The uncertainties regarding central vs. peripheral role of OXT are compounded by interpretational complexities due to the many psychological mechanisms that may mediate OXT effects on social behavior. One critical question is whether OXT effects on social behavior reflect its influences on specialized high-order social cognitive processes (e.g., trust, generosity, suspiciousness, mentalizing) or relatively broad states and orientations (e.g., general anxiety, affiliative motivation, global saliency of social cues). One reason to raise this question is the issue of explanatory parsimony. But this question is also important from the perspective of what cognitive science and psychology typically assume about the mind. On these standard assumptions, one would expect neurotransmitters and hormones, especially those acting peripherally, to work via relatively broad modulation, rather than via qualitative changes of computations on specific, higher-order content. Yet, many interpretations in human research on OXT and social behavior, and some of the public appeal of this research, rest on the possibility that this hormone and neurotransmitter selectively targets “the social brain” and has qualitatively distinct effects on complex, higher-order social-cognitive processes.
In what follows, we will argue that one such broad factor, anxiety, may account for a good chunk of the seemingly specific OXT effects on social cognition. We should say at the outset that, as outsiders to the field of OXT research but eager consumers of its findings, we find this possibility important not because of our particular interest in anxiety. Rather, we consider this possibility because it represents a genuine puzzle about the nature of the relationships between OXT and social cognition, and, more generally, the selectivity of mechanisms by which hormones and neurotransmitters influence higher-order mental processes.
3A. Anxiolytic Properties of Oxytocin
As is well known, anxiolytic effects of OXT have been demonstrated in a variety of species. These effects occur both after exogenous OXT administration and after endogenous release (Neumann, 2008
). It is likely that some mechanisms involve primarily central routes. For example, amygdala is rich in OXT receptors (Huber et al., 2005
). Some anxiolytic effects likely involve primarily peripheral routes. For example, OXT suppresses the “classic” stress hormones of the hypothalamic–pituitary–adrenal axis (Heinrichs et al., 2006
). Most likely, however, anxiolytic effects reflect a complex interaction of central and peripheral mechanisms. For example, OXT alters cardiovascular reactivity – a peripheral effect. However, OXT achieves this effect not only via its actions on the heart itself, but also centrally – via nucleus of the solitary tract (NTS), which integrates and relays incoming peripheral visceral inputs with central influences (Norman et al., 2011
). Further, though intranasal OXT administration reduces amygdala activity (Kirsch et al., 2005
), this “central” effect may involve peripheral influences (e.g., via vagal stimulation, Hassert et al., 2004
). In short, it is clear that a reduction of physiological and psychological reactivity to stressors is a common consequence of OXT, with many pathways leading to this outcome. These effects are what have motivated Sue Carter to describe chronic exposure to oxytocin as a “physiological metaphor for safety” (Carter, personal communication 2011).
3B. OXT Effects on Specific, Higher-Order Social Cognitive Processes
Given the anxiolytic properties of OXT, it is reasonable to wonder, as researchers have long wondered, to what extent specific higher-order social-cognitive effects observed in humans are due to OXT’s general anxiolytic effect. Consider some well-known findings. After intranasal OXT administration, participants display more trust in economic game involving allocating money to a stranger, with an anticipation of receiving greater returns (Kosfeld et al., 2005
), rate strangers’ faces higher on trustworthiness (Theodoridou et al., 2009
), receive higher “mind-reading” scores in a task that involves interpreting strangers eyes (Domes et al., 2007
), and show more in-group favoritism in a prisoner-dilemma task that involves allocation of money between arbitrarily assigned ‘in-group’ or “out-group” (De Dreu et al., 2010
). Note that all these effects involve high-order psychological processes (“trust”, “mentalizing”, “cooperation”) and are typically interpreted as suggesting that OXT selectively targets circuitry involved in sophisticated social-cognitive computations.
Findings suggesting that OXT influences complex social-cognitive circuitry are encountered also in translational research into mental health issues. Several studies now report that OXT administration positively influences complex psychiatric disorders such as autism and schizophrenia (for review, see Meyer-Lindenberg et al., 2011
). For example, in one study, schizophrenic patients received either 3 weeks of daily intranasal OXT (40 IU twice a day) or a placebo adjunctive to prescribed antipsychotics (Feifel et al., 2010
). Compared with the placebo + antipsychotics medications, the OXT group showed greater reduction of positive symptoms and, less robustly, negative symptoms of schizophrenia, as measured by standard scale PANSS, (Kay et al., 1987
). The positive symptoms in this scale include things like hallucinations, delusions, suspiciousness (paranoia), thoughts disorders, excitement, and hostility, whereas negative symptoms include flat affect, poor rapport, social withdrawal, etc. Endogenous peripheral OXT levels also have been correlated with severity of symptoms on the PANSS, especially in women (Rubin et al., 2010
). These results are particularly intriguing, as they imply that OXT may influences the very core mechanisms of belief formation, perhaps allowing one to reconceptualize schizophrenia as partially a disorder of “social trust”.
