Effects of oxytocin on agency in anxiously attached individuals: a replication and extension

Abstract

Oxytocin can promote prosocial behaviour but also result in anti-social effects in anxiously attached individuals. Such individuals desire interpersonal closeness but are preoccupied with abandonment; they also lack agency (self-efficacy, empowerment), possibly because of thwarted attachment needs. By augmenting affiliative strivings, oxytocin may threaten anxiously attached individuals’ already fragile sense of agency. Indeed, prior work shows that oxytocin increases communion, especially for avoidant individuals, but decreases agency for anxious individuals. Here, we replicated the oxytocin-avoidance-communion effect (P < .001), with some additional nuance, and the oxytocin-­anxiety-agency effect (P = .014) in a larger (N = 77; within-subject), gender diverse sample. We also investigated bodily agency using the rubber hand illusion paradigm. Anxious individuals also evidenced less bodily agency—experiencing greater ‘loss of own hand’—although this was not potentiated by oxytocin. These findings may explain some of oxytocin’s divergent social effects and highlight the tension that can arise between meeting needs for communion/security and agency in interpersonally vulnerable individuals.

Introduction

Oxytocin plays a key role in attachment/bonding and prosocial behaviour in non-human animals (Carter 1998, Walum and Young 2018), and evidence suggests that oxytocin plays a similar role in humans. To date, most of the human research has manipulated oxytocin via intranasal administration to establish causal effects. [Although some have questioned whether intranasally administered oxytocin enters the central nervous system (Leng and Ludwig 2016), there is now evidence supporting nose-to-brain transport (Martins et al. 2020; Yao et al. 2023)]; while evidence suggests that oxytocin can facilitate aspects of bonding and prosocial behaviour, oxytocin’s effects on such outcomes often depend on the person (Bartz et al. 2011, Bartz 2016).

On the one hand, oxytocin appears to be helpful for those who are less socially motivated and/or attuned. Oxytocin selectively facilitated social cognition/empathic accuracy in those who endorsed: more autism traits (Bartz et al. 2010, Althaus et al. 2016, Guastella and Hickie 2016, Bartz et al. 2019) and alexithymia (Luminet et al. 2011), and less empathic concern (Radke and de Bruijn 2015), trait empathy (Feeser et al. 2015), and emotional sensitivity (Leknes et al. 2013). Oxytocin selectively facilitated trust and cooperation in those who endorsed: more attachment avoidance (De Dreu 2012), a ‘pro-self’ value orientation (Declerck et al. 2014), competitiveness (Arueti et al. 2013), and less interpersonal trust (Merolla et al. 2013). Notably, Declerck et al. (2020) conducted a registered replication investigating the effect of oxytocin on trust; while they did not replicate the main faciliatory effect of oxytocin, additional analyses revealed a significant faciliatory effect of oxytocin on trust for individuals who endorsed less dispositional trust, consistent with De Dreu. Finally, Zhang et al. (2021) found that oxytocin reduced attachment insecurity, especially in avoidant individuals, and Wright et al. (2023) found that oxytocin specifically decreased attachment avoidance.

On the other hand, oxytocin does not appear to be helpful—and may be detrimental (socially)—to those who are insecure and preoccupied with interpersonal relationships. In one early study, Bartz et al. (2011) investigated the effect of oxytocin on trust in individuals with borderline personality disorder (BPD), a condition marked by impulsivity (especially impulsive aggression), emotional reactivity and abandonment concerns. Oxytocin did not facilitate trust—rather, oxytocin decreased trust in BPD, an effect that was independently replicated (Ebert et al. 2013). Subsequent work shows that normal variation in attachment anxiety (i.e. preoccupation with interpersonal relationships and concerns about abandonment) can moderate oxytocin’s effects. Oxytocin (vs. placebo) negatively biased recollections of maternal care and closeness in those who endorsed greater attachment anxiety (Bartz et al. 2010). Similarly, Wang et al. (2020) found that people higher in attachment anxiety were more attuned to negative emotional faces following oxytocin (vs. placebo) and Rockliff et al. (2011) found that those who endorsed lower levels of social safeness had more negative experiences during a compassion focused imagery task following oxytocin. Participants who experienced early parental separation (Meinlschmidt and Heim 2007), harsh parenting (Bakermans-Kranenburg et al. 2012), and childhood emotional maltreatment (Bhandari et al. 2014) also showed more negative (social) effects following oxytocin. Finally, oxytocin (vs. placebo) increased intimate partner violence inclinations among high trait aggressive individuals (DeWall et al. 2014).

