Anabolic/Androgenic Steroid Use and Aggression II: Does the Evidence Support a Causal Inference?

by Jack Darkes, PhD
Assistant Professor,
Department of Psychology
Director of Interventions,
Alcohol and Substance Use Research Institute
University of South Florida

Please send us your feedback on this article.

Introduction

Findings regarding the AAS use and aggression relationship are inconsistent and vary with the nature of the study and design. Although widely accepted as fact, a review finds little, if any, strong evidence for a direct causal relationship. This may, in part, stem from confusion regarding the specific questions that are answerable by various research designs. For instance, empirical studies generally report a strong association between AAS use and aggression, suggesting that AAS users are more likely than non-users to report high levels of aggression. Such reports suffer potential problems, such as memory biases that might distort reports of motivations and past behavior (e.g., Nisbett & Wilson, 1977), self-selection biases in AAS use, and no control for pre-existing biological and psychological factors. Experimental studies control for selection and pre-disposition via psychological testing, exclusion/inclusion criteria, and random assignment and find less evidence of an association and less support for a causal relationship. While the generality (external validity) of experimental designs must be considered, such issues also indicate that, as has been suggested the AAS/aggression relationship is complex (Beel, Maycock, & McLean, 1998; Uzych, 1992) and the question to be answered requires clarification.

Does AAS use cause aggression? This article will evaluate the support for such an assertion. Due to space limitations, study details are not presented, but can be found in Part I. For a further review of the issues involved refer to Sharp and Collins (1998) or Sapolsky (1997).

The evidence supports the complex multi-factorial nature of the AAS/aggression relationship. The ultimate questions to be answered reflect the discrepancy between the findings from different research methods. As a means of explaining this discrepancy, Part III will suggest a model in which aggression in AAS users is moderated by antecedent factors and partially mediated though proximal psychological variables.

This article will not argue that AAS users do not risk potential psychiatric (Pope & Katz, 1992) or medical (Korkia, 1998) consequences. The purpose is neither to endorse nor condemn AAS use, but to evaluate the complex relationship between AAS use and aggressive behavior in humans.

The Basis of Causal Inference

Evidence for causation accumulates from a variety of research designs. It does not result from a single study or experimental design, but accrues via convergent findings from a variety of investigative strategies. Several increasingly stringent levels of evidence can be required for in order to assert causation. In the exploratory stages, concurrent associations between use, levels of use and aggression can be examined. More rigorous studies follow users and assess behavior either before and after first use or as use fluctuates over time, to evaluate associations between patterns of use and behavior. In this way it can be determined whether aggression precedes, follows, or varies predictably with AAS use (prospective/longitudinal designs). In quasi-experimental designs pre-existing groups of users and non-users are "treated" and their behavior is compared. However, these designs are also subject to biases associated with self-selection and frequently offer little more than correlation. Finally, "true" experiments with random assignment to treatments (e.g., use or non-use) and evaluations of post-treatment behavior can be used. For the most part, the studies reviewed in Part I fit into these four categories.

The adage "Correlation does not equal correlation" defines a major problem with designs evaluating association solely via correlation. Simultaneous occurrence of behaviors (e.g., AAS use and aggression) is not sufficient evidence of causation. Based on correlations, it appears that AAS users report increased aggression compared to non-users. But correlation cannot determine causality or directionality: if the relationship were causal, the direction could not be determined solely through correlation of concurrent behaviors. Is aggression an effect or a cause of AAS use? Two competing explanations for AAS/aggression association could be supported by such results. Later this series will suggest bi-directionality as one explanation for the co-occurrence of aggression and AAS use. Those inclined toward aggression may be more likely to use, given AAS’ reputation for increasing aggression, and use in a manner that facilitates aggressive behavior. AAS and aggression are likely to both cause and effect each other.

Causation in the Endogenous Testosterone and Aggression Relationship

This directionality hypothesis is supported by reports showing that the T/dominance relationship is apparently bi-directional (e.g., Elias, 1981; Gladue, Boehler, & McCaul, 1989; Mazur & Booth, 1998). While empirical studies show correlations between T and aggression, experimental studies show bi-directionality in the relationship. An association between T and dominance does not exclusively indicate that high T levels cause dominant or aggressive behavior; increased T might also be a result rather than a cause. Although not universally supported (Suay et al., 1999), this finding suggests a complex process, as do reports of pre-contest increases in T.