So, yes, on first glance, these findings clearly suggest that OXT modulates complex social cognitive circuitry. Nevertheless, might non-specific reduction in anxiety contribute in a nontrivial way to all those effects? Let us look at some studies in more detail in light of this possibility, and discuss some counterarguments.
3C. But Oxytocin Works on Things that have Nothing to do with Anxiety
An understandable response to the challenge that anxiety reduction plays a major role in the results of administering nasal OXT to human subjects is to attack the very plausibility of such challenge. After all, it will be argued, the dependent measures in the above OXT studies seem far removed from anything to do with anxiety -- trust, generosity, mental state attributions, autism, or schizophrenia. On one view, the core feature connecting these constructs is the willingness or capacity to make inferences, especially favorable inferences, about other people’s dispositions and intentions. For example, such social inferences underlie subjects’ expectation that their partner will return, rather than pocket their money (i.e., trust, Kosfeld et al., 2005
); subjects’ interpretation of mental state from a stranger’s eyes (i.e., mentalizing, Domes et al., 2005), or suspiciousness about other people’s goals in schizophrenia (Feifel et al., 2010
). Why should these social-cognitive capacities be anxiety-sensitive?
One rebuttal takes the following form. In general, any social inference, and especially positive inference, is preconditioned on the subject’s willingness to engage in social interaction, and this willingness is anxiety-sensitive. That is, changes in anxiety levels may influence broad preconditions for many complex social computations, from basic social perception to understanding to greeting to cooperating to mating. As a result, though the initial cause might be quite simple and general, the downstream effects may be a quite specific, involving complex cognition and behavior. Consistent with this idea, evidence suggest that the effects observed in the trust studies do indeed appear to involve fairly general mechanisms of anxiety down-regulation, as mediated by the amygdala (Baumgartner et al., 2008
). Other recent work also suggests that complex mental state inferences in the mind-in-the-eye tasks depend on the general willingness to look into strangers’ eyes, which in turn appear anxiety-dependent (Evans et al., 2010
Our rebuttal also applies to the schizophrenia studies. As mentioned, early interpretations proposed that the improvements in positive symptoms were due to selective changes in trust. In practice, however, the assessment of these positive symptoms is hard to separate from the patients’ general willingness to engage with others. Notice that the rating of positive symptoms of schizophrenia on a scale like PANSS (Kay et al., 1987
) is performed by a psychiatrist who asks the patient questions about his social interactions (e.g., “do you talk to other people?”) and observes his/her interactions with others (e.g., does the patient show expressions of anger, resentment, sarcasm?). If a patient is less anxious, and thus more socially outgoing, he may show improvements on the index of positive symptoms, even with any reduction in the very core aspects of schizophrenia (e.g., auditory hallucinations, disorganized thought systems). Indeed, it’s well known that improvements in interpersonal engagement can occur even with low doses of standard antianxiety medications, or SSRIs (Knutson et al., 1998
). Consistent with our perspective, more recent and comprehensive interpretations of OXT effects on schizophrenia and autism tend to emphasize OXT’s broader effects on anxiety and affiliative motivation (Meyer-Lindenberg et al., 2011
3D. But We Did Control for Anxiety
Several researchers worry enough about the potential role of anxiety that they try to control for it, typically using a questionnaire. For example, in the work on trust (Kay et al., 1987
) and in the work on mentalizing (Domes et al., 2007
) researchers have found no significant effects of OXT on subjective experience of anxiety, as assessed by a German-language questionnaire called MDBF -- Multidimensional Mood State Questionnaire (Steyer et al., 1997
), available in English translation here: http://www.metheval.uni-jena.de/mdbf.php
. Accordingly, the researchers argued that anxiety reduction cannot be the underlying mechanisms for the reported high-order social cognitive effects. A closer look suggests there are several problems with this argument and the deployed methods.
First, the null findings in these particular behavioral experiments are puzzling given that other papers do report anxiolytic effects of OXT, even with similar doses and similar dependent measures. For example, in the context of stress-inducing situation (giving a public speech), OXT administration reduces self-reported anxiety using the MDBF questionnaire (Heinrichs et al., 2003
). A second and related issue concerns validity, or how accurately these measures reflect the relevant state. For example, the MDBF questionnaire employed to control for potential effects on anxiety in the trust research does not contain any question that specifically asks about anxiety, but rather assesses a general state of “calmness.” A third issue is that broad mood questionnaires, like the MDBF, tend to be rather insensitive, especially when it comes to relatively mild states. For example, in the English-speaking world, a questionnaire most similar to MDBF is PANAS (Watson et al., 1988
). This broad questionnaire is rather insensitive to mild states, qualitatively differentiated states, and also fails to accurately detect some important negative states, such as anger (Harmon-Jones et al., 2009
). Fourth is the tricky possibility that changes in low-level affective states do not always lead to changes in conscious experience. As a result, such states do not manifest on a questionnaire, even though they do manifest in behavior as when participants respond differently to relevant cues (Winkielman and Berridge, 2004
). For example, it is well known in psychiatry that anti-anxiety or anti-depressive medication can sometimes improve patients’ social behavior (e.g., greater sociability), before it improves self-reported affective experience (e.g., reduction in subjective feelings of sadness or anxiety).