How can oxytocin enhance socially oriented cognition and behaviour for some, but impede it for others? It is thought that hormones and other neurochemicals function as endocrine codes, cueing organisms to engage in adaptive behaviours (Roney et al. 2003)—in the case of oxytocin, affiliation (Bartz 2016, Mu et al. 2016). A related hypothesis is that oxytocin sensitizes neural circuits to social inputs (Marlin et al. 2015). These accounts explain why those who are less socially motivated and/or attuned (e.g., avoidantly attached) show the most robust increases in affiliation following oxytocin. However, heightening affiliative strivings and/or social tuning may backfire in anxiously attached individuals. Anxiously attached individuals desire interpersonal closeness but are preoccupied with losing their partner, require constant reassurance that they are loved, and fear being alone (Brennan et al. 1998); they also lack agency—that is, a sense of control, efficacy and empowerment—very possibly because of their thwarted attachment needs. By augmenting affiliative strivings, oxytocin may further antagonize anxiously attached individuals’ fragile sense of agency, and their subsequent feelings of vulnerability may impede prosociality. Supporting this, Bartz et al. (2015) found that oxytocin generally increased a communal orientation, characterized by warmth, kindness and understanding—especially for those who were avoidantly attached; however, oxytocin also selectively decreased agency (self-confidence, independence) in those who were anxiously attached.

The first aim of this research was to replicate the aforementioned oxytocin-attachment avoidance-communion and oxytocin-attachment anxiety-agency findings, and to assess whether these effects hold in a gender diverse sample (the original sample was all male). The second aim was to explore the effects of oxytocin—and attachment—on people’s ‘lower level’, non-conceptual sense of agency. Bartz et al. (2015) assessed agency (and communion) with the Extended Personality Attributes Questionnaire (Spence et al. 1979) by asking participants the extent to which such characteristics as ‘self-confident’ and ‘independent’ described them. This approach is widely used by personality and social psychologists who typically study the self by focusing on the conceptual self-concept—that is, the way people use language to describe themselves. Conversely, cognitive psychologists and philosophers of mind have addressed questions about the self by studying the bodily self, broadly defined as the implicit, pre-reflective awareness of the perceptual experience of one’s body in space (Gallagher 2000). In particular, the grounded cognition approach (Barsalou 2008) posits that all cognitive processes, including cognitive representations of the self, are rooted in bodily self-awareness, and conceptual self-hood is thought to be continuous with bodily self-hood. Thus, we included the widely used rubber hand illusion (RHI) paradigm (Botvinick and Cohen 1998) to assess the effects of oxytocin (and attachment) on agency experienced via the bodily self.

In the RHI, participants are seated with their hand placed on a table, obscured from view. A prosthetic hand is placed in front of the participant (in view). The experimenter strokes the rubber hand and the participant’s real hand, either synchronously or asynchronously. In synchronous trials, the congruence of visual and tactile input leads participants to believe the rubber hand belongs to them, and to experience a loss of ownership over their own hand (Botvinick and Cohen 1998, Tsakiris et al. 2010). Although most participants experience the illusion following synchronous stimulation, they report it less, or not at all, following asynchronous stimulation, because there is no sensory conflict to resolve. However, Krol et al. (2020) found people who have a less clear and stable sense of self are more vulnerable to the illusion even in the asynchronous condition, when the sensory input does not warrant this interpretation. Given that attachment anxiety, in general, is associated with reduced agency (Bartz et al. 2015), one might expect this reduced agency to also be reflected in their bodily self-representation and that more anxious individuals would experience greater loss of their own hand during the RHI. [As noted, little work has compared assessments of the self-concept and bodily self; however, preliminary data indicate that agency as assessed by the EPAQ is significantly associated with self-reported experience of hand loss during the RHI (Bartz, unpublished data; see Supplementary data for details), which supports the overlap between these two aspects of the self.] Moreover, if oxytocin exacerbates anxiously attached individuals’ already fragile sense of agency, then we would expect the association between attachment anxiety and hand loss to be potentiated in the oxytocin (vs. placebo) condition.

Transparency and openness

We report how we determined our sample size, all data exclusions, and all manipulations. The materials used in this research are described in the Methods section and are widely available. The study is a replication of previously published research (Bartz et al. 2015) and we precisely followed the data analytic plan used in the original study. The data-analysis scripts are posted at https://osf.io/a3d2y/.

Methods

Participants

Participants were recruited from the McGill community and surrounding area; they were required to be between 18 and 40 years old and in good health (see Supplementary data for inclusion/exclusion criteria and complete demographics). Seventy-eight participants met eligibility criteria and were enrolled; one participant was discontinued on day one and four participants did not return for day two but these four were accounted for with our analytic strategy. Thus, the final sample was 77 participants. Mean age was 22.27 years (SD = 3.36); there were 32 males and 44 females (one participant did not indicate gender). Sample size was constrained by the resource intensive and costly nature of pharmacological challenge research; however, we believe our sample size is reasonable. First, it is almost double the size of Bartz et al. (2015) sample. Second, our within-subjects design affords more statistical power than the typical between-subjects design used in oxytocin administration studies and is optimal for testing hypotheses involving individual difference moderated effects as we aim to do.

Participants gave informed consent and were compensated $70–80. The study was approved by the McGill University Institutional Review Board.

Study design and procedures

Following Bartz et al. (2015), we used a randomized, double-blind, placebo-controlled challenge procedure in which participants received 24 IU intranasal oxytocin (Syntocinon, Novartis) or placebo (containing all ingredients except the active substance) on two occasions, 1–3 weeks apart. Drug and placebo were imported from Victoria Apotheke Zürich; drug/placebo administration order was counter-balanced.

