of social interaction on OXT concentrations (F (2, 6.51) = 11.631, p = .007). Mean ES in experiments that used ELISA was significantly larger than zero, whereas the ones that utilized HPLC were significantly lower than zero; still, the latter effect should be interpreted with caution due to the ES being represented by only one value.

Table 4.

Results for Moderators in Pre-Post Design

Moderator variables	Number of studies (m)	Number of effect sizes (k)	β0, mean g (95% CI)	Omnibus test	p-value	Residual heterogeneity	p- value	τ2
Study sample
Age (years)	22	55	0.004 (−0.003; 0.010)	F(1, 8.41) = 1.527	.250	Qe(53) = 40.982	.886	0.007
Sex (% of females)	25	65	0.001 ( −0.001; 0.002)	F(1, 2.59) = 2.536	.223	Qe(63) = 76.419	.119	0.026
Race/ethnicity (% of Whites)	10	42	−0.006 (−0.015; 0.004)	F(1, 5.81) = 2.146	.195	Qe(40) = 38.905	.519	0.025
Sample size	27	72	0.001 ( −0.005; 0.005)	F(1, 1.41) = 0.400	.614	Qe(70) = 81.396	.166	0.026
Breastfeeding status				F(1, 3.88) = 0.507	.517	Qe(23) = 20.000	.642	<.0001
Breastfeeding	10	17	0.110 (−0.023; 0.243)		.093
Not breastfeeding	2	8	0.044 (−0.864; 0.952)		.652
Relations between interacting participants				F(4, 0.43) = 0.425	.831	Qe(65) = 71.037	.284	0.027
Parent and child	16	42	0.121 (−0.008; 0.251)		.064
Romantic couple	2	13	−0.007 (−0.370; 0.357)		.851
Couple of familiar individuals	1	1	*
Dyad of strangers	2	2	−0.226 (−1.262; 0.810)		.221
Group of 3+ people except for parents with children	5	9	−0.010 (−0.479; 0.460)		.951
Hypnotist and a participant	3	5	0.088 (−0.734; 0.909)		.405
Psychotherapist and a participant	1	1	*
Social interaction
Types of interactions				F(6, 0.52) = 0.971	.719	Qe(65) = 55.491	.794	0.015
Playing	14	30	0.067 (−0.039; 0.174)		.192
Speaking	3	11	−0.029 (−0.228; 0.170)		.501
Tactile	4	14	0.282 (−0.560; 1.124)		.298
Speaking and Tactile	1	4	0.012 (−0.183; 0.207)		.731
Singing	2	5	−0.252 (−2.142; 1.638)		.339
Hypnosis	2	4	0.089 (−0.732; 0.910)		.399
Other	3	4	0.198 (−0.464; 0.861)		.259
Experiment location				F(2, 6.26) = 0.640	.558	Qe(65) = 75.345	.179	0.029
Laboratory condition	15	41	0.098 (0.009; 0.187)		.033
Home	4	7	−0.010 (−0.295; 0.275)		.910
Other	6	20	0.110 (−0.325; 0.546)		.533
Duration of interaction (min)	25	68	0.004 ( −0.003; 0.012)	F(1, 5.85) = 1.831	.226	Qe(66) = 62.881	.586	0.019
Pre-post interaction trend of assessed participants’ experience				F(2, 2.44) = 2.098	.295	Qe(23) = 25.484	.326	0.062
Positive	5	21	0.056 (−0.339; 0.452)		.711
Negative	1	1	0.010 (0.010; 0.010)		<.0001
No change/neutral	3	4	−0.023 (−0.205; 0.159)		.651
OXT collection and analysis
Time of day of OXT collection				F(1, 2.32) = 0.313	.625	Qe(59) = 73.193	.101	0.032
Afternoon and evening (after 12:00 p.m.)	17	56	0.079 (−0.046; 0.204)		.197
Morning and afternoon	3	5	0.021 (−0.365; 0.406)		.836
OXT collection time after the beginning of interaction (min)	23	64	0.001 ( −0.002; 0.003)	F(1, 3.84) = 0.377	.574	Qe(62) = 67.483	.295	0.028
OXT collection time after the end of interaction (min)	22	58	0.004 (−0.011; 0.018)	F(1, 2.95) = 0.668	.475	Qe(56) = 46.518	.813	0.010
Restrictions before an experiment (“no” means no restrictions):
Eating				F(1, 11.92) = 0.288	.601	Qe(70) = 81.930	.156	0.027
Yes	13	26	0.105 (−0.096; 0.305)		.274
No	17	46	0.061 (0.002; 0.121)		.043
Caffeine consumption				F(1, 15.78) = 0.077	.785	Qe(70) = 82.422	.147	0.026
Yes	12	19	0.064 (−0.046; 0.174)		.224
No	17	53	0.086 (−0.050; 0.222)		.198
Alcohol consumption				F(1, 8.3) = 0.990	.348	Qe(70) = 82.290	.149	0.026
Yes	8	15	0.023 (−0.091; 0.137)		.639
No	21	57	0.093 (−0.026; 0.213)		.117
Drinking any liquids				F(1, 3.66) = 1.809	.256	Qe(70) = 77.771	.245	0.022
Yes	4	8	−0.085 (−0.549; 0.380)		.582
No	23	64	0.109 (0.008; 0.210)		.036
Drinking (except for water)				F(1, 7.23) = 0.087	.776	Qe(70) = 82.138	.152	0.026
Yes	7	9	0.103 (−0.094; 0.300)		.243
No	22	63	0.075 (−0.038; 0.187)		.180
Smoking				F(1, 15.98) = 0.689	.419	Qe(70) = 82.284	.150	0.026
Yes	12	23	0.035 (−0.072; 0.143)		.481
No	17	49	0.100 (−0.038; 0.238)		.142
Doing exercise				F(1, 3.41) = 3.702	.139	Qe(70) = 75.419	.308	0.021
Yes	23	62	−0.125 (−0.532; 0.282)		.369
No	4	10	0.114 (0.013; 0.215)		.029
Restrictions on use of hormonal contraception, steroid medications, or other medications that would likely influence OXT concentrations				F(1, 1.15) = 0.398	.631	Qe(70) = 82.051	.154	0.026
Yes	2	6	0.003 (−1.519; 1.525)		.986
No	25	66	0.084 (−0.017; 0.186)		.099
For women who are breastfeeding: was biomaterial collected between at least 30 min after and 30 min before breastfeeding				F(1, 8.62) = 1.151	.313	Qe(34) = 38.015	.292	0.015
Yes	6	9	0.156 (−0.018; 0.330)		.069
No	15	27	0.055 (−0.091; 0.201)		.431
Presence of separation procedure				F(1, 18.55) = 1.234	.281	Qe(70) = 74.124	.345	0.023
Yes	16	46	0.034 (−0.070; 0.138)		.491
No	12	26	0.145 (−0.054; 0.343)		.138
Biological fluid collected				F(2, 5.96) = 2.039	.212	Qe(69) = 77.296	.231	0.026
Blood	7	24	−0.009 (−0.133; 0.115)		.859
Saliva	17	41	0.095 (−0.054; 0.245)		.194
Urine	4	7	0.151 (−0.054; 0.357)		.096
Techniques for samples’ collection				F(4, 4.58) = 1.292	.393	Qe(67) = 76.921	.191	0.029
Venipuncture	3	5	0.011 (−0.790; 0.812)		.952
Intravenous catheter/cannula	4	19	−0.022 (−0.131; 0.088)		.525
Absorbent device	10	24	0.111 (−0.131; 0.353)		.323
Passive drool method	8	17	0.069 (−0.020; 0.158)		.105
Urine miscellaneous collection	4	7	0.151 (−0.058; 0.359)		.100
Use of the concentration procedure				F(1, 16.08) = 4.466	.051	Qe(70) = 70.164	.472	0.020
Yes	16	41	0.149 (0.011; 0.287)		.037
No	11	331	−0.024 (−0.144; 0.095)		.649
Use of the extraction procedure				F(1, 15.61) = 0.005	.944	Qe(70) = 82.415	.147	0.026
Yes	10	21	0.082 (0.003; 0.162)		.044
No	17	51	0.077 (−0.069; 0.223)		.275
Methods for OXT analysis				F(2, 6.51) = 11.631	.007	Qe(69) = 72.864	.352	0.022
ELISA	21	52	0.118 (0.010; 0.227)		.035
RIA	5	19	−0.046 (−0.346; 0.257)		.667
HPLC	1	1	−0.143 (−0.143; −0.143)		<.0001
Study quality	27	72	−0.133 (−0.920; 0.654)	F(1, 11.83) = 0.136	.719	Qe(70) = 82.318	.149	0.026

Note. p-values < .05 are bolded.

Differences for none of the binary categorical moderators (like Yes/No) reached statistically significant level (p < .05) between “Yes” and “No.”

^*for this moderator, the cluster-robust Wald test could not be performed due to non-positive definite variance-covariance matrix; it was omitted from the analysis.

The assessment of subgroups of moderators generated several statistically significant effects (Table 4). Findings indicated that the mean ES for analyses of interactions between parent and child was marginally significantly different from zero (p = .064); its magnitude was small (g = 0.121 (95% CI −0.008 to 0.251)) and reflected the increase in the concentration of OXT after the interaction. Interestingly, in interactions between parent and child, parental (g = 0.096 (95% CI 0.011 to 0.180), p = .030), but not children’s (g = 0.182 (95% CI −0.127 to 0.491), p = .199) mean ESs were significantly higher than zero.

Having no restrictions on eating, drinking any liquids, or doing exercise prior to the interaction was characterized by mean ES significantly different from zero (0.061 (95% CI 0.002 to 0.121), 0.109 (95% CI 0.008 to 0.210), and 0.114 (95% CI 0.013 to 0.215), respectively), all of which could be regarded as small. The negative trend of assessed participants’ experience between the beginning and the end of interaction was associated with mean ES significantly higher than zero (0.010); still, the latter effect should be interpreted with caution due to the ES being represented by only one value. The mean ES for the use of the extraction procedure was small but significantly different from zero (0.082 (95% CI 0.003 to 0.162)).

The significant contribution of the baseline OXT concentration to social behavior, which will be presented later in the meta-analysis section for correlation studies, led us to check whether the baseline OXT concentration could have a moderating effect in the pre-post design. Yet, this moderator did not have a significant effect (F(1, 3.63) = 0.038, p = .857).

Regarding the analysis of the possible causes of heterogeneity among the published results, the inclusion of all moderators into the model was accompanied by residual heterogeneity that was no longer statistically significant (p > .05, Table 4), indicating that these factors may have caused the heterogeneity across the results.

Studies with Within-Group Between-Conditions (Control vs. Interaction) Design

Overall effect

In this study design, all OXT values were presented as OXT concentrations after the experimental procedure (control or social interaction). The overall mean ES of the effect of social interaction on OXT concentrations was g = 0.256 (95% CI −0.360 to 0.872) and was not significantly different from zero (p = .312 based on m = 5 studies, k = 8 ESs), indicating that social interaction did not alter OXT concentrations compared to conditions with no social interactions (Table 3 and Figure 8).

Fig. 8. Forest Plot for Within-Group Between-Conditions (Control vs. Interaction) Design.

Note. The overall mean ES of social interaction on OXT concentrations was not significantly different from zero (g = 0.256, p = .312). If there was more than one effect size in an article, decimal points after the article year indicate different effect sizes originating from the same article.

Publication bias

We did not observe any asymmetry in the funnel plot (Figure 9). The intercepts from Egger’s regression test did not significantly deviate from zero (b = 0.673, t(6) = −0.49, p = .640). The MLMA Egger test yielded nonsignificant slopes (b = −6.931, p = .812). This suggests that there was no indication of publication bias for studies with within-group between-conditions design.

Fig. 9.

Funnel Plot for Within-Group Between-Conditions Design

Heterogeneity in effect sizes

The results of the test for heterogeneity revealed significant variation between all ESs in the data set (Q(df = 7) = 46.061, p < .0001). The I² index was 85.38%, indicating large heterogeneity. Heterogeneity is regarded as substantial when less than 75% of the total amount of variance can be attributed to sampling variance (14.62%). Such a level of heterogeneity, once again, necessitates moderator analyses. Variance component estimates showed significant variation between studies (σ² = 0.216, χ² = 4.08, p = .044) but not between ESs within studies (σ² = 0.0, χ² = 0.0, p = 1.0).

