Elsevier

Aggression and Violent Behavior

Volume 10, Issue 2, January–February 2005, Pages 241-261
Aggression and Violent Behavior

Testosterone and aggression: A reanalysis of Book, Starzyk, and Quinsey's (2001) study

https://doi.org/10.1016/j.avb.2004.01.001Get rights and content

Abstract

A critique is presented of the meta-analysis of testosterone and aggression by Book, Starzyk, and Quinsey [Aggression and Violent Behaviour 6 (2001) 579], and the results of a reanalysis of their data are reported. We identified the following problems with their analysis: Secondary, rather than primary, sources were used in the initial literature review; 15 studies were included that should have been omitted; there were no decision rules for calculating effect sizes, leading to inaccuracies in most of these; the statistical test used to compare categories was of low power; the coding of study characteristics was inaccurate. A reanalysis that corrected these problems produced a lower mean weighted correlation (r=.08 instead of the reported r=.14). The conclusions from our categorical comparisons were different from those of Book et al.: Neither of their positive findings (a decline with age; lower correlations in morning than afternoon samples) were confirmed. We found significant differences for sex, age, offender status, and source of hormone measure, all of which are different from those in the original analysis.

Introduction

Book, Starzyk, and Quinsey (2001) undertook a meta-analysis of studies linking testosterone and aggression. While such quantitative syntheses are important for advancing knowledge in a particular area and are generally preferable to qualitative reviews Cooper, 1984, Schmidt, 1992, there are certain problems with undertaking a meta-analytic review that these authors appear not to have addressed. As a consequence, some of their conclusions are likely to be unsound. In this commentary, we outline these problems under the following headings: (1) the derivation of hypotheses by qualitative reviews of secondary rather than primary sources in the background literature; (2) the inclusion of studies that should have been omitted, because they were duplicate entries from the same sample, were not concerned with the association between testosterone and aggression, or did not involve a measure of aggression; (3) ambiguity in the decision rules for calculating effect sizes for the association between testosterone and measures of aggression, leading to inaccuracies in the majority of their effect-size values; (4) statistical problems in comparing categories; (5) inaccuracies in coding the study characteristics, leading to inaccuracies in the categorical analysis.

In addition to outlining these problems, we have examined all studies used by Book et al. (2001) in their analysis, reassessed whether they should be included, recalculated the effect sizes, recoded the categorical variables, and analyzed these values. We then compared the conclusions from this reanalysis with those reached in the original analysis. We conclude that there are sufficient grounds for questioning most of the conclusions in the original analysis. Finally, we make some general observations about the utility of meta-analytic reviews.

Section snippets

Review of the literature

It is important that research questions are accurately identified in the review of previous literature. This requires accurate characterization of what is already established and separating this from widely repeated but erroneous statements found in secondary sources and texts. Book et al. (2001) relied on a few book chapters for most of their background review, leading to the presentation of secondary sources as empirical studies and to some inaccurate characterization of what is established

Which studies should be included in the meta-analysis?

Inspection of Table 1 by Book et al. (2001) indicates several studies that should not have been included in a meta-analysis of the association between testosterone and aggression. There are three reasons why some studies should have been excluded: (1) duplication of the same sample, (2) studies involving the effect of competition on testosterone levels, and (3) studies involving no aggression measures.

We identified two cases of sample duplication. In one, the same sample was entered three

Calculation of effect sizes

Effect sizes listed by Book et al. (2001) were r values for the association between testosterone and measures of aggression. This may appear straightforward, but in many studies, there are multiple measures of testosterone (and other androgen levels), and multiple measures of aggression. Thus, Table 1 of the earlier review by one of the present authors (Archer, 1991) showed several different correlations for each study. Unless clear selection rules are followed, the same data may lead to

Statistical problems in comparing categories

Meta-analysis is as important for identifying comparisons between subgroups as it is for enabling an overall effect size to be calculated. The method adopted by Book et al. (2001) is not entirely clear. It appeared to be a comparison of the numbers of expected and observed correlations in different categories (p. 591), using χ2. It is more usual to compare across different categories using a categorical model test equivalent to a one-way analysis of variance Hedges & Olkin, 1985, Johnson, 1989.