After the screening, participants completed questionnaires, including an attachment measure (below), and then self-administered the oxytocin/placebo. Approximately 45-mins later, and after an unrelated graphesthesia task, participants underwent two rounds of the RHI paradigm: one involving synchronous and one involving asynchronous stimulation (stimulation order was randomized and counterbalanced across participants but consistent across testing days). After each round, participants completed the Rubber Hand Questionnaire (Longo et al. 2008). Between the two RHI rounds, participants completed measures assessing agency and communion (below) and an unrelated task (Bartz et al. 2019). On day two, participants received the alternate pharmacological compound with identical study procedures.

Rubber hand illusion (Botvinick and Cohen 1998)

Participants were seated at a table with their right hand placed in front of them palm down; a barrier was placed to the left of their hand, obscuring their hand from view. A prosthetic hand was positioned approximately 15 cm to the left of the participant’s hand, on the left side of the barrier so that the prosthetic hand was visible. Latex gloves were placed on the rubber hand and participant’s real hand, as skin colour differences can affect the illusion (e.g., Farmer et al. 2012). The experimenter faced the participant and stroked the prosthetic hand and the participant’s unseen hand using identical paintbrushes. The hands were stimulated on the second, third, and fourth fingers from the second knuckle to the tip of the finger, at a rate of ∼1 stroke/sec. In the synchronous condition, the real and rubber hands were stroked simultaneously; in asynchronous condition, a stroke was delivered to the real hand followed by a stroke to the prosthetic hand, on the same anatomical location, but 500-ms later. In both conditions, participants were instructed to keep their own hand still and to focus on the prosthetic hand. Consistent with other work, stimulation lasted for 2 mins in each condition (Tsakiris et al. 2010).

Measures

Attachment was assessed with the Experience in Close Relationships Scale (ECR; Brennan et al. 1998), a widely used self-report measure of adult attachment. The ECR consists of 36 items assessing the two dimensions of anxiety (e.g., preoccupation and worry about abandonment) and avoidance (e.g., detachment from relationships, preference for self-reliance). Participants indicate, on a seven-point scale, the extent to which they agree or disagree with each item in terms of ‘how they generally experience close relationships’. Example anxiety items include: ‘I often wish that other people’s feelings for me were as strong as my feelings for them,’ ‘I worry a fair amount about losing close relationships,’ ‘I worry about being alone,’ and ‘If I can’t get other people to show interest in me, I get upset or angry’. Example avoidance items include: ‘I get uncomfortable when people want to be very close to me,’ ‘I try to avoid getting too close to people,’ ‘I want to get close to people, but I keep pulling back,’ and ‘I find it difficult to allow myself to depend on others’.

Of note, it is typical in attachment research to look at both the main and interactive effects of attachment anxiety and avoidance to capture the full spectrum of attachment differences (Fig. 1). The two high anxious profiles (often referred to as ‘preoccupied’ and ‘fearful’) want emotionally close relationships, but the former (characterized by low avoidance) are primarily concerned that they are not loved enough, and do not like being alone, whereas the latter (characterized by high avoidance) are afraid to get close to others as they fear rejection. The two high avoidant profiles (often referred to as ‘fearful’ and ‘dismissive’) also differ in important ways: As noted, fearfuls want closeness but fear rejection so avoid intimacy whereas dismissives are comfortable without close relationships and ‘compulsively’ (Bartholomew and Horowitz 1991) prioritize independence and self-sufficiency. Finally, the low anxious, low avoidant (‘secures’) are comfortable with both intimacy and autonomy. While we did not have specific predictions about these more nuanced classifications, we tested for interactions given (i) this is common in attachment research, (ii) Bartz et al. (2015) original study tested for interactive effects, and (iii) including the interaction term allows us to assess how more secure participants respond to oxytocin for comparisons with prior work. (A complete discussion of this topic is beyond the scope of this paper but interested readers are referred to Mikulincer and Shaver 2016; also see Supplementary data for additional details.)

Figure 1.

Attachment sub-types as a function of anxiety and avoidance (adapted from Mikulincer and Shaver 2016).

Agency and communion were assessed with the Extended Personality Attributes Questionnaire (Spence, Helmreich, and Holahan 1979). This questionnaire consists of 32 characteristics describing aspects of the self and their opposite/contradiction (e.g., ‘very independent’ and ‘not at all independent’), with the letters A through E between the anchors. Participants select the letter that ‘corresponds to where they fall on the spectrum’. Eight items assess agency (‘independent,’ ‘active,’ ‘competitive,’ ‘never gives up easily,’ ‘feels very superior,’ ‘makes decisions easily,’ ‘self-confident,’ and ‘stands up well under pressure’) and eight items assess communion (‘kind,’ ‘gentle,’ ‘able to devote self to others,’ ‘helpful to others,’ ‘aware of other’s feelings,’ ‘understanding of others,’ ‘warms in relation to others’ and ‘emotional’). The EPAQ also includes undesirable forms of agency and communion but, following Bartz et al. (2015), we focused on the positive items.