Moderator analyses

Moderator analyses were conducted to identify possible characteristics of the study sample, social interaction, or OXT collection and analysis that could moderate the effect of social interaction on OXT concentrations. Table 5 presents the results of the moderators and the omnibus test statistic (p < .05). It also presents several characteristics that moderated the effect of social interaction on OXT concentrations. Even still, some effects should be interpreted with caution due to the small number of articles and the magnitude of ESs included in this analysis.

Table 5.

Results for Moderators for Within-Group Between-Conditions Design

Moderator variables	Number of studies (m)	Number of effect sizes (k)	β0, mean g (95% CI)	Omnibus test	p-value	Residual heterogeneity	p-value	τ2
Study sample
Age (years)	4	5	0.033 (−0.210; 0.277)	F(1, 1.33) = 0.970	.470	Qe(3) = 39.538	< .0001	0.339
Sex (% of females)	4	7	0.004 (−0.011; 0.018)	F(1, 1.01) = 9.477	.199	Qe(5) = 31.165	< .0001	0.243
Sample size	5	8	0.016 (−0.002; 0.034)	F(1, 1.43) = 33.697	.058	Qe(6) = 8.725	.190	0.029
Relations between interacting participants				F(1, 2.85) = 51.310	.007	Qe(6) = 4.465	.614	< .0001
Dyad of strangers	3	3	0.627 (0.392; 0.861)		.008
Group of 3+ people except for parents with children	2	5	−0.257 (−1.672; 1.158)		.261
Social interaction
Types of interactions				F(2, 0.97) = 59.117	.097	Qe(5) = 2.972	.704	< .0001
Trust-related	3	3	0.627 (0.392; 0.861)		.008
Singing	1	3	−0.136 (−0.137;−0.136)		.0001
Other	1	2	−0.354 (−0.877; 0.170)		.074
Experiment location				F(1, 1.92) = 211.986	.006	Qe(4) = 0.658	.956	< .0001
Laboratory condition	3	3	0.627 (0.392; 0.861)		.008
Other	1	3	−0.136 (−0.137; −0.136)		.0001
Duration of interaction (min)	3	6	0.003 (−0.039; 0.045)	F(1, 1.81) = 0.148	.741	Qe(4) = 28.322	< .0001	0.392
OXT collection and analysis
Time of day of OXT collection				F(1, 2.85) = 51.310	.007	Qe(6) = 4.465	.614	< .0001
Afternoon and evening (after 12:00 p.m.)	2	5	−0.257 (−1.672; 1.158)		.261
Morning and afternoon	3	3	0.627 (0.392; 0.861)		.008
OXT collection time after the beginning of interaction (min)	3	6	0.001 (−0.120; 0.121)	F(1, 1.11) = 0.002	0.975	Qe(4) = 29.014	< .0001	0.313
OXT collection time after the end of interaction (min)	3	6	−0.004 (−0.256; 0.247)	F(1, 1.02) = 0.040	.874	Qe(4) = 20.555	.0004	0.239
Restrictions before an experiment (“no” means no restrictions):
Eating				F(1, 2.99) = 3.663	.152	Qe(6) = 41.367	< .0001	0.233
Yes	1	3	−0.136 (−0.142; −0.131)		.002
No	4	5	0.360 (−0.466; 1.186)		.259
Caffeine consumption						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Alcohol consumption						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Drinking any liquids				F(1, 2.99) = 3.663	.152	Qe(6) = 41.367	< .0001	0.233
Yes	1	3	−0.136 (−0.142; −0.131)		.002
No	4	5	0.360 (−0.466; 1.186)		.259
Drinking (except for water)						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Smoking						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Doing exercise				F(1, 2.99) = 3.663	.152	Qe(6) = 41.367	< .0001	0.233
Yes	1	3	−0.136 (−0.142; −0.131)		.002
No	4	5	0.360 (−0.466; 1.186)		.259
Restrictions on use of hormonal contraception, steroid medications, or other medications that would likely influence OXT concentrations						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
For women who are breastfeeding: was biomaterial collected between at least 30 min after and 30 min before breastfeeding						Qe(6) = 31.405	< .0001	0.172
No	5	7	0.287 (−0.270; 0.844)		.226
Presence of separation procedure				F(1, 2.99) = 3.663	.152	Qe(6) = 41.367	< .0001	0.233
Yes	1	3	−0.136 (−0.142; −0.131)		.002
No	4	5	0.360 (−0.466; 1.186)		.259
Biological fluid collected				F(1, 2.85) = 51.310	.007	Qe(6) = 4.465	.614	< .0001
Blood	3	3	0.627 (0.392; 0.861)		.008
Saliva	2	5	−0.257 (−1.672; 1.158)		.261
Techniques for samples’ collection				F(2, 0.97) = 59.117	.097	Qe(5) = 2.972	.704	< .0001
Venipuncture	3	3	0.627 (0.392; 0.861)		.008
Absorbent device	1	3	−0.136 (−0.137;−0.136)		.0001
Passive drool method	1	2	−0.354 (−0.877; 0.170)		.074
Use of the concentration procedure						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Use of the extraction procedure						Qe(7) = 46.061	< .0001	0.216
No	5	8	0.256 (−0.360; 0.872)		.312
Methods for OXT analysis				F(1, 2.99) = 3.663	.152	Qe(6) = 41.367	< .0001	0.233
ELISA	4	5	0.360 (−0.466; 1.186)		.259
RIA	1	3	−0.136 (−0.142; −0.131)		.002
Study quality	5	8	−3.677 (−6.103; −1.252)	F(1, 2.62) = 27.516	.019	Qe(6) = 7.424	.283	0.016

Note. p-values < .05 are bolded.

Race/ethnicity (% of Whites) and breastfeeding status data were not available for all effect sizes.

Differences for none of the binary categorical moderators (like Yes/No) reached statistically level significance (p < .05) between “Yes” and “No.”

Study quality was a significant moderator, as the omnibus test was significant (F (1, 2.62) = 27.516, p = .019), and the regression coefficient was significant (−3.677; t(6) = −5.246, p = .019). The negative regression coefficient implies that the studies with higher quality have lower reported effects.

The relationships between the interacting participants moderated the effect of social interaction on OXT concentrations (F(1, 2.85) = 51.310, p = .007). The mean ES was significantly different from zero in couples of strangers and could be regarded as high (g = 0.627 (95% CI 0.392 to 0.861)). This reflects that higher OXT concentrations were revealed in the presence of social interaction compared with its absence. All other statistically significant moderators were directly related to the division based on relations between interacting participants: for example, membership in a dyad of strangers, but not in a group of 3+ people, was exclusively associated with a trust-related type of interaction (Table 2), laboratory condition of the experiment, morning and afternoon time of OXT collection, blood source of OXT, and venipunctures used for its collection. All these subgroups were characterized by large mean ESs (>0.62) that were significantly different from zero. In contrast, membership in a group of 3+ people, but not in a dyad of strangers, was exclusively associated with a non-trust-related type of interaction (choir singing and group cooking), non-laboratory condition of the experiment, the afternoon and evening (after 12:00 p.m.) time of OXT collection, saliva source of OXT, and absorbent device or passive drool method used for its collection, all characterized by small to medium negative mean ES, part of which was statistically different from zero. Therefore, group membership was revealed to be the most significant moderator between OXT concentrations and social interaction. Having restrictions on eating, drinking any liquids, doing exercise prior to the interaction, or the presence of separation procedure was characterized by mean ES significantly different from zero (all −0.136); nonetheless, the latter effect should be interpreted with caution because ESs were derived from one study.

Regarding the analysis of the possible causes of heterogeneity across the study results, the inclusion of several moderators into the model was accompanied by residual heterogeneity that was no longer statistically significant (p > .05, Table 5). Sample size, relationships between interacting participants, types of interactions, experiment location, time of day of OXT collection, biological fluid collected, techniques for sample collection, and study quality are among the factors that may represent causes of heterogeneity across study results.

Correlational Design

Overall effect

The overall mean ES of the correlations between measures of social constructs and OXT concentrations was z = 0.137 (95% CI 0.057 to 0.218) and was significantly different from zero (p = .002 for m = 31 studies, k = 143 ESs), indicating a positive correlation (Table 5 and Figure 10). Fisher’s z of 0.137 is equal to a correlation coefficient r of .136 and could be considered small to medium.

Fig. 10. Forest Plot for Correlational Design.

Note. The overall mean ES of the correlations between measures of social constructs and OXT concentrations was significantly and positively different from zero (z = 0.137, p = .002). If there was more than one ES in an article, decimal points after the article year indicate different effect sizes originating from the same article.

Publication bias

We did not observe asymmetry in the funnel plot (Figure 11). The intercepts from Egger’s regression test did not significantly deviate from zero (b = 0.020, t(141) = 1.683, p = .095). The MLMA Egger test yielded nonsignificant slopes (b = −0.034, p = .519). This suggests that there was no indication of publication bias for studies with correlational design.

Fig. 11.

Funnel Plot for Correlational Design

Heterogeneity in effect sizes

The results of the test for heterogeneity revealed significant variation between all ESs in the data set (Q(df = 142) = 488.479, p < .0001). The I² index was 80.88%, indicating large heterogeneity. Heterogeneity is considered substantial because less than 75% of the total amount of variance can be attributed to sampling variance (19.12%), making it relevant to perform moderator analyses. Variance component estimates showed significant variation between studies (σ² = 0.035, χ² = 33.95, p < 0.0001), as well as between ESs within studies (σ² = 0.014, χ² = 33.52, p < 0.0001).

Moderator analyses

Moderator analyses were conducted to identify possible characteristics of the study sample, social interaction, or OXT collection and analysis that could moderate the correlations between the measures of social interaction and OXT concentrations. Table 6 presents the results of the moderators and the omnibus test statistic (p < .05). From Table 6, it is evident that none of the considered characteristics moderated the effect of social interaction on OXT concentrations.

Table 6.

