Is testosterone linked to human aggression? A meta-analytic examination of the relationship between baseline, dynamic, and manipulated testosterone on human aggression.

Abstract Testosterone is often considered a critical regulator of aggressive behaviour. There is castration/replacement evidence that testosterone indeed drives aggression in some species, but causal evidence in humans is generally lacking and/or—for the few studies that have pharmacologically manipulated testosterone concentrations—inconsistent. More often researchers have examined differences in baseline testosterone concentrations between groups known to differ in aggressiveness (e.g., violent vs non-violent criminals) or within a given sample using a correlational approach. Nevertheless, testosterone is not static but instead fluctuates in response to cues of challenge in the environment, and these challenge-induced fluctuations may more strongly regulate situation-specific aggressive behaviour. Here, we quantitatively summarize literature from all three approaches (baseline, change, and manipulation), providing the most comprehensive meta-analysis of these testosterone-aggression associations/effects in humans to date. Baseline testosterone shared a weak but significant association with aggression (r = 0.054, 95% CIs [0.028, 0.080]), an effect that was stronger and significant in men (r = 0.071, 95% CIs [0.041, 0.101]), but not women (r = 0.002, 95% CIs [−0.041, 0.044]). Changes in T were positively correlated with aggression (r = 0.108, 95% CIs [0.041, 0.174]), an effect that was also stronger and significant in men (r = 0.162, 95% CIs [0.076, 0.246]), but not women (r = 0.010, 95% CIs [−0.090, 0.109]). The causal effects of testosterone on human aggression were weaker yet, and not statistically significant (r = 0.046, 95% CIs [−0.015, 0.108]). We discuss the multiple moderators identified here (e.g., offender status of samples, sex) and elsewhere that may explain these generally weak effects. We also offer suggestions regarding methodology and sample sizes to best capture these associations in future work.