Loss of own hand was assessed with the Rubber Hand Questionnaire (Longo et al. 2008). This questionnaire consists of 18 items assessing participants’ subjective sense of the illusion across three domains: embodiment of the rubber hand, loss of own hand, and movement. Participants rate their agreement with each statement using a seven-point scale ranging from −3 (strongly disagree) to 3 (strongly agree). Because we were interested in assessing reductions in agency, our primary outcome was the five-item loss of own hand subscale (e.g., ‘it seemed like I was unable to move my hand,’ ‘it seemed like my hand had disappeared,’ and ‘it seemed like my hand was out of my control’); as noted in footnote 2, this subscale, specifically, has been associated with self-reported agency as assessed by the EPAQ (see Supplementary Table S2). However, we also conducted exploratory analyses on the embodiment subscale (see Supplementary Table S3).

Data analysis

We used multilevel modelling to test our hypotheses. Binary variables were computed to estimate the effect of drug (Oxytocin = 1; Placebo = 0), and RHI stimulation condition (Asynchronous = 1; Synchronous = 0), drug order (Oxytocin 1^st = 1, Placebo 1^st = 0), and RHI stimulation condition order (Asynchronous 1^st = 1, Synchronous 1^st = 0). Attachment anxiety and attachment avoidance were mean centred. All models met assumptions for multilevel modelling (see online data and materials).

Our first two models tested the replication effects; specifically, that following oxytocin (vs. placebo): 1) more (vs. less) avoidantly attached individuals describe themselves as more communal (oxytocin × attachment avoidance interaction predicting communion) and 2) more (vs. less) anxiously attached individuals describe themselves as less agentic (oxytocin × attachment anxiety interaction predicting agency). Following Bartz et al. (2015), each model included attachment avoidance, attachment anxiety, and drug condition, and all two-way interactions, as well as age (a covariate). We initially controlled for drug administration order; however, order was not a significant predictor in either model, so we report the more parsimonious model without drug order (drug order was not a significant predictor in Bartz et al.’s study either so was removed from their model). Finally, Bartz et al. (2015) also tested the drug × avoidance × anxiety interaction (as is standard when looking at attachment effects and moderated attachment effects); however, the interaction effect was not significant, so was removed from their final analysis. For the sake of completeness, we included this three-way interaction, as well as the avoidance × anxiety interaction (as noted in the Methods section, including the interaction also allowed us to look at the secure group (low anxiety + low avoidance) for comparisons with prior work). A random intercept was added to test for random variance between participants in both models (1|Subject). In both models, entering a random effect for drug condition resulted in a failure to converge; thus, no random slopes were estimated in models 1 and 2.

In model 3, we focused on our measure of bodily agency. Model 3 was similar to models 1 and 2 but included the additional RHI variables. Specifically, we regressed loss of own hand on attachment anxiety and attachment avoidance, with drug condition (oxytocin vs. placebo) and RHI stimulation condition (synchronous vs. asynchronous) entered as fixed factors of interest. We also included the drug × attachment anxiety interaction. Note: in an initial model, we included the drug × attachment avoidance interaction, the drug × anxiety × avoidance interaction, and we controlled for drug administration order and RHI stimulation condition order. As this produced a highly complex model with several non-significant effects, we pared it down, removing the non-significant three-way interaction, the non-significant drug × attachment avoidance interaction, and the drug and RHI order variables, which were not related to own hand loss. The pattern of results did not change throughout this process; thus, we report the more parsimonious model below. Model 3 included a random intercept for individual subjects, with random slopes for drug condition and RHI stimulation condition.

Data were analysed using R (R Core Team 2023); multilevel models were estimated using the lmer function in the lme4 package (Bates et al. 2015). All models were fit using restricted maximum likelihood and degrees of freedom were estimated using Satterthwaite’s method, and the variance-covariance structure was estimated using the default settings in lmer. P-values for fixed effects were calculated using lmertest. Simple slopes for three-way interactions were calculated using the simple_slopes function in reghelper, and simple slopes for two-way interactions were calculated using the sim_slopes function in the interactions package. Profile likelihood confidence intervals for all fixed effects were calculated using the confint function in the stats package. Finally, figures were created using the package ggplot, version 3.2.1 (Wickham 2016).

Results

Model 1: Effects of oxytocin and attachment (anxiety and avoidance) on communion

To determine the degree of variance in communion scores attributable to clustering, we first ran a null model; 86% of variance in communion was attributed to clustering (ICC = 0.86), supporting the use of multi-level modelling for these data.

Table 1 summarizes the results of model 1. The model intercept represents the average level of communion for the average participant (average level of attachment anxiety, attachment avoidance, and age) in the placebo condition, controlling for drug/placebo administration order. The average participant had communion score of 3.89 (out of 5) following placebo. There was no conditional effect of oxytocin on communion; however, there was a significant conditional effect of attachment avoidance, such that for every one-unit increase in attachment avoidance above the mean there was a 0.21 unit decrease in communion in the placebo condition for participants with mean levels of attachment anxiety and age. This is consistent with the preliminary findings in Bartz et al. (2015) showing a negative association between attachment avoidance and communion in a non-drug sample (their preliminary study). The two-way interaction effects were not significant; however, there was a significant three-way interaction between attachment anxiety, attachment avoidance, and drug.