Results for Moderators for Correlational Design

Moderator variables	Number of studies (m)	Number of effect sizes (k)	β0, mean g (95% CI)	Omnibus test	p-value	Residual heterogeneity	p-value	τ2
Study sample
Age (years)	27	133	0.004 (−0.039; 0.048)	F(1, 1.86) = 0.219	.689	Qe(131) = 439.477	< .0001	0.042
Sex (% of Females)	28	125	−0.0003 (−0.002; 0.001)	F(1, 4.81) = 0.465	.527	Qe(123) = 384.210	< .0001	0.040
Race/ethnicity (% of Whites)	10	31	−0.007 (−0.020; 0.006)	F(1, 4.02) = 2.123	.218	Qe(29) = 131.279	< .0001	0.052
Sample size	31	143	−0.0003 (−0.002; 0.001)	F(1, 5.3) = 0.520	.501	Qe(141) = 465.518	< .0001	0.037
Breastfeeding status				F(1, 2.04) = 1.718	.319	Qe(93) = 337.690	< .0001	0.039
Breastfeeding	16	69	0.097 (−0.012; 0.206)		.079
Not breastfeeding	7	26	0.144 (0.015; 0.274)		.036
Relations between interacting participants				F(3, 1.08) = 1.061	.587	Qe(139) = 458.437	< .0001	0.037
Parent and child	22	105	0.116 (0.010; 0.223)		.033
Romantic couple	5	20	0.158 (−0.038; 0.354)		.088
Dyad of strangers	2	6	0.326 (−0.343; 0.996)		.102
Hypnotist and a participant	2	12	0.139 (−1.450; 1.729)		.466
Social interaction
Types of interactions				F(3, 1.53) = 0.102	.950	Qe(133) = 438.593	< .0001	0.041
Playing	21	103	0.127 (0.017; 0.238)		.026
Speaking	3	12	0.178 (0.032; 0.324)		.035
Tactile	1	2	*
Speaking and Tactile	3	10	0.164 (−0.421; 0.749)		.349
Hypnosis	2	12	0.139 (−1.450; 1.729)		.466
Trust-related	1	4	*
Valence of interaction				F(3, 1.11) = 12.034	.185	Qe(139) = 451.843	< .0001	0.033
Positive	28	102	0.148 (0.065; 0.230)		.001
Negative	8	29	0.073 (−0.049; 0.195)		.203
Synchrony	4	8	0.298 (0.035; 0.562)		.040
Unclear	1	4	−0.106 (−0.312; 0.101)		.129
Approach for assessment of social constructs				F(2, 3.38) = 1.557	.332	Qe(140) = 488.440	< .0001	0.043
Behavioral analysis	22	96	0.167 (0.061; 0.274)		.004
Questionnaires or interviews about current interaction	5	24	0.156 (−0.138; 0.449)		.206
Questionnaires or interviews about general traits associated with social behavior aspects	7	23	0.045 (−0.122; 0.211)		.526
Type of correlation				F(1, 1.31) = 1.079	.453	Qe(140) = 466.103	< .0001	0.033
Cross-sectional	29	129	0.123 (0.042; 0.204)		.004
Longitudinal	3	13	0.154 (−0.021; 0.328)		.062
Experiment location				F(2, 1.67) = 0.038	.964	Qe(138) = 472.640	< .0001	0.039
Laboratory condition	20	94	0.130 (0.021; 0.239)		.022
Home	9	37	0.141 (−0.027; 0.309)		.090
Other	2	10	0.162 (−0.156; 0.481)		.170
Duration of interaction (min)	26	127	0.003 (−0.009; 0.014)	F(1, 2.75) = 0.668	.479	Qe(125) = 411.358	< .0001	0.036
OXT collection and analysis
OXT parameter				F(4, 1.05) = 1.773	.497	Qe(129) = 413.923	< .0001	0.039
Baseline OXT	20	88	0.152 (0.056; 0.248)		.004
OXT change	5	18	0.209 (0.031; 0.388)		.033
Log (post-interaction OXT) minus log (baseline OXT)	2	10	−0.151 (−0.747; 0.446)		.192
AUC	2	5	0.321 (−1.976; 2.619)		.387
Post-interaction OXT	6	13	0.114 (−0.204; 0.432)		.398
Composite averaged from baseline and post-interaction assessments	1	1	*
24 h cumulative	1	6	*
Time of day of OXT collection				F(1, 8.17) = 0.407	.541	Qe(107) = 374.595	< .0001	0.024
Afternoon and evening (after 12:00 p.m.)	16	77	0.131 (0.025; 0.236)		.019
Morning and afternoon	6	32	0.187 (−0.008; 0.382)		.056
OXT collection time after the beginning of interaction (min)	19	94	0.001 (−0.007; 0.009)	F(1, 4.18) = 0.224	.660	Qe(92) = 353.690	< .0001	0.048
OXT collection time after the end of interaction (min)	11	37	−0.009 (−0.054; 0.036)	F(1, 1.43) = 1.650	.368	Qe(35) = 162.647	< .0001	0.068
Restrictions before an experiment (“no” means no restrictions):
Eating				F(1, 6.91) = 0.039	.848	Qe(141) = 487.221	< .0001	0.036
Yes	10	41	0.126 (−0.070; 0.321)		.174
No	23	102	0.142 (0.067; 0.218)		.0008
Caffeine consumption				F(1, 3.8) = 0.063	.815	Qe(141) = 485.039	< .0001	0.037
Yes	4	18	0.104 (−0.368; 0.576)		.530
No	27	125	0.143 (0.058; 0.227)		.002
Alcohol consumption				F(1, 2.38) = 0.665	.489	Qe(141) = 483.856	< .0001	0.036
Yes	28	129	0.234 (−0.293; 0.762)		.196
No	3	14	0.129 (0.044; 0.214)		.005
Drinking any liquids						Qe(142) = 488.479	< .0001	0.035
No	31	143	0.137 (0.057; 0.218)		.002
Drinking (except for water)				F(1, 4.65) = 0.002	.964	Qe(141) = 466.878	< .0001	0.038
Yes	5	23	0.151 (−0.610; 0.912)		.609
No	27	120	0.136 (0.024; 0.249)		.020
Smoking				F(1, 3.8) = 0.063	.815	Qe(141) = 485.039	< .0001	0.037
Yes	4	18	0.104 (−0.368; 0.576)		.530
No	27	125	0.143 (0.058; 0.227)		.002
Doing exercise						Qe(142) = 488.479	< .0001	0.035
No	31	143	0.137 (0.057; 0.218)		.002
Restrictions on use of hormonal contraception, steroid medications, or other medications that would likely influence OXT concentrations				F(1, 2.4) = 0.060	.826	Qe(141) = 480.830	< .0001	0.037
Yes	3	12	0.156 (−0.139; 0.451)		.149
No	28	131	0.136 (0.047; 0.225)		.004
For women who are breastfeeding: was biomaterial collected between at least 30 min after and 30 min before breastfeeding				F(1, 18.18) = 0.337	.569	Qe(84) = 343.142	< .0001	0.038
Yes	12	51	0.106 (−0.062; 0.274)		.192
No	10	35	0.161 (0.033; 0.289)		.020
Presence of separation procedure				F(1, 17.96) = 0.013	.912	Qe(141) = 482.488	< .0001	0.037
Yes	12	64	0.134 (−0.001; 0.268)		.052
No	20	79	0.141 (0.048; 0.234)		.005
Biological fluid collected				F(2, 1.48) = 3.664	.267	Qe(140) = 470.636	< .0001	0.029
Blood	19	74	0.195 (0.121; 0.269)		< .0001
Saliva	12	56	0.028 (−0.096; 0.153)		.618
Urine	3	13	0.172 (−0.146; 0.490)		.135
Techniques for samples’ collection				F(4, 1) = 1.517	.538	Qe(133) = 463.113	< .0001	0.032
Venipuncture	13	57	0.206 (0.119; 0.294)		.0003
Intravenous catheter/cannula	4	12	0.180 (−0.174; 0.534)		.201
Absorbent device	10	50	0.041 (−0.102; 0.183)		.529
Passive drool method	2	6	−0.049 (−0.810; 0.711)		.562
Urine miscellaneous collection	3	13	0.180 (−0.115; 0.475)		.117
Use of the concentration procedure				F(1, 3.69) = 0.502	.521	Qe(141) = 476.911	< .0001	0.038
Yes	11	51	0.109 (−0.021; 0.239)		.088
No	22	92	0.152 (0.056; 0.249)		.004
Use of the extraction procedure				F(1, 3.05) = 3.436	.159	Qe(141) = 465.128	< .0001	0.032
Yes	8	33	0.205 (0.091; 0.319)		.006
No	24	110	0.116 (0.029; 0.204)		.012
Methods for OXT analysis				F(1, 3.81) = 0.186	.689	Qe(141) = 487.500	< .0001	0.037
ELISA	27	131	0.132 (0.043; 0.220)		.005
RIA	4	12	0.182 (−0.167; 0.530)		.193
Study quality	31	143	0.211 (−0.121; 0.543)	F(1, 12.57) = 1.893	.193	Qe(141) = 477.280	< .0001	0.034

Note. p-values < .05 are bolded.

Differences for none of the binary categorical moderators (like Yes/No) reached statistically level significance (p < .05) between “Yes” and “No.”

^*for these moderators, the cluster-robust Wald test could not be performed due to non-positive definite variance-covariance matrix; they were omitted from the analysis.

The assessment of subgroups of moderators generated several statistically significant effects (Table 6). Among types of participants, the mean ES for parent and child (z = 0.116 (95% CI 0.010 to 0.223)) was significantly larger than zero and could be regarded as low. Interestingly, in interactions between parent and child, parental (z = 0.136 (95% CI 0.021 to 0.251), p = .023), but not children’s (z = −0.022 (95% CI −1.338 to 1.295), p = .951) mean ESs were significantly higher than zero. Among types of interaction, the two types that significantly differed from zero were playing, with a positive small to medium ES (z = 0.127 (95% CI 0.017 to 0.238)) and speaking, with a positive small to medium ES (z = 0.178 (95% CI 0.032 to 0.324)). Positive interactions (z = 0.148 (95% CI 0.065 to 0.230)) and interactions with synchrony (z = 0.298 (95% CI 0.035 to 0.562)) were significantly larger than zero, i.e., positively correlated with OXT concentrations with medium ES. The only approach for the assessment of social constructs that had a statistically significant effect was behavioral analysis (z = 0.167 (95% CI 0.061 to 0.274)). The only experiment location with a statistically significant effect was the laboratory condition (z = 0.130 (95% CI 0.021 to 0.239)). Among the OXT parameters used, baseline OXT (z = 0.152 (95% CI 0.056 to 0.248)) and OXT change (z = 0.209 (95% CI 0.031 to 0.388)) statistically significantly differed from zero. The afternoon and evening (after 12:00 p.m.) time of OXT collection was associated with significantly different from zero ES (z = 0.131 (95% CI 0.025 to 0.236)). In participants who had no restrictions on eating, drinking any liquids (including caffeine and alcohol consumption), smoking, doing exercise, use of medicines (hormonal contraception, steroid medications, or other medications that would likely influence OXT concentrations), breastfeeding before experiments, or who had no separation procedure before the experiment, mean ES was positive and differed statistically significantly from zero. A cross-sectional design, negative breastfeeding status, blood collection, venipuncture, absence of concentration procedure, both presence and absence of extraction procedure, and ELISA method of analyses all had positive mean ES that differed statistically significantly from zero.

Regarding the analysis of the possible causes of heterogeneity among study results, the inclusion of moderators did not eliminate any residual heterogeneity, which indicates that none of these moderators were likely the source of it.

Discussion

This meta-analysis is the first comprehensive review of research on relationships between OXT concentration and human social interactions. Its objective was to describe the included studies in terms of characteristics of study design, study sample, social interaction, and procedures for OXT collection and analysis, to measure the overall effect, and to investigate potential moderators of ESs. Our search was not limited in terms of the language of publication, the country where the study was performed, the type of publication (theses and dissertations were included), or the publication date. Furthermore, we contacted numerous authors to obtain additional information when it was missing from their articles. As a result, we can safely assume that these data accurately reflect the state of research in this field. We revealed that despite increasing in number over the past few decades, studies investigating this subject are still rather far from providing a clear picture of the association of interest.

Studies with Pre-Post Design

In pre-post design, we have observed that social interaction does not significantly change OXT concentrations following interaction (Table 3). In the psychology literature, OXT is widely referred to as an endocrinal indicator of social bonding and affiliation (Feldman, 2012; Insel, 1992); therefore, the finding was surprising given the current theoretical framework and efforts made by the scientific community to study the relationships between OXT concentrations and social interactions. Such a result can be explained by the high level of heterogeneity among the studies included in this analysis. As we mentioned before, there is no gold standard paradigm for measuring social interactions, so a variety of protocols and designs were included in the pool of analyzed articles. However, it is noteworthy that the diversity in several aspects was not evenly distributed (Table 1). For instance, we identified seven categories of interacting participants, yet most of the studies were conducted with parents and children. Similarly, out of eight experimental paradigms, most of the studies used play as an interaction procedure. Also, a significant asymmetry was observed in terms of the age of participants: there were more adults than children; more females than males; and more Whites than any other races or ethnicities combined.

The results of the test for heterogeneity revealed significant variation between the ESs in the data sets. To test whether moderating factors could explain heterogeneity, we performed moderator analyses with three major groups of variables related to the study sample, social interaction, and OXT collection and analysis, respectively (Table 4). We found that all moderators explained heterogeneity across the results. Surprisingly, the characteristics of the study sample had no significant moderating effects between social interaction and OXT concentrations. These findings contradict the literature, which has suggested the possible contribution of age, sex, and race in the modulation of social behavior by OXT (Torres et al., 2018). A meta-analysis devoted to confounders showed that endogenous OXT concentrations increased significantly with increasing percentages of females and the mean age of participants (Engel et al., 2019). The cited study included only adult participants and was focused on baseline endogenous OXT, and the presence of social interaction was not among the inclusion criteria of the study, making it difficult to compare previous results to our sample directly.