Recoding and reanalysis of study characteristics

Although we have presented a reanalysis of Book et al.'s (2001) comparisons between categories, based on their figures, our discussion in 2 Review of the literature, 3 Which studies should be included in the meta-analysis? showed that both the selection of samples and the effect sizes they used were unsound. We therefore decided to undertake a reanalysis of categorical variables with our revised figures. Although Book et al. provided a table for the reliability of the coded variables, there are

Conclusions

The specific conclusions from our meta-analysis of the studies used by Book et al. (2001) are that their analysis is unsound in several respects: the choice of studies to include, the calculation of study-level effect sizes, and coding of the variables. Each of these has repercussions for the conclusions we can draw from the analysis. Instead of a mean weighted correlation of r=.14, reanalysis produced a considerable smaller value (r=.08), although this was still significantly different from

References (62)

  • A.S. Scerbo et al.

    Salivary testosterone and cortisol in disruptive children: Relationship to aggressive, hyperactive, and internalizing behaviors

    Journal of the American Academy of Child and Adolescent Psychiatry

    (1994)
  • E.A. Shirtcliff et al.

    Gender differences in the validity of testosterone measured in saliva by immunoassay

    Hormones and Behavior

    (2002)
  • J. Archer

    The influence of testosterone on human aggression

    British Journal of Psychology

    (1991)
  • J. Archer et al.

    The association between testosterone and aggression among young men: Empirical findings and a meta-analysis

    Aggressive Behavior

    (1998)
  • J. Bain et al.

    Sex hormones in murderers and assaulters

    Behavioral Science and the Law

    (1987)
  • T. Banks et al.

    Salivary testosterone and cortisol in a delinquent and violent urban subculture

    Journal of Social Psychology

    (1996)
  • M. Berman et al.

    The effects of hormones, Type A behavior pattern, and provocation on aggression in men

    Motivation and Emotion

    (1993)
  • A. Booth et al.

    Testosterone and men's marriages

    Social Forces

    (1993)
  • J.H. Brooks et al.

    Serum testosterone in violent and nonviolent young offenders

    Journal of Clinical Psychology

    (1996)
  • K. Christiansen et al.

    Hormonal, anthropometrical, and behavioral correlates of physical aggression in !Kung San men of Namibia

    Aggressive Behavior

    (1992)
  • H.M. Cooper

    The integrative research review

    (1984)
  • J.M. Dabbs et al.

    Saliva testosterone and criminal violence in young adult prison inmates

    Psychosomatic Medicine

    (1987)
  • J.M. Dabbs et al.

    Age, testosterone, and behavior among female prison inmates

    Psychosomatic Medicine

    (1997)
  • J.M. Dabbs et al.

    Salivary testosterone and cortisol among late adolescent male offenders

    Journal of Abnormal Child Psychology

    (1991)
  • M. Daly et al.

    Homicide

    (1988)
  • M. Daly et al.

    Killing the competition: Female/female and male/male homicide

    Human Nature

    (1990)
  • M. Daly et al.

    Evolutionary psychology of male violence

  • J. Ehrenkranz et al.

    Plasma testosterone: Correlation with aggressive behavior and social dominance in man

    Psychosomatic Medicine

    (1974)
  • M. Elias

    Serum cortisol, testosterone, and testosterone-binding globulin responses to competitive fighting in human males

    Aggressive Behavior

    (1981)
  • B.A. Gladue

    Aggressive behavioral characteristics, hormones, and sexual orientation in men and women

    Aggressive Behavior

    (1991)
  • B.A. Gladue et al.

    Hormonal responses to competition in human males

    Aggressive Behavior

    (1989)
  • Cited by (158)

    • Biochemical Factors in Aggression and Violence

      2022, Encyclopedia of Violence, Peace, & Conflict: Volume 1-4, Third Edition
    View all citing articles on Scopus
    View full text