Table 1.

Results of multilevel model analyses predicting communion.

	Communion
Predictors	Estimates	Std. error	CI	Statistic	P
Intercept	3.91	0.05	3.81, 4.01	t(79) = 77.53	<.001
Attachment anxiety	0.01	0.04	−0.07, 0.09	t(79) = 0.33	.743
Oxytocin	0.03	0.02	−0.02, 0.07	t(220) = 1.19	.235
Attachment avoidance	−0.21	0.04	−0.30, −0.13	t(79) = −4.85	<.001
Age	0	0.01	−0.03, 0.03	t(72) = 0.12	.906
Attachment anxiety × oxytocin	0.01	0.02	−0.02, 0.05	t(220) = 0.72	.474
Attachment anxiety × avoidance	−0.02	0.03	−0.08, 0.04	t(78) = −0.61	.546
Attachment avoidance × oxytocin	−0.02	0.02	−0.06, 0.02	t(220) = −0.98	.329
Attachment anxiety × avoidance × oxytocin	0.05	0.01	0.03, 0.08	t(220) = 3.93	<.001
Random effects
σ2	0.03
τ00 Subject	0.17
N Subject	77
Observations	300
Marginal R2/conditional R2	0.25/0.88

Bold indicates statistically significant effects.

To probe the nature of this interaction, we conducted post-hoc simple slopes analyses; Table 2 reports the effect of drug at each level of anxiety and avoidance (low, average, high). Results showed a significant, negative slope of drug for high avoidant/low anxious (‘dismissive’) participants, such that oxytocin (vs. placebo) decreased communion by 0.09 units for these individuals. Conversely, there was a significant, positive slope of drug for high avoidant/high anxious (‘fearful’) participants, such that oxytocin increased communion by 0.1 units for these individuals. There was also a significant, positive slope of drug for low avoidant/low anxious (‘secure’) participants, such that oxytocin increased communion by 0.1 units for these individuals. There were no other significant simple slopes. See Fig. 2 for a visualization of the results for each attachment profile.

Table 2.

Simple effects of oxytocin on communion and agency at each level of attachment anxiety and avoidance.

		Communion
Anxiety	Avoidance	Category	Est.	Std. error	95% CI	Statistic	P-value
Low (−1.32)	Low (−1.17)	‘Secure’	0.11	0.04	0.04, 0.19	t(219) = 2.97	.003
Low (−1.32)	Mean (−0.01)		0.01	0.03	−0.05, 0.07	t(220) = 0.36	.717
Low (−1.32)	High (1.14)	‘Dismissive’	−0.09	0.05	−0.18, 0.00	t(219) = −1.99	.048
Mean (−0.02)	Low (−1.17)		0.05	0.03	−0.01, 0.11	t(220) = 1.56	.120
Mean (−0.02)	Mean (−0.01)		0.03	0.02	−0.02, 0.07	t(220) = 1.19	.234
Mean y (−0.02)	High (1.14)		0.00	0.03	−0.06, 0.06	t(220) = 0.12	.908
High (1.29)	Low e (−1.17)	‘Preoccupied’	−0.02	0.05	−0.11, 0.07	t(220) = −0.37	.716
High (1.29)	Mean (−0.01)		0.04	0.03	0.02, 0.10	t(220) = 1.32	.187
High (1.29)	High (1.14)	‘Fearful’	0.10	0.04	0.02, 0.18	t(221) = 2.32	.021

		Agency
Anxiety	Avoidance	Category	Est.	Std. error	95% CI	Statistic	P-value
Low (−1.32)	Low (−1.17)	‘Secure’	0.08	0.03	0.02, 0.15	t(219) = 2.46	.015
Low (−1.32)	Mean (−0.01)		0.03	0.03	0.02, 0.08	t(219) = 1.07	.285
Low (−1.32)	High (1.14)	‘Dismissive’	−0.02	0.04	−0.10, 0.05	t(219) = −0.60	.548
Mean (−0.02)	Low (−1.17)		−0.02	0.03	−0.07, 0.03	t(219) = −0.69	.488
Mean (−0.02)	Mean (−0.01)		−0.03	0.02	−0.07, 0.01	t(219) = −1.47	.144
Mean (−0.02)	High (1.14)		−0.04	0.03	−0.09, 0.02	t(220) = −1.37	.172
High (1.29)	Low (−1.17)	‘Preoccupied’	−0.12	0.04	−0.20, −0.04	t(219) = −3.01	.003
High (1.29)	Mean (−0.01)		−0.09	0.03	−0.14, −0.03	t(220) = −3.15	.002
High (1.29)	High (1.14)	‘Fearful’	−0.05	0.04	−0.12, 0.02	t(221) = −1.36	.177

Bold indicates statistically significant effects.

Figure 2.