Notably, the mean ES in a specific subgroup was marginally significantly larger than zero, specifically in parent/child dyads (Table 4), and statistically significantly larger than zero in parents from parent/child dyads. This finding could be explained by the attachment between parents and children, which is the most fundamental and essential social bond (Hirschi, 1969; Wiatrowski et al., 1981). Interestingly, participants’ appraisal of their experience related to social interaction did not influence the mean ES, except for the data on negative experience, which can hardly be relied upon because they were based on only one ES value. The reason for the absence of the impact of positive experience may be the non-objectivity of these data because they were self-reported, and there was no available data for the dynamics of participants’ objective behavioral characteristics in response to social interaction.

Although the factors related to the OXT collection procedure did not moderate the effect of social interaction on OXT concentrations, the absence of several restrictions (eating, drinking any liquids, or exercising) was accompanied by mean ES statistically significantly higher than zero (Table 4). At the same time, the presence of these restrictions was not accompanied by mean effects that were significantly different from zero. This may indicate the stimulating impact of these procedures on the concentrations of OXT in addition to the effect of the social interaction per se. Nevertheless, these effects should be interpreted with caution because, within the studies where the restrictions were stated, it is impossible to determine how closely the participants followed these rules. Furthermore, in those studies where the restrictions were not listed, it is difficult to ascertain which restrictions may have been followed by participants voluntarily by chance. In the abovementioned study on confounders of endogenous OXT concentrations (Engel et al., 2019), instructions to refrain from eating, drinking any liquid, caffeine or alcohol consumption, exercising, and fasting did not have a significant effect. Again, this discrepancy can be explained by methodological differences between the two meta-analyses.

Regarding the methods and procedures for measuring the concentrations of OXT, the ELISA, but not the RIA method, has yielded significant effects (Table 4). This information can be interpreted as ELISA being more sensitive to detecting the relationship between OXT concentration and social interaction. Still, it should be noted that in our sample for the meta-analysis of pre-post design studies, ELISA was more frequently accompanied by an extraction procedure than the RIA method, as indicated by both the number of studies and the number of sample sizes. This was the opposite of what we found in the whole sample of 63 studies, as well as in the subsample of studies for the meta-analysis of correlational design. Therefore, the observed result may be attributed to the extraction procedure, which could potentially lead to a substantial enrichment and concentration of analytes, resulting in increased precision and reduced matrix interference (Algoe et al., 2017). Consistent with this is the fact that the presence, but not the absence of the extraction procedure had a small but significantly different from zero ES (Table 4). It should be mentioned that a separate meta-analysis did not find any evidence to suggest that extraction had an effect on the outcome (Valstad et al., 2017).

Previously, significant disagreements were shown among methods used to measure OXT (blood plasma and central): RIA with extraction, ELISA with/without extraction, and ELISA with filtration (Lefevre et al., 2017). As the authors noted, this absence of agreement between methods could be expected for OXT plasma concentration given the influence of any pre-processing procedure (none, filtration, or extraction). However, it is surprising that central OXT concentrations, where there are supposedly no proteins interfering with OXT, were also unrelated between these methods. These discrepancies could probably be explained by three different specific antibodies that could potentially vary in the epitopes they recognize, leading to variable results between immunoassays. This raises the necessity for the standardization and characterization of the supplies used for OXT analysis in more detail. Furthermore, as MacLean et al. (2019) noted, “…discrepancies in data generated by different methods of measurement are not necessarily an indicator that some methods are valid whereas others are not. Rather, we propose that current challenges in the measurement of oxytocin may be analogous to the parable of the blind men and the elephant, with different methods of sample preparation and measurement being sensitive to different states in which the oxytocin molecule can exist” (p. 225). In conclusion, according to our analysis, the ELISA method with extraction procedure could be more promising in detecting the relationship between OXT concentration and social interaction, and the data on RIA need to be treated with caution as they come from only 5 studies.

OXT is a biomarker primarily synthesized in the brain, which has both benefits and drawbacks for analysis. On the one hand, this means that, unlike other biomarkers of behavior (cortisol, testosterone, estrogen, or progesterone), which originate primarily from tissue outside of the nervous system and hence could represent only indirect measures of brain function, OXT is synthesized in the nervous system and is more likely to be directly related to behavior (Carter et al., 2007). On the other hand, the question of correspondence between central and peripheral OXT concentrations is highly debated (Valstad et al., 2017). Theoretically, after its synthesis in the brain, OXT can be detected in biological fluids (blood, saliva, and urine) in an amount directly proportional to its concentrations in the brain; thus, this peripheral amount could reflect a causal mechanistic relationship with the processes occurring in the brain and, accordingly, with behavior. Practically, central and peripheral OXT concentrations might be correlated under specific stress conditions, which vary with the method used (Lefevre et al., 2017). The data from the pre-post design presented in this study represent another argument for the lack of connection between OXT responses and social interaction: a clear peak-like relationship between the concentration of OXT and the time of its collection with respect to the interaction (Figure 5) was not shown in any of the studied biological fluids. Nevertheless, regarding OXT metabolic clearance, only 10% of infused synthetic OXT was shown to be metabolized in nonpregnant women’s plasma in two hours (Takeda et al., 1989), which could explain a lack of decrease in OXT concentrations after social interactions.

Studies with Within-Group Between-Conditions (Control vs. Interaction) Design

In the studies with within-group between-condition design, participants were consecutively involved in social interaction and a control condition; ultimately, social interaction generally did not lead to changes in the OXT concentrations (Table 3). Although several moderators impacted the main effect (Table 5), all of them were interrelated, i.e., clustered within two independent sub-cohorts of participants, making it difficult to determine the primary and secondary factors that influenced the concentration of OXT. It remains unclear, for example, in couples of strangers, whether the morning time of OXT collection, their trust-related type of interaction, or blood source of OXT determined the large mean ES because no other time of OXT collection or type of social interaction was available for this subgroup of participants. These uncertainties, coupled with the small number of articles for this design type, indicate that the result should be interpreted with caution.

Correlational Design

In contrast to the two causal designs described above, the correlational design showed a significant positive main effect between OXT concentration and social interaction (Table 3). We performed moderator analyses with three major groups of variables related to the study sample, social interaction, and OXT collection and analysis, respectively, in order to explain the significant level of heterogeneity found (Table 6). The inclusion of any moderators in search of potential causes of heterogeneity among the study results did not lead to the elimination of residual heterogeneity, which indicates that none of these moderators taken separately can potentially be a source of it.

We showed that correlations between OXT concentrations and positive social interactions and interactions with synchrony had the largest and positive mean effects that differed significantly from zero (Table 6). Regardless, for a negative valence of interactions, there were no significant differences from zero, raising the question regarding the specificity of the OXT concentration measure concerning the valence of interaction to be answered in the future.

The mean ES for correlational studies was significantly larger than zero for parents with children, similar to this trend in pre-post interaction design (Table 6), and again mean ES was statistically significantly larger than zero in parents from parent/child dyads. The two types of interaction that significantly differed from zero were speaking and playing, the latter of which is related to the result described above for parents with children: in correlational design, playing was studied only in this group of participants (Table 2). Among approaches for the assessment of social constructs, the maximal, and only statistically robust method was behavioral analysis. Therefore, this reflects that behavioral analysis allows for obtaining the most objective and reliable data.

Among all OXT parameters correlated with indicators of social interaction, correlations with baseline OXT had significantly different from zero mean effects (Table 6). It has been suggested that baseline OXT “may be considered a biomarker of certain personality traits related to social-affiliative functioning, such as affiliative tendency, empathy, social fitness, social engagement, or reciprocal relational style” (Zilcha-Mano et al., 2021, p.527). This may be partly due to the fact that the baseline OXT is individually stable over months and even years in mothers, fathers, infants, and children (Feldman et al., 2013; Feldman et al., 2007; Schneiderman et al., 2012). Interestingly, correlations with the OXT change were also statistically significant, which suggests that, at least in correlational design, the dynamics of change in OXT concentrations during social interactions are of importance.

The absence of numerous restrictions before an experiment consistently contributed to the association between OXT concentration and social interaction (Table 6): the absence of restrictions on eating, caffeine consumption, alcohol consumption, drinking any liquids, drinking (except for water), smoking, doing exercise, breastfeeding, use of medicines, and on interactions between participants before the beginning of experiments (i.e., the lack of a separation procedure) was accompanied by mean ES statistically significantly different from zero (higher in all cases). Compared to causal designs, the absence of a larger number of restrictions in correlational design was accompanied by significantly higher than zero ES. At the same time, the presence of these restrictions was not associated with mean effects that were significantly different from zero. This may indicate an impact of these procedures on concentrations of OXT in addition to the effects of social interaction.

In a correlational design, only the ELISA method provided a positive ES that differed significantly from zero (Table 6), similar to the findings for the pre-post interaction design (Table 4). As previously mentioned for the pre-post interaction design, this result suggests that the ELISA method is more sensitive in detecting the relationship between OXT concentration and social interaction. Though, unlike the pre-post interaction design, it appears that the presence of the extraction procedure does not play a significant role since both the presence and absence of the extraction procedure had a positive ES that differed significantly from zero, as shown in Table 6. Additionally, unlike the sample used for the meta-analysis of pre-post design studies, the subsample of studies used for the meta-analysis of the correlational design showed that the RIA method is more commonly accompanied by an extraction procedure than the ELISA method; still, even with this extraction procedure, the RIA method still did not show a positive ES that differed significantly from zero.

Among the biological fluids studied, blood collection had the highest positive mean ES, significantly different from zero (Table 6). On the one hand, blood is the most reliable source given that OXT is directly released into the bloodstream from the neurohypophysis; nonetheless, blood proteins readily sequester OXT, leading to challenges in detection (MacLean et al., 2019). On the other hand, OXT’s mode of entry into saliva remains poorly understood, though it presents a relatively clean matrix, limiting opportunities for binding to large proteins (MacLean et al., 2019). Considering that we found a relationship between OXT concentrations and social interaction in correlational design, blood collection seems more suitable for detecting the relationship between OXT concentrations and social interaction compared to other biological fluids. Yet, as Lefevre et al. (2017) mentioned, correlating peripheral OXT with behavioral scales seems to be a suboptimal method to investigate the relationship between the OXT system and behavior because peripheral OXT might be a noisy proxy of central OXT. Finally, it could be useful to compare the data from this meta-analysis with similar meta-analyses with data generated from animal studies. It would allow us to better understand whether the absence of effects in causal designs is species-specific or fundamental. Unfortunately, no such meta-analysis is currently available.

Limitations and Directions for Future Research

This study had several limitations. First, a marked heterogeneity across studies was observed. Many sources of heterogeneity were revealed in non-correlational designs, but for correlational design, subgroup analyses were unsuccessful in identifying the sources of heterogeneity. In order to address this limitation, random-effects models were used in all the analyses. Nevertheless, there may be other important moderators that we did not consider, such as genetic and epigenetic factors, because they both have been associated with social functioning (Kumsta et al., 2013; Parker et al., 2014). Yet, none of the included studies analyzed epigenetic regulation of the oxytocinergic signaling, and only three investigated genetic single nucleotide polymorphism (SNP) variability of the OXT receptor gene, OXTR (Baião et al., 2019; Feldman et al., 2013; Feldman et al., 2012), which was insufficient for analysis in our work. Second, a limited number of studies were devoted to participants and types of social interaction other than parents playing with children. It would be important to consider more studies that use other types of social interaction. Third, the use of correlational design in more than half of the studies included in this review limited the possibility of establishing the presence, direction, and magnitude of any causal relationship between OXT concentrations and social interactions. Finally, most of the results included in our analyses were obtained using immunoassays (ELISA and RIA) and none using mass spectrometry. Considering that these techniques detect analytes in fundamentally different ways and are characterized by different strengths and limitations (MacLean et al., 2019), it would be useful to compare them in a meta-analytic framework.