To some extent, these problems can be addressed by using physiological measures that more directly tap into biological mechanisms of anxiety. One such measure, for example, is affective startle modulation, where startle responses are assessed in the presence of an emotionally-relevant stimulus. Fittingly, a recent study on rats showed that fear-potentiated startle response is reduced after peripheral administration of OXT (Ayers et al., 2011
; Missig et al., 2010
). Relevant to our concerns, in these studies the specific aspect of anxiety that was influenced was background anxiety. It would be thus important to know whether the specific OXT effects on trust, mental inferences from the eyes, or schizophrenia would hold when statistically controlling for the effects on such sensitive anxiety measures.
3E. But Oxytocin Does Work Selectively
The issue of non-specific anxiety may seem initially irrelevant for studies which report OXT effects that are interactive, i.e., selectively influence the targeted behavior but not unrelated behavior. For example, in the trust studies, research showed OXT effects on social trust, measured by amount invested, in a condition when participants played with another person, but not on nonsocial “trust”, in a condition when participants played against a computer (Kosfeld et al., 2005
). Another recent study showed that OXT increases variables related to ingroup-favoritism, but no reduction in variables related to outgroup derogation (De Dreu et al., 2010
). Importantly, these selective effects were pivotal for both papers’ argument against a more parsimonious, general-state interpretation.
Appealing as dissociations are, any argument from a dissociation has to meet high standards, as has long been appreciated (Teuber, 1955
). For the argument to succeed it is necessary that both measures, or both conditions have reasonably equal sensitivity and that subjects are equally attentive and motivated to respond on both measures or in both conditions. These are fairly steep requirements. A dissociation argument is weakened if the measure or condition where the OT was absent was perceived by subject to be less important, more boring, more confusing, etc. Note therefore that in the Kosfeld et al study, the comparison was human risk vs. computer risk – conditions that clearly differ in their interest potential and anxiety-inducing properties. The DeDreu et al study, has been reinterpreted to explain the observed dissociation in the ingroup/outgroup behavior results in terms of exactly such non-specific factors (Chen et al., 2011
More generally, both biological and psychological factors can easily turn a relatively broad physiological effect into what may deceivingly appear to be a narrow behavioral effect. This hazard is, indeed, “old news”, well appreciated at least since the famous studies on two-factor theory of emotion showed that enhancement of relatively non-specific arousal (via epinephrine injection) can channel into a variety of emotion-related behaviors and experiences based on subtle contextual cues (Schacter and Singer, 1962
). A more recent, and more relevant example comes from a recent study on lactating human mothers, which presumably have higher levels of OXT (Hahn-Holbrook et al., 2011
). Compared to controls, lactating mothers were found to be simultaneously less stressed, but at the same time more aggressive. The proposed explanation is that fear usually has aggression-constraining properties, so, paradoxically, OXT-related fear reduction, in proper context may lead to greater aggressive behavior. The role of psychological situational factors in channeling OXT effects was recently emphasized by a comprehensive review that pointed out the many inconsistent findings in the human OXT literature can be at least partly understood as reflecting contextual influences (Bartz et al., 2011
3F. Anxiolytics as Control
Given the above, it is unfortunate that human OXT studies rarely control for anxiety, and when they do, these controls are weak. It is also surprising that, so far as we know, very few human studies, if any, include anti-anxiety substances as a control. This is regrettable since rat studies show that the effects of oxytocin and benzodiazepines, for example, can be quite similar (Neumann, 2008
). One interesting control in human studies would be different medications for treatment of anxiety disorders. In the context of the earlier section (2), it would be especially interesting to contrast medications known to work peripherally (e.g., beta-blockers, like Propranolol) and centrally (e.g., Lorazepam). Finally, it would be critical to include measures that differentiate between social and non-social effect of such interventions.
3G. Parsimony, specification, and discriminative validity
Perhaps the core issue discussed in this section is an instance of the more general scientific problem of phenomenon description: specificity versus parsimony and consilience with the rest of the relevant science. To put it another way, how in science should we select a characterization for a phenomenon to avoid misspecification and instead maximize accuracy and specificity (construct and discriminative validity), along with its heuristic value and fit across different levels of explanations? For example, is dopamine best characterized as a “pleasure molecule” or is there a deeper and better characterization? (Berridge, 2007
). Are we getting at the core features of the phenomenon? Is the actual effect narrow, or are we making a general effect appear highly specific as a consequence of using a narrow set of dependent variables? Are we including conditions and measures that ensure sufficient discriminative validity?
Fair enough, there is no perfect algorithm for finding the optimal characterization, only scientific judgment. Nonetheless, sensitivity to the possibilities of mis-characterization may guard against premature fixation of a description. Naturally enough, the media prefer fetching characterizations that capture the public’s imagination, but in the long run, indulging these whims can embarrass the science. They also hurt science by covering interesting complexity, since after all, nonapeptides, like OXT, may not have a “function”, and may exert different and even opposing influences on behavior, based upon the sophisticated pattern of neuromodulation in the brain and a particular social arrangement (Goodson and Kabelik, 2009