Effects of oxytocin (vs. placebo) on communion for each attachment profile. To aid data visualization, we categorized participants into attachment profiles based on whether their score was above or below the sample mean in anxiety and avoidance: dismissive (high avoidance, low anxiety), fearful (high avoidance, high anxiety), preoccupied (low avoidance, high anxiety), and secure (low avoidance, low anxiety). Raw communion scores are plotted on the Y-axis with separate plots for each attachment profile following placebo and oxytocin administration. Dots = individual communion scores, distributions on left of each panel (purple) = scores following placebo, and distributions on right of each panel (yellow) = scores following oxytocin. Dots connected by dashed lines (red) = group mean for each drug condition. Solid lines within each plot = the median. The whiskers on each plot specify the 25–75 percentiles. The two high avoidant profiles (dismissive and fearful) reported less communion overall compared to the low avoidant profiles (preoccupied and secure). Oxytocin resulted in a decrease in communion for dismissive participants, whereas oxytocin increased communion for fearful and secure participants. Oxytocin did not impact the level of communion reported by preoccupied participants. (Note: This figure provides visualization of the data and is not based on any specific analysis performed.)

Model 2: Effects of oxytocin and attachment (anxiety and avoidance) on agency

As in model 1, we first ran a null model indicating that 89% of the variance in agency can be attributed to clustering (ICC = 0.89) and supporting the use of a multi-level modelling. The model intercept represents the average level of agency for the average participant (average level of attachment anxiety, attachment avoidance, and age) in the placebo condition.

Table 3 summarizes the results of model 2. The average participant had an agency score of 3.46 (out of 5) on the placebo day. There was no conditional effect of oxytocin on agency; however, there was a significant conditional effect of attachment anxiety on agency, such that for every one-unit increase in attachment anxiety above the mean there was a 0.14-unit decrease in the degree of agency reported in the placebo condition for participants with mean levels of attachment avoidance, and age. Again, this is consistent with Bartz et al. (2015) observed negative association between attachment anxiety and agency in a non-drug sample (their preliminary study). There were no conditional effects of attachment avoidance or age (Table 3).

Table 3.

Results of multilevel model analyses predicting agency.

	Agency
Predictors	Estimates	Std. error	CI	Statistic	P
Intercept	3.46	0.05	3.36, 3.56	t(77) = 67.47	<.001
Attachment anxiety	−0.15	0.04	−0.22, −0.07	t(77) = −3.64	<.001
Oxytocin	−0.03	0.02	−0.07, 0.01	t(219) =−1.51	.134
Attachment avoidance	0	0.04	−0.09, 0.09	t(77) = −0.06	.95
Age	−0.01	0.02	−0.04, 0.02	t(72) = −0.79	.431
Attachment anxiety × oxytocin	−0.04	0.01	−0.07, −0.01	t(219) = −2.95	.004
Attachment anxiety × attachment avoidance	−0.04	0.03	−0.10, 0.03	t(77) = −1.10	.274
Attachment avoidance × oxytocin	−0.01	0.02	−0.04, 0.03	t(220) = −0.45	.655
Attachment anxiety × attachment avoidance x oxytocin	0.03	0.01	0.01, 0.05	t(220) = 2.47	.014
Random effects
σ2	0.03
τ00 Subject	0.18
N Subject	77
Observations	300
Marginal R2/conditional R2	0.193/0.900

Bold indicates statistically significant effects.

Additionally, consistent with Bartz et al. (2015), there was a significant two-way interaction between attachment anxiety and drug. Simple slopes analysis on this interaction revealed that the slope for attachment anxiety was significant and negative in both conditions but significantly stronger in the oxytocin condition (placebo: β = −0.14 (95% CI: −0.22, −0.07), t(219) = −3.64, P = 0.001; oxytocin: β = −0.19 (95% CI: −0.27, −0.11), t(219) = −4.73, P < 0.001). Essentially, oxytocin potentiated the negative association between attachment anxiety and agency.

Furthermore, we observed a significant three-way interaction between attachment anxiety, attachment avoidance, and drug. [Note: the attachment anxiety by oxytocin interaction remains significant when the three-way, attachment anxiety × attachment avoidance × oxytocin interaction is removed from the model (B = −0.05, P = .002).] Again, we conducted simple slopes analyses to probe this effect. Table 2 reports the effect of drug at each level of anxiety and avoidance. Results showed a significant, negative slope of drug for low avoidant/high anxious (‘preoccupied’) participants, such that oxytocin decreased agency by 0.12 units for these individuals. Results also showed a significant, negative slope of drug for average avoidant/high anxious participants, such that oxytocin decreased agency by 0.09 units for these individuals. Results showed a similar pattern for high avoidant/high anxious (‘fearful’) participants, but the effect not statistically significant (β = −0.05; P = 0.177). Conversely, there was a significant, positive slope of drug for low avoidant/low anxious (‘secure’) participants, such that oxytocin increased agency by 0.08 units for these individuals. There were no other significant simple slopes. See Fig. 3 for a visualization of the results.

Figure 3.