Conclusions

In summation, this review provides only partial support for the link between social interaction and peripheral OXT concentrations, which are assumed to serve as surrogates of central nervous system OXT. The range of approaches for the study of hormonal regulation of human social interaction and the role of OXT in it is excessively wide, as wide as the range of associated methodological discrepancies. Hence, it is important to prioritize the development of standardized and reliable assessment models for social interactions, like the Trier Social Stress Test used in studies of the hormonal stress response. Moreover, it should be noted that social interactions may be more complex and less amenable to standardization than acute psychosocial stress and therefore require more extensive and nuanced assessments.

Perhaps one of these models may be based on interactions between parents and infants because this form of interaction showed one of the strongest and marginally statistically significantly different from zero ES in a pre-post design and statistically significantly different from zero ES in a correlational design. Also, it is seemingly important for this model to account for the behavioral and psychological aspects of social interaction in parallel to hormonal data. Future research should continue to refine these behavioral measures.

In standardized assessment models expected in the future, the presence of a separation procedure should be carefully considered to avoid overestimation of OXT concentrations during social interactions. Also, because several factors not directly related to social interaction, such as eating, drinking liquids, or doing exercise, have been shown to moderate the relationship between OXT concentrations and social interaction, they should be controlled more comprehensively.

Particular attention should be focused on the physiological and molecular aspects of OXT. Due to the invasive nature of the collection process, it is unlikely that the measurement of central OXT concentration will become widely adopted. As a result, it is essential to make efforts to characterize the role of peripheral OXT in its various forms, such as free, bound, or degraded, in the context of social interactions. In addition, it would also be useful to distinguish the source of peripheral OXT: whether it is the central nervous system or peripheral organs. As we mentioned earlier, various peripheral organs have been reported to produce OXT (uterus, testis, heart, and others), so these OXT sources could also contribute to the OXT concentration in social interaction settings, but the specific contribution from each of the sources has not yet been studied. The development of a model that considers all current pitfalls would require compiling the scattered efforts being made to elucidate neurochemical mechanisms of sociality in humans.

Public Significance Statement.

This comprehensive review and meta-analysis of studies on the relationships between endogenous oxytocin concentration and human social interactions highlight a lack of convergence in current research findings and consensus in their interpretation. Whereas social interaction alone does not consistently trigger the anticipated hormonal responses, correlations between social interaction indicators and oxytocin concentration demonstrate the presence of an association. These findings underscore the need for standardized and reliable approaches to research the neurochemical mechanisms of sociality, enabling advancements in understanding its texture, disserting the etiology of related disorders, and promoting healthy social interactions.

Appendix

Overview of the Studies with Pre-Post Design

Reference	Data included in meta- analysis, coordinates in forest plots	Hedges’ g	Mean baseline oxytocin concentration (pg/ml unless otherwise stated)	N of OXT donors	Sex (% of females)	Age [range], M±SD or SE	Ethnicity (%)	Interaction type	Participants type	Duration of interaction	OXT collection time in relation to the beginning of interaction	OXT source	OXT analysis method	Immunoassay kit manufacturer	Sample extraction utilized?	Results	Additional description of the sample/design
Abraham et al. (2014)		NI	NI	20	100	34.05±4.54 y (M±SD)	NI	P	P+C	NI	0 min; +x min	S	ELISA**	Enzo	No	NI	heterosexual primary-caregiving mothers
		NI	NI	21	0	35.0±2.58 y (M±SD)										NI	heterosexual secondary-caregiving fathers
		NI	NI	23		38.5±3.17 y (M±SD)										NI	homosexual primary-caregiving fathers
		NI	NI			36.52±5.47 y (M±SD)										NI	homosexual primary-caregiving fathers
Baião et al. (2019)	Baião (2019).1	0.073	22.99	38	NI	*[40-76 m], 57.70±7.25 m (M±SD)	W(100)	P	P+C	15	0 min; +15 min	S	ELISA	Enzo	Yes - solid phase (Strata-X)	NI	low scores maternal behavior
	Baião (2019).2	−0.070	23.54	50												NI	high scores maternal behavior
Bellosta-Batalla et al. (2020)	Bellosta-Batalla (2020)	−0.064	144.38	26	NI	*[20-50 y]	NI	P	G	90	−10 min; +100 min	S	ELISA	Arbor Assays	No	0
Bick et al. (2013)	Bick (2013).1	0.146	13.54 pg/ml adjusted for creatinine	41	100	*[28-68 y], 42.1±10.1 y (M±SD)	W(47), B(46), H(7)	P	P+C	40	0 min; +30 min	U	ELISA	Assay Design	Yes - solid phase (Sep-Pak C18)	0	first assessment (2 months of the relationship)
	Bick (2013).2	0.238	16.04 pg/ml adjusted for creatinine	32												0	second assessment (3 months after the first assessment)
Brondino et al. (2017)	Brondino (2017).1	1.501§	6.75	11	100	21±1.26 y (M±SD)	NI	S	F1+F2	15	0 min; +15 min	S	ELISA	Enzo	No	+	gossip conversation
	Brondino (2017).2	0.127	7.92			20.64±0.92 y (M±SD)										0	non-gossip emotional conversation
	Brondino (2017).3	−0.306	6.52	22	100	20.82±1.09 y (M±SD)			S1+S2							0	non-gossip neutral conversation
Christensen et al. (2014)		NI	NI	NI	NI	*28.0±7 y (M±NI)	NI	TR	F1+F2	10	−5 min; +12 min	P	RIA	Bachem	Yes - solid phase (Oasis HLB)	NI	family members
		NI	NI						S1+S2							NI	unfamiliar partners
Cong et al. (2015)	Cong (2015).1	0.062	50.49	26	100	31.5±6.8 y (M±SD)	W(76.9), B(19.2), H(3.8), A(26.9), NH(73.1)	T	P+C	30	−10 min; +25-30 min	S	ELISA	Salimetrics	No	NI
	Cong (2015).2	0.009	50.49								−10 min; +55-60 min					NI
	Cong (2015).3	0.073	41.25	19	0	35.6±5.9 y (M±SD)	W(68.4), B(21), H(26.3), A(10.5), NH(73.7)				−10 min; +25-30 min					NI
	Cong (2015).4	0.068	41.25								−10 min; +55-60 min					NI
Ditzen et al. (2007)		NI	NI	22	100	26.6±4.2 y (M±SD)	NI	S	R1+R2	10	0 min; +10 min	P	RIA	NI	No	NI
Elmadih et al. (2014)	Elmadih (2014).1	0.036	301.87	14	100	27.64±4.77 y (M±SD)	W(100)	P	P+C	10	0 min; +10 min	P	ELISA	Abcam	No	0	mothers with lower sensitivity
	Elmadih (2014).3	−0.008	301.87								0 min; +15 min					0
	Elmadih (2014).2	−0.129	235.09	15		30.40±5.37 y (M±SD)					0 min; +10 min					0	mothers with higher sensitivity
	Elmadih (2014).4	−0.287	235.09								0 min; +15 min					0
Feldman, Gordon and Zagoory-Sharon (2010)	Feldman (2010a).1	0.318	10.46 pg OT/mg protein	55	65	28.3±5.11 y (M±SD; mothers); 29.6±4.78 y (M±SD; fathers)	NI	P	P+C	15	0 min; +15 min	S	ELISA	Assay Design	No	+
	Feldman (2010a).2	0.379	16.37 pg OT/mg protein		NI	[4-6 m], 157.1±11.9 d (M±SD)										+
Feldman, Gordon, Schneiderman, et al. (2010)	Feldman (2010b).1	0.346	6.01	32	100	28.7±5.29 y (M±SD; mothers); 29.1±4.28 y (M±SD; fathers)	NI	P	P+C	15	0 min; +30 min	S	ELISA	Assay Design	No	+	mothers with high affectionate contact
	Feldman (2010b).2	−0.286	5.7	19												0	mothers with low affectionate contact
	Feldman (2010b).3	0.266	7.86	18	0											+	fathers with high stimulatory contact
	Feldman (2010b).4	−0.140	6.65	11												0	fathers with low stimulatory contact
Feldman et al. (2011)		NI	NI	112	63	28.7±5.29 y (M±SD; mothers); 29.1±4.28 y (M±SD; fathers)	NI	S	P+C	15	0 min	P	ELISA	Assay Design	No	NI
		NI	NI					U								NI
Feldman et al. (2013)	Feldman (2013).1	0.012	7.86	49	100	*27.24±3.67 y (M±SD)	W(100)	P	P+C	7	NI	S	ELISA	Assay Design	No	NI
	Feldman (2013).2	−0.186	11	46	0	*29.45-3.87 y (M±SD)										NI
	Feldman (2013).3	−0.178	7.18	48	NI	*38.9±2.68 m (M±SD)				24	0 min; +24 min					NI
Fries et al. (2005)		NI	18.99 μg/mg of creatinine	21	57	54.2 m (M)	NI	P	P+C	30	−x d; +45–50 min	U	HPLC	Not applicable	Yes - solid phase (Oasis or Strata)	NI
	Fries (2005).1	−0.143	18.99 μg/mg of creatinine						S1+S2							NI
Gouin et al. (2010)		NI	NI	74	50	*[22-73 y], 38.47±11.99 y (M±SD)	W(91.9), B(4.1), H(2.7), A(1.3)	S	R1+R2	50	−190 min; +65 min	P	ELISA	Assay Design	No	0
Grape et al. (2002)		NI	0.469	8	75	[28-53 y], 40.2 y (M)	NI	Sing	S1+S2	45	−x min; +75 min	Se	ELISA	Peninsula	No	NI	amateur singers
		NI	0.620		50	[26-49 y], 36.4 y (M)										NI	professional singers
Grewen et al. (2005)		NI	1.65	38	100	*[20-49 y], 28.66±1.09 y (M±SE; females)	W(82)	S+T	R1+R2	10	−2 min; +17 min	P	RIA	Amico lab	Yes - liquid-liquid (acetone-ether)	+
		NI	1.65								−2 min; +14 min					NI
		NI	1.65								−2 min; +20 min					NI
		NI	1.53		0		W(79)				−2 min; +14 min; +17 min; +20 min					0
Grewen et al. (2010)	Grewen (2010).1	0.425	4.67	15	100	*[21.0-45.0 y], 30.8±1.4 y (M±SE)	W(50), B(50)	P	P+C	5	0 min; +12 min	S	ELISA	Assay Design	Yes - solid phase (Strata-X)	+
	Grewen (2010).2	0.017	4.82	20							0 min; +5 min	P	ELISA			NI
	Grewen (2010).3	−0.308	4.82								0 min; +12 min					NI
Heinrichs et al. (2001)	Heinrichs (2001)	0.010	9.6	23	100	30.2±1.0 y (M±SE)	NI	T	P+C	15	0 min; +25 min	P	RIA	NI	No	0
Kasos et al. (2018)	Kasos (2018).1	−0.170	536.82	18	100	*23.28±3.54 y (M±SD)	NI	O (hypnosis)	O	NI	0 min; +x min	S	ELISA	Enzo	Yes - solid phase (Sep-Pak C18)	0	clients
	Kasos (2018).2	0.464	263.02	5		54.2±11.43 y (M±SD)										0	hypnotists
Keeler et al. (2015)	Keeler (2015).1	−0.075	201.8	4	50	over 18 years of age	NI	Sing	G	5.63	−5 min; +5.63 min	P	ELISA	Enzo	No	NI	standard vocal performance
	Keeler (2015).2	0.212	184.6							6.02	−5 min; +6.02 min					NI	improvised vocal performance
Kim et al. (2014)		NI	1.75	50	100	[19-41 y], 28.0±4.6 y (M±SD)	W(62), O(38)	P	P+C	5	0 min; +5 min	P	RIA	NI	Yes - liquid-liquid (acetone-ether)	NI
Krause et al. (2016)	Krause (2016)	0.221	0.83	44	100	[21.9-44.2 y], 33.6±5.4 y (M±SD)	NI	O (interactions during psychological assessment)	Ps+Part	9-34	0 min; +9-34 min	P	RIA	RIAgnosis	No	+
Levi-Shachar et al. (2020)	Levi-Shachar (2020)	0.024	40.44	33	45	8.95±1.75 y (M±NI)	NI	S	P+C	5	0 min; +20 min	S	ELISA	Enzo	No	0
Light et al. (2005)		NI	NI	59	100	[20-49 y]	NI	S+T	R1+R2	10	−2 min; 13 min	P	RIA	Amico lab	Yes - liquid-liquid (acetone-ether)	NI
Markova (2018)	Markova (2018).1	0.044	166	34	100	*31.60±3.58 y (M±SD)	W(100)	P	P+C	10	0 min; +10 min	S	ELISA	Enzo	No	NI
	Markova (2018).2	−0.097	193.9	28	NI	*[4 m], 139.43±19.42 d (M±SD)										NI
Matsunaga et al. (2020)	Matsunaga (2020)	0.025	83.32	24	100	[27-45 y], 32.58±4.76 y (M±SD)	NI	T	P+C	15	NI	S	ELISA	Enzo	No	NI
Melton et al. (2019)	Melton (2019).1	0.037	3.05 ng	16	50	*[25-40 y], 33 (Mdn)	NI	P	R1+R2	60	0 min; +60 min	U	ELISA	Enzo	Yes - solid phase (Strata-X)	NI
	Melton (2019).2	0.582	4.58 ng	18				O (visiting art class with an instructor and spouse)	G							NI
Pratt et al. (2015)	Pratt (2015).1	0.096	8.04	51	100	*38.66±4.40 y (M±SD)	NI	P	P+C	15	0 min; +25 min	U	ELISA	Assay Design	Yes - solid phase (Oasis HLB)	0
	Pratt (2015).2	0.194	9.91		NI	*6.33±1.25 y (M±SD)				25						0
Schladt et al. (2017)	Schladt (2017).1	−0.252	4.09	38	55	[18-29 y], 22-23 y (Mdn)	NI	Sing	G	10	0 min; +10 min	S	RIA	RIAgnosis	No	-
	Schladt (2017).2	−0.328	4.09							20	0 min; +20 min					-
	Schladt (2017).3	−0.384	4.09								0 min; +40 min					-
Smith et al. (2013)	Smith (2013).1	0.107	1.96	119	0	*29.3±6.6 y (M±SD)	W(91), O(9)	S+T	R1+R2	11.3	0 min; +11.3 min	P	RIA	Amico lab	No	0	positive spouse contact day 1
	Smith (2013).7	−0.018	1.96								0 min; +21.3 min					0
	Smith (2013).3	−0.115	2.08	57				S		9	0 min; +9 min					0	positive spouse contact day 2
	Smith (2013).9	−0.097	2.08								0 min; +19 min					0
	Smith (2013).5	−0.032	1.85	56							0 min; +9 min					0	neutral spouse contact day 2
	Smith (2013).11	0.024	1.85								0 min; +19 min					0
	Smith (2013).2	−0.022	1.88	119	100	*27.9±6.6 y (M±SD)		S+T		11.3	0 min; +11.3 min					0	positive spouse contact day 1
	Smith (2013).8	0.041	1.88								0 min; +21.3 min					0
	Smith (2013).4	0.084	1.92	57				S		9	0 min; +9 min					0	positive spouse contact day 2
	Smith (2013).10	−0.078	1.92								0 min; +19 min					0
	Smith (2013).6	0.091	1.84	56							0 min; +9 min					0	neutral spouse contact day 2
	Smith (2013).12	0.051	1.84								0 min; +19 min					0
Strathearn et al. (2009)	Strathearn (2009).1	0.569	1.46	15	100	28.0±4.3 y (M±SD)	W(67), O(33)	P	P+C	5	0 min; +5 min	Se	RIA	NI	No	NI	mothers with secure attachment
	Strathearn (2009).2	−0.608	1.76			29.6±3.6 y (M±SD)	W(53), O(47)									NI	mothers with insecure attachment
Strathearn et al. (2012)		NI	NI	55	100	[19-41 y], 28±4.5 y (M±SD)	W(67), B(13), H(20)	P	P+C	5	0 min; +5 min	P	RIA	Amico lab	No	NI
Tse et al. (2018)	Tse (2018).1	0.340	44.26	61	100	34.98±5.67 y (M±SD)	NI	P	P+C	10	−10 min; +15 min	S	ELISA	Assay Design	No	NI
Uvnas-Moberg et al. (1991)		NI	16.30-22.80 pM	10	80	[21-55 y], 40 y (M; females); [40-59 y], 50 y (M; males)	NI	S	G	5	−5 min; −4 min; −3 min; −2 min; −1 min; +1 min; +2 min; +3 min; +4 min; +5 min; +23 min; +24 min	P	RIA	NI	Yes - solid phase (Sep-Pak C18)	NI
Varga and Kekecs (2014)	Varga (2014).1	−0.248	1.97	12	0	*29.62±6.69 y (M±SD)	NI	O (hypnosis)	O	NI	0 min; NI	S	ELISA	Enzo	Yes - solid phase (Sep-Pak C18)	0	clients
	Varga (2014).2	0.364	3.03	4	0	Adults										0	hypnotists
Vittner et al. (2018)	Vittner (2018).1	0.767	161.97	28	100	32±1.13 y (M±SD)	W(68), B(11), H(18), A(3)	T	P+C	60	−10 min; +50 min	S	ELISA	Enzo	No	+	mothers interacting with infants
	Vittner (2018).2	0.373	161.97								−10 min; +105 min					NI
	Vittner (2018).3	0.802	134.71	28	32	[3-10 d]	W(61), B(14), H(18), A(7)				−10 min; +50 min					+	infants interacting with mothers
	Vittner (2018).4	0.512	134.71								−10 min; +105 min					NI
	Vittner (2018).5	0.483	142.99	27	0	*33±1.38 y (M±SD)	W(68), B(14), H(14), A(3)				−10 min; +50 min					+	fathers interacting with infants
	Vittner (2018).6	0.104	142.99								−10 min; +105 min					NI
	Vittner (2018).7	0.914	130.71	28	32	[3-10 d]	W(61), B(14), H(18), A(7)				−10 min; +50 min					+	infants interacting with fathers
	Vittner (2018).8	0.656	130.71								−10 min; +105 min					NI
Yirmiya et al. (2020)		NI	28.84	53	68	11.55±1.14 y (M±SD)	NI	P	P+C	NI	0 min; +x min	S	ELISA	Assay Design	No	0	mothers with high maternal safety signals
		NI	41.69													0	mothers with low maternal safety signals
Yuhi et al. (2018)	Yuhi (2018).1	−0.049	125	9	0	[21-69 y], 27.6±5.3 y (M±SE)	O(W/A) (11.1), A(88.9)	O (group cooking)	G	60	0 min; +70 min	S	ELISA	Enzo	No	0
	Yuhi (2018).2	0.185	153		100	[21-50 y], 33.9±3.9 y (M±SE)	O(W/A) (33.3), A(66.7)									0
Zyga (2019)		NI	54.38	20	38.1	*[3-5.9 y], 4.38± 0.85 (M±SD)	majority W	P	P+C	5	−60 min; +10 min	S	ELISA	Arbor Assays	Yes	NI

