A multiverse meta-analysis of intranasal oxytocin administration studies

Intranasal oxytocin administration has been used to experimentally manipulate central oxytocin levels to investigate its effects on physiology and cognition. However, reported outcomes have varied substantially. Over 20 meta-analyses to date have examined the effect of oxytocin administration on cognition and behavior to better understand these inconsistent findings. These meta-analyses, like primary studies, involve different methodological decisions such as inclusion criteria and analytical methods, which can influence their conclusions. To evaluate the impact of methodological decisions on meta-analysis outcomes, we adopted a multiverse approach, exploring a wide range of possible meta-analyses using diverse methodological approaches. This included varying inclusion criteria, data synthesis models, and methods for correcting publication bias. We analyzed 372 effect sizes from 107 studies on the effects of oxytocin administration on cognition, which yielded 59 distinct meta-analyses incorporating plausible analytic decisions. Our results indicate that summary effect estimates for intranasal oxytocin meta-analyses can vary considerably depending on the methodological decisions made by meta-analysts (d = -0.08 to 0.22). Notably, the inclusion of populations with psychiatric or neurodevelopmental conditions and multiple oxytocin administrations yielded larger effect sizes. Furthermore, summary effect estimates were inflated when publication bias was not accounted for. These findings underscore the importance of meta-analysis pre-registration, which facilitates the transparent evaluation of meta-analyst decision-making, and the importance of correcting summary effect size estimates for publication bias. The analysis script and data have been published on the Open Science Framework (https://osf.io/s5gx4/).