Effects of oxytocin on agency for each attachment profile. Interpretation of the features of Fig. 3 is identical to Fig. 2, with the exception that the Y-axis represents agency. Anxiously attached participants (i.e. fearful and preoccupied) reported less agency overall compared to the low anxious profiles (dismissive and secure). Oxytocin resulted in a decrease in agency for preoccupied participants; a similar pattern was observed for fearful participants, but the effect was not statistically significant. Conversely, oxytocin resulted in an increase in agency for secure participants. Oxytocin did not impact the level of agency reported by dismissive participants. (Note: This figure provides visualization of the data and is not based on any specific analysis performed.)

Model 3: Effects of oxytocin and attachment (anxiety and avoidance) on own hand loss

The null model predicting own hand loss indicated that 54% of the variance in own hand loss can be attributed to clustering (ICC = 0.54). Thus, a multilevel modelling approach is appropriate.

The results of model 3 are presented in Table 4. We observed a statistically significant conditional effect of RHI condition, such that following placebo, own hand loss ratings were 0.5 units lower (indicating less loss), on average, in the asynchronous (vs. synchronous) condition (β = −0.49 (95% CI: −0.72, −0.26), t(141.18) = −4.16, p <.001), which confirms that we successfully manipulated the illusion via simulation. We also observed a significant conditional effect of anxious attachment, such that in the placebo condition, a one-unit increase in anxious attachment (relative to the mean) resulted in a 0.25 unit increase in loss of own hand. However, there was no conditional effect of oxytocin (β = 0.13; P = 0.286), and the drug × attachment anxiety interaction was not significant (β = 0.076; P = 0.392). Essentially, the pattern was the same in the placebo and oxytocin conditions—specifically, simple slopes analysis also revealed a significant, positive effect of anxious attachment on loss of own hand in the oxytocin condition (β = 0.31 (95% CI: 0.156—0.472), t(71.11) = 3.91, P = .008). In sum, as with self-conceptions of agency, anxiously attached individuals also show a weaker sense of bodily agency, but this effect was not potentiated by oxytocin. Finally, there was an unexpected RHI order effect such that receiving the asynchronous condition first resulted in a 0.6 unit increase in loss of own hand. [Regarding our exploratory analyses on the embodiment subscale, there was a conditional effect of stimulation condition, such that following placebo, asynchronous (vs. synchronous) stimulation resulted in less embodiment of the rubber hand (β = −1.06; P < .001). Other than an unexpected effect of drug administration order (β = 0.26; P < .005), there were no other significant effects, although there was a trending effect of attachment anxiety such that more anxious participants reported greater embodiment (β = 0.15, t(71) = 1.76, p = .083). See Supplementary data for complete results.]

Table 4.

Results of multilevel model analyses predicting loss of own hand.

	Loss of own hand
Predictors	Estimates	Std. error	CI	Statistic	P
Intercept	−1.50	0.19	−1.86, −1.13	t(105) = −8.08	<0.001
Oxytocin	0.13	0.12	−0.11, 0.38	t(129) = 1.07	0.286
Asynchronous (RHI stimulation)	−0.49	0.12	−0.73, −0.26	t(141) = −4.16	<0.001
Attachment anxiety	0.25	0.09	0.06, 0.43	t(72) = 2.69	0.009
Attachment avoidance	−0.02	0.10	−0.22, 0.19	t(73) = −0.16	0.871
RHI stimulation condition order	0.58	0.20	0.19, 0.98	t(72) = 2.96	0.004
Attachment avoidance × oxytocin	−0.02	0.09	−0.19, 0.15	t(70) = −0.22	0.826
Attachment anxiety × oxytocin	0.07	0.08	−0.09, 0.22	t(69) = 0.86	0.392
Asynchronous (RHI stimulation) × oxytocin	−0.27	0.14	−0.56, 0.01	t(74) = −1.89	0.062
Random effects
σ2	0.39
τ00 Subject	1.39
τ11 Oxytocin	0.34
τ12   Asychronous (RHI Stimulation)	0.25
ρ01	−0.55
ρ02	−0.74
N Subject	77
Observations	300
Marginal R2/conditional R2	0.19/0.76

Bold indicates statistically significant effects.

Discussion

Prior research shows that oxytocin increases a communal orientation, characterized by warmth, kindness and understanding, especially for avoidantly attached individuals who typically lack communion. Additionally, oxytocin has been shown to selectively decrease agency—a sense of efficacy, empowerment, and self-confidence—for anxiously attached individuals who typically lack agency. These findings may explain some peculiarities in research investigating the social effects of oxytocin—namely, that oxytocin is helpful (socially) for those who are less socially motivated and/or attuned but can exacerbate insecurities in those who are preoccupied with closeness. Overall, we replicated these effects, with additional nuance.

Oxytocin again enhanced communion for more avoidant individuals, but this effect was specific fearful avoidants—in fact, oxytocin attenuated communion for dismissive avoidants. In contrast to fearful individuals, who want interpersonal closeness but fear rejection, dismissive individuals are thought to have no such desires. Our findings suggest that the beneficial effects of oxytocin on communion for avoidant individuals may require an intact desire for social connection (Bartz et al. 2015). Relatedly, it may be difficult to augment communion in dismissive individuals because they have successfully deactivated their attachment system. Fraley and Shaver (1997) observed a similar dissociation in their study investigating attachment and the suppression of unwanted thoughts (i.e. suppressing thoughts about their romantic partner leaving them). In contrast to fearful participants—who evidenced distress during this thought suppression task—dismissive participants showed no rebound effect or physiological arousal, suggesting attachment system deactivation. Oxytocin may only augment a communal orientation if the attachment system is not deactivated. That said, these more nuanced between-subject effects should be treated with caution until replicated. Finally, we found that oxytocin also augmented communion for secure individuals, who reported greater warmth, kindness and understanding following oxytocin, which is consistent with the prosocial effects reported in the literature.