Note.

Sex (% of females): values are rounded up to a whole number.

Age: M = mean; Mdn = median; SD = standard deviation; SE = standard error of mean; y = years; m = months; w = weeks; d = days.

Ethnicity: H = Hispanic; W = White; B = Black; NH = non-Hispanic; A = Asian; O = other.

Interaction type: T = tactile/physical/skin-to-skin/holding interaction; S = speaking; P = play/game; S+T = Speaking and Tactile; Sing = singing; TR = Trust-related interaction; O = other.

Participants type: P+C = parent+child; G = group of 2+ people, but not parents with children; R1+R2 = two members of romantic couple; F1+F2 = two adults, friends or relatives; S1+S2 = two strangers; Ps+Part = psychologist+participant; O = other.

OXT source: S = saliva; P = blood plasma; Se = blood serum; U = urine.

Result: + = significant (p≤0.05) OXT pre to post increase; - = significant (p≤0.05) OXT pre to post decrease; 0 = no significant changes.

NI = no information.

^*denotes cases 1) when the age of OXT donors was indicated by the authors without taking into account the exclusion of several participants or samples from the OXT analysis, 2) when the age of OXT donors was not indicated for the current time point but instead for the previous time point plus average/fixed time period

^**all ELISAs were competitive enzyme-linked immunosorbent assays

^§denotes outlier removed from meta-analyses

Overview of the Studies with Between-Groups Comparison and Between-Conditions (Within-Group) Comparison Design

Reference	Data included in meta-analysis, coordinates in forest plots	Hedges’ g	Mean baseline oxytocin concentration (pg/ml unless otherwise stated)	Compared conditions/groups	N of OXT donors	Sex (% of females)	Age [range], M±SD or SE	Ethnicity (%)	Interaction type	Participants type	Duration of interaction	OXT collection time in relation to the beginning of interaction	OXT source	OXT analysis method	OXT parameter	Immunoassay kit manufacturer	Sample extraction utilized?	Results
Between-conditions comparison
Keri et al. (2009)	Kéri (2009)	0.492	NI	control; trust	50	68	47.8±7.3 y (M±SD)	NI	TR	S1+S2	NI	NI	P	ELISA	post-interaction OXT	Assay Design	No	+
Keri and Kiss (2011)	Kéri (2011)	0.693	NI	control; trust	60	42	29.4±8.0 y (M±SD)	NI	TR	S1+S2	NI	NI	P	ELISA**	post-interaction OXT	Assay Design	No	+
Kiss et al. (2011)	Kiss (2011).1	0.662	NI	control; trust	82	NI	30.7±8.0 y (M±SD)	NI	TR	S1+S2	60	+50-60 min	P	ELISA	post-interaction OXT	Assay Design	No	+
Schladt et al. (2017)	Schladt (2017).4	−0.106	4.09	solo; choir	38	55	[18-29 y], 22-23 y (Mdn)	NI	Sing	G	10	0 min; +10 min	S	RIA	post-interaction OXT	RIAgnosis	No	-
	Schladt (2017).5	−0.174	4.09								20	0 min; +20 min						-
	Schladt (2017).6	−0.128	4.09								20	0 min; +40 min						-
Yuhi et al. (2018)	Yuhi (2018).3	−0.589	125	individual cooking; group cooking	9	0	[21-69 y], 27.6±5.3 y (M±SE)	O(W/A) (11.1), A(88.9)	O (group cooking)	G	60	0 min; +70 min	S	ELISA	post-interaction OXT	Enzo	No	-
	Yuhi (2018).4	−0.215	153			100	[21-50 y], 33.9±3.9 y (M±SE)	O(W/A) (33.3), A(66.7)										0
Between-groups comparison
Ditzen et al. (2007)		NI	NI	control; social support group	44	100	26.6±4.2 y (M±SD; interaction); *26.8±4.7 y (M±SD; control)	NI	S	R1+R2	10	0 min; +10 min	P	RIA	post-interaction OXT	NI	No	NI
Smith et al. (2013)		NI	2.30 (no contact), 1.96 (positive spouse contact)	no contact; positive spouse contact	177	0	*29.3±6.6 y (M±SD)	W(91), O(9)	S+T	R1+R2	11.3	0 min; +11.3 min	P	RIA	post-interaction OXT	NI	No	0
		NI	1.95 (no contact), 1.88 (positive spouse contact)			100	*27.9±6.6 y (M±SD)											0

Note.

Sex (% of females): values are rounded up to a whole number.

Age: M = mean; Mdn = median; SD = standard deviation; SE = standard error of mean; y = years; m = months; w = weeks; d = days.

Ethnicity: H = Hispanic; W = White; B = Black; NH = non-Hispanic; A = Asian; O = other.

Interaction type: T = tactile/physical/skin-to-skin/holding interaction; S = speaking; P = play/game; Sing = singing; TR = Trust-related interaction; O = other.

Participants type: P+C = parent+child; G = group of 2+ people, but not parents and children; R1+R2 = two members of romantic couple; F1+F2 = two adults, friends or relatives; S1+S2 = two strangers; Ps+Part = psychologist+participant; O = other.