We also replicated the oxytocin-attachment anxiety-agency effect, again with additional nuance. Anxious and especially preoccupied individuals showed a decrease in agency following oxytocin. More precisely, attachment anxiety/preoccupation was negatively associated with agency in the oxytocin and placebo conditions, but the effect was stronger in the former condition, suggesting that oxytocin potentiates the association between interpersonal anxiety and (lack of) agency. This observation bolsters the idea that for anxious/preoccupied individuals, enhancing affiliation may further threaten their sense of agency. A similar but non-significant pattern was observed for the fearful profile; of note, fearful participants reported the lowest level of agency overall; the lack of an effect for fearful participants may thus be due to floor effects (indeed, analyses collapsing across the preoccupied and fearful profiles showed a similar and significant pattern to that observed for the preoccupied profile). Finally, oxytocin increased agency for secure participants; together with the secure-communion effect, this observation suggests that oxytocin may boost positive self-impressions in general for secure individuals. Notably, this oxytocin-secure-agency effect is consistent with the attachment theory notion of a ‘secure base’—i.e. that the caregiver’s presence gives (secure) individuals confidence to explore the environment.

Turning to our measure of bodily agency, we observed a similar pattern to that for conceptual agency, although the drug by anxiety interaction was not significant. As with conceptual agency, anxiously attached individuals reported less bodily agency—specifically, experiencing greater loss of own hand during the rubber hand illusion even when the illusion is not warranted (in the asynchronous condition). The present work thus shows that attachment anxiety is associated with reduced conceptual and interpersonal agency, but also embodied agency. To our knowledge, this is the first study to link attachment anxiety with losses in the bodily self. Notably, experiences of attachment security in childhood are thought to play a critical role in the development of self-organization (Fonagy et al. 2004). Our findings support this idea and suggests that attachment continues to play a role in self-organization—both at the conceptual and bodily level—in adulthood. Our observation of a link between attachment anxiety and (lower) bodily agency also aligns with clinical theory and research on borderline personality disorder, which is associated with (extreme) attachment anxiety (Bartz et al. 2011) and identity disturbance/unstable sense of self (American Psychiatric Association 2013).

As noted at the outset, it is thought that oxytocin functions as an endocrine code, cueing organisms to engage in affiliation (Mu et al. 2016). This motivational state may then interact with people’s unique interpersonal contingencies, differentially affecting agency and communion. Alternatively (or in addition), it is thought that oxytocin primes neural circuits to be more sensitive to social inputs (Marlin et al. 2015); for more anxious individuals, this priming may bring to mind interpersonal loss (cf Bartz et al. 2010), which then threatens their sense of self. The present findings are consistent with these ideas but highlight that heightening affiliative strivings and/or sensitivity to social inputs can produce different results depending people’s relational histories and resulting interpersonal expectancies and coping strategies (Bartz 2016).

In conclusion, we replicate two previously reported interaction effects of oxytocin—namely, that oxytocin augments communion for more avoidantly attached individuals, and oxytocin further decreases agency for more anxiously attached individuals. We also show that this effect extends to women, an important contribution to the oxytocin literature. That said, our sample was relatively young (< 40 years) and consisted of mostly White/European and Asian participants, so future work is needed to test these effects in older and/or other racial/ethnic samples. Moreover, notwithstanding the merits of our within subjects design to test individual differences in response to oxytocin, our sample size may not have been adequate to separately assess the four different attachment profiles, so the more nuanced between-person effects (i.e. attachment subtypes) should be treated with caution until replicated. In addition to replication, we also sought to extend this prior research by looking at the effects of oxytocin and attachment on bodily agency. Overall, more anxious individuals experienced greater loss of hand during the rubber hand illusion paradigm, although oxytocin did not potentiate this effect. In addition to shedding light on some counterintuitive effects of oxytocin observed in the literature, this work highlights the tension that can arise in satisfying needs for communion and agency, or security and authenticity. Most of the time, these two motives (or needs) operate independently; however, for some individuals, or in some contexts, they can come into conflict.

Author contributions

Jennifer Bartz (Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Writing—original draft and review/editing, Supervision, Project Administration, Funding acquisition), Willis Klein (Data curation, Formal Analysis, Validation, Visualization, Writing—original draft and review/editing), Amy Gregory (Formal analysis, writing—original draft and review/editing)

Supplementary data

Supplementary data are available at SCAN online.

Conflict of interest: The authors have no interests to declare.

Funding

This research was funded by a Natural Sciences and Engineering Research Council (NSERC) Development Grant (#428387) awarded to J. A. Bartz.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.