OXT source: S = saliva; P = blood plasma; Se = blood serum; U = urine.

Result: + = significant (p≤0.05) increase in OXT concentrations: interaction higher than control; - = significant (p≤0.05) decrease in OXT concentrations: interaction lower than control; 0 = no significant changes.

NI = no information.

^*denotes cases 1) when the age of OXT donors was indicated by the authors without taking into account the exclusion of several participants or samples from the OXT analysis, 2) when the age of OXT donors was not indicated for the current time point but instead for the previous time point plus average/fixed time period

^**all ELISAs were competitive enzyme-linked immunosorbent assays

Overview of the Studies with Associative (Correlational and Regression) Design

Reference	Data included in meta- analysis, coordinates in forest plots	Fisher’s z	Mean baseline oxytocin concentration (pg/ml unless otherwise stated)	N of OXT donors	Sex (% of females)	Age [range], M±SD or SE	Ethnicity (%)	Interaction type	Participants type	Duration of interaction	OXT collection time in relation to the beginning of social interaction	OXT source	OXT analysis method	Immunoassay kit manufacturer	Sample extraction utilized?	OXT parameter	Approach for assessment of social constructs	Categories of social constructs	Details of social constructs	Types of approaches for assessment of social constructs	Results	Type of statistical analysis	Additional description of the sample/design
Correlations
Algoe et al. (2017)	Algoe (2017).1	0.203	NI	126	NI	*[18-50 y], 23.7±5.64 y (M±SD)	W(70.9), B(7.4), A(15.9), NH(90.7), O(5.8)	S	R1+R2	5	NI	U	ELISA**	Enzo	Yes - solid phase (Strata-X)	24 h cumulative	Customary behavioral analysis	P	perceived expresser responsiveness	Q-current	+	PC
	Algoe (2017).2	0.288	NI																experienced love		+
	Algoe (2017).3	0.030	NI																experienced reward		0
	Algoe (2017).4	0.245	NI																perceived expresser gratitude		+
	Algoe (2017).5	0.234	NI																perceived expresser love		+
	Algoe (2017).6	0.030	NI																perceived expresser reward		0
Apter-Levi et al. (2014)	Apter-Levi (2014).1	0.060	388.05 (mothers), 391.18 (fathers)	119	60	28.9±5.22 y (M±SD; mothers); 29.3±4.26 y (M±SD; fathers)	NI	P	P+C	15	NI	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	object salience	B	0	PC
	Apter-Levi (2014).2	0.193																	social salience		+
	Apter-Levi (2014).3	−0.040																	joint attention		0
	Apter-Levi (2014).4	0.110																S	gaze synchrony		0
	Apter-Levi (2014).5	−0.020																P	stimulatory contact		0
	Apter-Levi (2014).6	0.313																	affectionate contact		+
Atzil et al. (2011)	Atzil (2011)	0.710	NI	19	100	*[22-37 y]	NI	P	P+C	NI	NI	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	S	parent-infant synchrony	B	+	PC
Atzil et al. (2017)	Atzil (2017)	0.549	NI	17	100	*[21-42 y]	NI	P	P+C	2	NI	P	ELISA	Enzo	No	baseline OXT	Customary behavioral analysis	S	vocalization synchrony	B	+	PC
Bick et al. (2013)	Bick (2013).3	0.354	13.54 pg/ml adjusted for creatinine	41	100	*[28-68 y], 42.1±10.1 y (M±SD)	W(47), B(46), H(7)	P	P+C	40	0 min; +30 min	U	ELISA	Assay Design	Yes - solid phase (Sep-Pak C18)	OXT change	Maternal Delight Scale	P	delight	B	+	PC	first assessment (2 months of the relationship)
	Bick (2013).4	0.523	16.04 pg/ml adjusted for creatinine	32																	+		second assessment (3 months after the first assessment)
Feldman et al. (2007)	Feldman (2007).1	0.255	NI	62	100	[18.4-43.2 y], 27.8±0.70 y (M±SE)	NI	P	P+C	15	−28 w	P	ELISA	R&D	No	baseline OXT	Coding Interactive Behavior Manual-Newborn Version	P	maternal-behavior composite	B	+	PC	OXT collected at the first trimester
	Feldman (2007).2	0.288	267.87 pM			*[18.4-43.2 y], 27.8±0.70 y (M±SE)					NI										+		OXT collected at the first postpartum month
Feldman, Gordon and Zagoory-Sharon (2010)	Feldman (2010a).7	0.234	10.46 pg OT/mg protein	55	65	28.3±5.11 y (M±SD; mothers); 29.6±4.78 y (M±SD; fathers)	NI	P	P+C	15	0 min	S	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	social engagement	B	0	PC
	Feldman (2010a).8	0.277																	affect synchrony		+
	Feldman (2010a).3	0.310									+15 min					post-interaction OXT			social engagement		+
	Feldman (2010a).4	0.343																	affect synchrony		+
	Feldman (2010a).9	0.299	16.37 pg OT/mg protein		NI	[4-6 m], 157.1±11.9 d (M±SD)					0 min					baseline OXT			social engagement		+
	Feldman (2010a).10	0.277																	affect synchrony		+
	Feldman (2010a).5	0.436									+15 min					post-interaction OXT			social engagement		+
	Feldman (2010a).6	0.332																	affect synchrony		+
Feldman, Gordon, Schneiderman, et al. (2010)	Feldman (2010b).5	0.288	6.17	71	100	28.7±5.29 y (M±SD)	NI	P	P+C	15	0 min	S	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	affectionate contact	B	+	PC
	Feldman (2010b).6	0.377	365.59									P									+
		NI	6.17									S							stimulatory contact		0
		NI	365.59									P									0
		NI	7.09	41	0	29.1±4.28 y (M±SD)						S							affectionate contact		0
		NI	405.10									P									0
	Feldman (2010b).7	0.343	7.09									S							stimulatory contact		+
	Feldman (2010b).8	0.412	405.10									P									+
Feldman et al. (2011)	Feldman (2011).1	0.321	365.59 (mothers), 405.10 (fathers)	112	63	28.7±5.29 y (M±SD; mothers); 29.1±4.28 y (M±SD; fathers)	NI	P	P+C	15	0 min	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	positive engagement	B	+	PC
	Feldman (2011).6	0.288																	positive communicative sequences		+
	Feldman (2011).4	0.354																	affect synchrony		+
	Feldman (2011).5	−0.100																N	interactive stress		0
	Feldman (2011).8	0.172	6.17 (mothers), 7.09 (fathers)									S						P	positive communicative sequences		0
	Feldman (2011).2	0.266																	positive engagement		+
	Feldman (2011).16	0.277																	affect synchrony		+
	Feldman (2011).7	0.020																N	interactive stress		0
	Feldman (2011).13	−0.288															Adult Attachment Style		attachment anxiety	Q-general	-
	Feldman (2011).14	−0.332																	attachment avoidance		-
	Feldman (2011).15	0.299															Yale Inventory of Parent Thought and Action (YIPTA)		parental preoccupations		+
	Feldman (2011).9	0.080	10.34	71	100	28.7±5.29 y (M±SD)						U			Yes - solid phase (Oasis HLB)		Customary behavioral analysis	P	positive engagement	B	0
	Feldman (2011).11	−0.070	10.34																positive communicative sequences		0
	Feldman (2011).10	−0.050	10.34																affect synchrony		0
	Feldman (2011).3	0.485	10.34															N	interactive stress		+
	Feldman (2011).12	0.354	10.34														Parenting Stress Index (PSI)		parenting stress	Q-general	+
Feldman et al. (2012)	Feldman (2012)	0.310	379.54 pM	272	56	[21-37 y]	W(100)	P	P+C	NI	0 min	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	parental touch	B	+	PC
Feldman et al. (2013)	Feldman (2013).4	0.388	7.18	48	NI	*38.9±2.68 m (M±SD)	W(100)	P	P+C	24	0 min; +24 min	S	ELISA	Assay Design	No	composite averaged from baseline and post-interaction assessments	Coding Interactive Behavior (CIB)	P	social reciprocity	B	+	PC
Gonzaga et al. (2006)	Gonzaga (2006).1	0.563	NI	25	100	*[23-35 y], 28.12±4.0 y (M±SD)	W(68), B(4), H(8), A(20)	S	S1+S2	NI	NI	P	RIA	NI	No	OXT change	Relived Emotion Task (RET)	P	affiliation cues	B	+	PC
	Gonzaga (2006).2	−0.040	NI																sexual cues		0
	Gordon (2017).12	0.020	NI																parent stimulatory affectionate touch		0
Gordon et al. (2010a)	Gordon (2010a).1	0.343	337.35	61	100	*27.24±3.67 y (M±SD)	NI	P	P+C	10	NI	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	affectionate parenting behavior	B	+	PC
	Gordon (2010a).2	−0.224	337.35																stimulatory parenting behavior		0
	Gordon (2010a).3	0.080	401.98	58	0	*29.45±3.87 y (M±SD)													affectionate parenting behavior		0
	Gordon (2010a).4	0.310	401.98																stimulatory parenting behavior		+
Gordon et al. (2010c)	Gordon (2010c)	0.658	NI	43	0	28.08±4.19 y (M±SD)	NI	P	P+C	12	NI	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	P	affect synchrony	B	+	PC
Gordon et al. (2017)	Gordon (2017).1	0.255	NI	71	100	*27.72±3.52 y (M±SD)	NI	P	P+C	5	0 min	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	S	parent-infant synchrony	B	+	PC
	Gordon (2017).2	0.192	NI																		0
	Gordon (2017).3	−0.266	NI															P	parent stimulatory affectionate touch		-
	Gordon (2017).4	0.343	NI	55														S	parent-infant synchrony		+
	Gordon (2017).5	0.321	NI															P	parent affectionate touch		+
	Gordon (2017).6	−0.100	NI																parent stimulatory affectionate touch		0
	Gordon (2017).7	0.141	NI	72	0	*29.281±3.52 y (M±SD)												S	parent-infant synchrony		0
	Gordon (2017).8	0.151	NI															P	parent affectionate touch		0
	Gordon (2017).9	0.100	NI																parent stimulatory affectionate touch		0
	Gordon (2017).10	0.070	NI	49														S	parent-infant synchrony		0
	Gordon (2017).11	0.070	NI															P	parent affectionate touch		0
Grewen et al. (2005)	Grewen (2005).1	0.400	1.53	38	0	*[20-49 y], 28.66±1.09 y (M±SE; females); 29.26±0.93 y (M±SE; males)	W(79)	S+T	R1+R2	10	−2 min	P	RIA	Amico lab	Yes - liquid-liquid (acetone-ether)	baseline OXT	Social Relationships Index (SRI)	P	support	Q-general	+	PC
	Grewen (2005).2	0.377	1.65		100		W(82)														+
	Grewen (2005).3	0.354	1.65								+14 min	P	RIA			post-interaction OXT					+
Julian et al. (2018)	Julian (2018).1	0.090	NI	33	100	[21-38 y], 27.18±4.60 у (M±SD)	NI	P	P+C	25	0 min; +45 min	S	ELISA	Enzo	No	AUC, the area under the curve	Maternal Warmth and Control Scale	P	positive parenting	B	0	PC
	Julian (2018).2	−0.310	NI															N	negative parenting		0
	Julian (2018).3	0.100	NI													baseline OXT		P	positive parenting		0
	Julian (2018).4	−0.299	NI															N	negative parenting		0
Kasos et al. (2018)	Kasos (2018).3	−0.131	NI	18	100	*23.28±3.54 y (M±SD)	NI	O (hypnosis)	O	NI	0 min; +x min	S	ELISA	Enzo	Yes - solid phase (Sep-Pak C18)	OXT change	Dyadic Interactional Harmony Questionnaire (DIH)	P	communion subscale	Q-current	0	SC	OXT concentration in client, DIH rated by the client
	Kasos (2018).4	−0.485	NI																		0		OXT concentration in client, DIH rated by the hypnotist
	Kasos (2018).5	0.388	NI														Phenomenology of Consciousness Inventory (PCI)		positive affect		0		OXT concentration in client, PCI rated by the client
	Kasos (2018).6	0.255	NI															N	negative affect		0		OXT concentration in client, PCI rated by the client
Kiss et al. (2011)	Kiss (2011).2	0.213	NI	82	NI	30.7±8.0 y (M±SD)	NI	TR	S1+S2	60	+50-60 min	P	ELISA	Assay Design	No	post-interaction OXT	Experience in Close Relationship (ECR)	N	attachment anxiety	Q-general	0	PC	non-trust related condition
	Kiss (2011).3	0.523	NI																attachment avoidance		+
	Kiss (2011).4	0.485	NI																attachment anxiety		+		trust related condition
	Kiss (2011).5	0.255	NI																attachment avoidance		+
Light et al. (2005)	Light (2005).1	0.321	NI	59	100	[20-49 y]	NI	S+T	R1+R2	10	−2 min	P	RIA	Amico lab	Yes - liquid-liquid (acetone-ether)	baseline OXT	Physical Affection Scale	P	hugs	Q-general	+	PC
	Light (2005).2	0.299	NI																massages		+
	Light (2005).3	0.060	NI								+13 min					post-interaction OXT			hugs		0
		NI	NI																massages		NI
		NI	NI								−2 min					baseline OXT			kissing		0
		NI	NI								+13 min					post-interaction OXT					0
		NI	NI								−2 min					baseline OXT			hand-holding		0
		NI	NI								+13 min					post-interaction OXT					0
		NI	NI								−2 min					baseline OXT			sitting/lying close		0
		NI	NI								+13 min					post-interaction OXT					0
		NI	NI								−2 min					baseline OXT			support		0
MacKinnon et al. (2014)	MacKinnon (2014).1	−0.002	NI	287	100	*31.40±4.60 y (M±SD)	NI	P	P+C	5	− 31 w	P	ELISA	Enzo	No	baseline OXT	Global Rating Scales (GRS)	N	depressive behavior	B	0	PC	OXT collected at the 12-14 weeks gestation
	MacKinnon (2014).2	−0.020	NI								− 11 w										0		OXT collected at the 32–34 weeks gestation
	MacKinnon (2014).3	−0.080	296.76								NI										0		OXT collected at the 7-9 weeks postpartum
	MacKinnon (2014).4	0.080	NI								− 31 w							P	sensitivity		0		OXT collected at the 12-14 weeks gestation
	MacKinnon (2014).5	0.060	NI								− 31 w							N	intrusiveness		0
	MacKinnon (2014).6	0.060	NI								− 31 w							N	remoteness		0
	MacKinnon (2014).7	−0.020	NI								− 11 w							P	sensitivity		0		OXT collected at the 32-34 weeks gestation
	MacKinnon (2014).8	0.004	NI								− 11 w							N	intrusiveness		0
	MacKinnon (2014).9	0.080	NI								− 11 w								remoteness		0
	MacKinnon (2014).10	−0.030	296.76								NI							P	sensitivity		0		OXT collected at the 7-9 weeks postpartum
	MacKinnon (2014).11	0.020	296.76								NI							N	intrusiveness		0
	MacKinnon (2014).12	−0.050	296.76								NI								remoteness		0
MacKinnon et al. (2018)	MacKinnon (2018).1	0.010	NI	189	100	35.56±4.36 y (M±SD)	NI	P	P+C	5	−2-3 y	P	ELISA	Enzo	No	baseline OXT	Emotional Availability Scales (EAS)	P	sensitivity	B	0	PC	OXT collected at the 32-34 weeks gestation
	MacKinnon (2018).2	0.070	NI																structuring		0
	MacKinnon (2018).3	0.030	NI																non-intrusive behavior		0
	MacKinnon (2018).4	0.040	NI																non-hostility		0
Markova (2018)	Markova (2018).3	0.592	166	34	100	*31.60±3.58 y (M±SD	W(100)	P	P+C	10	0 min; +10 min	S	ELISA	Enzo	No	AUCI, the area under the curve with respect to increase	Customary behavioral analysis	P	game rate	B	+	SC
	Markova (2018).4	0.386	166																		+
	Markova (2018).5	−0.439	193.9	28	NI	*[4 m], 139.43±19.42 d (M±SD)										post-interaction OXT					-
	Markova (2018).6	−0.430	193.9																game time		-
	Markova (2018).7	−0.454	193.9	24												AUCI, the area under the curve with respect to increase			game rate		-
Markova and Siposova (2019)	Markova (2019).1	−0.369	NI	32	100	*[22-38 y], 31.60±3.57 y (M±SD)	W(100)	P	P+C	10	0 min; +10 min	S	ELISA	Enzo	No	baseline OXT	Customary behavioral analysis	P	warm sensitivity	B	-	PC
	Markova (2019).2	−0.457	NI	37																	-
	Markova (2019).3	−0.361	NI	35												post-interaction OXT					-
	Markova (2019).4	−0.541	NI	21												baseline OXT					-		mothers in the high attunement group
	Markova (2019).5	−0.568	NI	24																	-		mothers in the high attunement group
Samuel et al. (2015)	Samuel (2015).1	0.119	283.39	90	100	32.68±4.19 y (M±SD)	NI	P	P+C	5	NI	P	ELISA	Enzo	No	baseline OXT	Global Rating Scales (GRS)	P	sensitivity	B	0	PC
	Samuel (2015).2	0.225	283.39																nonintrusiveness		+
	Samuel (2015).3	−0.095	283.39																nonremoteness		0
	Samuel (2015).4	−0.023	283.39																nondepressive		0
Schneiderman et al. (2012)	Schneiderman (2012).1	0.299	509.83 (women), 480.76 (men)	120	50	22.84±4.50 y (M±SD; females), 25.03±8.78 y (M±SD; males)	NI	S	R1+R2	5	NI	P	ELISA	Assay Design	No	baseline OXT	Coding Interactive Behavioral Manual (CIB)	P	interactive reciprocity	B	+	PC
	Schneiderman (2012).2	0.192															Romantic Couple Interview	N	worries	Q-general	+
	Schneiderman (2012).3	0.100																	attachment anxiety		0
	Schneiderman (2012).4	−0.070																	attachment avoidance		0
Smith et al. (2013)	Smith (2013).15	−0.075	1.96	170	0	*29.3±6.6 y (M±SD)	W(91), O(9)	S+T	R1+R2	11.3	0 min	P	RIA	Amico lab	No	baseline OXT	Dyadic Adjustment Scale (DAS) and the Marital Adjustment Test (MAT)	P	relationship quality composite	Q-general	0	PC
	Smith (2013).16	0.026	1.96	119							0 min; +11.3 min					OXT change					0
	Smith (2013).17	−0.131	1.88	169	100	*27.9±6.6 y (M±SD)					0 min					baseline OXT					0
	Smith (2013).18	−0.091	1.88	119							0 min; +11.3 min					OXT change					0
Tse et al. (2017)	Tse (2017).1	−0.277	44.26	61	100	35 y (M)	NI	P	P+C	10	−10 min; +10 min	S	ELISA	Assay Design	No	log (post-interaction OXT) minus log (baseline OXT)	Customary behavioral analysis	Unclear valence	touch frequency	B	-	SC
	Tse (2017).2	−0.090	44.26																eye gaze frequency		0
	Tse (2017).3	−0.100	44.26																eye gaze duration		0
	Tse (2017).4	−0.203	44.26																touch duration		0
	Tse (2017).5	−0.100	44.26														Positive Affect and Negative Affect Scale (PANAS)	P	positive affect	Q-current	0
	Tse (2017).6	0.141	44.26															N	negative affect		0
Tse et al. (2018)	Tse (2018).2	−0.277	44.26	61	100	34.98±5.67 y (M±SD)	NI	P	P+C	10	−10 min; +15 min	S	ELISA	Assay Design	No	log (post-interaction OXT) minus log (baseline OXT)	Positive Affect and Negative Affect Scale (PANAS)	P	positive affect	Q-current	-	PC	baseline positive affect
	Tse (2018).3	−0.161	44.26															N	negative affect		0		baseline negative affect
	Tse (2018).4	−0.277	44.26															P	positive affect		-		post-interaction positive affect
	Tse (2018).5	−0.080	44.26															N	negative affect		0		post-interaction negative affect
Varga and Kekecs (2014)	Varga (2014).3	0.709	1.97	12	0	*29.62±6.69 y (M±SD)	NI	O (hypnosis)	O	NI	0 min; NI	S	ELISA	Enzo	Yes - solid phase (Sep-Pak C18)	OXT change	Dyadic Interactional Harmony Questionnaire (DIH)	P	communion	Q-current	+	PC	clients
	Varga (2014).4	0.371	1.97																intimacy		0
	Varga (2014).6	0.430	1.97														Archaic Involvement Measure (AIM)		archaic involvement		0
	Varga (2014).8	0.356	1.97														Dyadic Interactional Harmony Questionnaire (DIH)		playfulness		0
	Varga (2014).5	0.297	3.03	4		Adults													intimacy		0		hypnotists
	Varga (2014).7	−0.336	3.03														Archaic Involvement Measure (AIM)		archaic involvement		0
	Varga (2014).9	−0.322	3.03														Dyadic Interactional Harmony Questionnaire (DIH)		communion		0
	Varga (2014).10	0.103	3.03																playfulness		0
Vittner et al. (2019)	Vittner (2019).1	−0.460	NI	28	0	33±1.38 y (M±SD)	W(68), B(14), H(14), A(3)	T	P+C	60	NI	S	ELISA	Enzo	No	NI	Parental Risk Evaluation Engagement Model Instrument (PREEMI)	P	Composite	Q-general	-	PC
	Vittner (2019).2	0.234	NI		100	32±1.13 y (M±SD)	W(68), B(11), H(18), A(3)														0
Regressions
Gordon et al. (2010b)		NI	291.23 (time 1), 325.8 (time 2),	37	100	26.26±3.94 y (M±SD)	NI	P	P+C	NI	−4 m; NI	P	ELISA	Assay Design	No	baseline OXT	Customary behavioral analysis	S	triadic synchrony	B	+	regression	two visits for OXT collection: at the second m after the child's birth and when the infant was approximately 6 m old
		NI	306.01 (time 1), 329.71 (time 2)		0	28.81±4.73 y (M±SD)															+
Miura et al. (2014)		NI	NI	50	100	[24-44 y], 35.9±3.9 y (M±SD)	NI	P	P+C	5	NI	U	RIA	NI	No	NI	Interaction Rating Scale (IRS)	P	respect for autonomy development	B	-	regression
		NI	NI	30	0	[31-42 y], 36.8±2.8 y (M±SD)										NI					0

Note.

Sex (% of females): values are rounded up to a whole number.

Age: M = mean; SD = standard deviation; SE = standard error of mean; y = years; m = months; w = weeks; d = days.

Ethnicity: H = Hispanic; W = White; B = Black; NH = non-Hispanic; A = Asian; O = other.

Interaction type: T = tactile/physical/skin-to-skin/holding interaction; S = speaking; P = play/game; Sing = singing; TR = Trust-related interaction; O = other.

Participants type: P+C = parent+child; G = group of 2+ people, but not parents and children; R1+R2 = two members of romantic couple; F1+F2 = two adults, friends or relatives; S1+S2 = two strangers; Ps+Part = psychologist+participant; O = other.

OXT source: S = saliva; P = blood plasma; Se = blood serum; U = urine.

Interaction categories: P = positive constructs; N = negative constructs; S = synchrony.

Types of approaches for assessment of social constructs: B = analysis of behavior; Q-current = questionnaire regarding current interaction procedure; Q-general = questionnaire regarding general social constructs.

Result: + = significant (p≤0.05) positive correlation; - = significant (p≤0.05) negative correlation; 0 = no significant changes.

Type of statistical analysis: PC = Pearson correlation; SC = Spearman correlation.

NI = no information.

^*denotes cases 1) when the age of OXT donors was indicated by the authors without taking into account the exclusion of several participants or samples from the OXT analysis, 2) when the age of OXT donors was not indicated for the current time point but instead for the previous time point plus average/fixed time period

^**all ELISAs were competitive enzyme-linked immunosorbent assays