The complexity of the T and aggression relationship bears highlighting. Variations in T and aggression might be expected to shown less psychological or environmental influence and to show strong evidence of a causal relationship, but this has not been the case (see Sapolsky, 1997). Dabbs’ (1996) quote bears repeating: "Relatively few people out of the entire population engage in criminal behavior, regardless of their testosterone levels (p. 180)." Hence, another shortcoming in inferring causation from correlation –third variables. Numerous characteristics differentiate incarcerated individuals from the general population. Hence, some unmeasured variable might influence both T and aggression. Such cases [prisoners (Dabbs, 1996) or individual "pathological" cases (instances of ‘roid rage, see Taber, 1999)] while convenient, have already shown aggression. They offer little information about the process in general because they have already exhibited the association and their anecdotal reports and recollections of the events and factors involved (again see Taber, 1999) are subject to several biases (see Nisbett & Wilson, 1977)

Do AAS cause aggression in humans?

Given the criteria needed to support a causal inference for AAS and aggression and the nature of the existing literature, the answer is complex. This section will evaluate the studies reviewed in Part I as evidence to support a causal inference.

Case Studies.

Shortcomings associated with case studies (e.g., Pope & Katz, 1990) and other idiographic evaluations (e.g., Thilbin, Kristiansson, & Rajs, 1997) have already been discussed. Although rich in detail, they represent an exploratory level of analysis. Generalizing from such data to the population is a tenuous proposition and, for the most part, not possible. Such nightmarish examples (e.g., Taber, 1999) capture attention, but provide little more than headlines regarding AAS use and aggression.

Research based on correlation dominates the literature on AAS and aggression. In all fairness, most authors do not strongly assert that their data prove that AAS cause aggression, but that, in their samples, AAS users reported more aggression than non-users. For example, current users reported higher levels of anger-arousal and hostile outlook (Lefavi, Reeve, & Newland, 1990). Mood disorder (Pope & Katz, 1994), self-reported aggression and aggressive traits (Galligani, Renck, & Hansen, 1996; Perry, Anderson, & Yates, 1990; Yates, Perry, & Murray, 1992) "abnormal" personality traits (Cooper, Noakes, Dunne, Lambert, & Rochford, 1996) and aggressive mood (Bond, Choi, & Pope, 1995) were increased in AAS users. Unfortunately, except for sporadic comments in discussion sections (which never make the headlines), the shortcomings of these data as evidence of causation are rarely mentioned. For instance, the assertion, with no concrete non-circular definition of AAS abuse, that "There is evidence from both case reports and empirical studies that abuse of these drugs causes significant psychiatric and medical effects (p. 832)(Porcerelli & Sandler, 1998)" seems premature. Not all users who could be classified as high dose users (abusers) exhibit psychopathology (e.g., Pope & Katz, 1992), and case studies and concurrent assessments of use and aggression cannot prove causation.

Why? First, consider motivations to use AAS. Increased size, strength, and enhanced physical performance are potential candidates. Body dysmorphia, a body image disturbance wherein even strong and/or large individuals see themselves as small and weak (Pope, Katz, & Hudson, 1993; Wroblewska, 1997) has been offered as one motivation for males to use AAS. What characteristics might co-exist with these motivations? Users of AAS reportedly showed lower levels of social physique anxiety than non-users (Schwerin et al., 1996), putatively as a result of use. Such users might exhibit higher levels of social physique anxiety prior to use. Upper body strength esteem and body dissatisfaction also reliably predicted AAS use (Schwerin et al., 1997). This may answer the question "Why get bigger or stronger?" AAS use often occurs in conjunction with other antisocial behaviors, such as truancy and other substance abuse (Kindlundh, Isacson, Berglund, & Nyberg, 1999) and the antisocial traits of AAS users have been characterized as similar to those in alcoholics (Yates, Perry, & Anderson, 1990). Psychological variables such as expectations of positive effects from use (Lovstakken, Peterson, & Homers, 1999) might also differentiate AAS users from non-users. Although these differences between users and non-users are one reason empirical studies cannot show causation, such variables are rarely controlled for in AAS research. And, while size and strength increases are the goal of many who engage in resistance training, only a minority of resistance trainers choose use AAS and a minority of those exhibit overt aggression. Physical enhancement may be only one potential motive for using AAS.

Secondly, the inability of correlation to establish causal direction is problematic. If concurrent measurement find that AAS users report more aggression than non-users, it is unclear which factor is "causing" the other. The factors mentioned above may either cause or result from AAS use. Characteristically aggressive individuals could be more likely to use AAS. Or AAS could cause aggression. Or they could interact. For instance, AAS use could enhance the expression of existing aggressive tendencies. Concurrent associations between variables cannot answer these questions.

Large-scale empirical and individual case studies are best used to explore possible associations between behaviors, not to show causation. These studies suggest that AAS users frequently report increased aggression and at levels higher than non-users. This reliable and consistent association serves only to suggest that further study of potential causation is warranted.

The study of repeated patterns of association between AAS use and aggression over time is more difficult to execute than concurrent assessment. However, assessing over time allows the evaluation of order of occurrence as logical evidence of causation. If aggressive behavior increases following initiation of drug use and decreases during non-use periods, support for causation is bolstered.

Choi, Parrott, & Cowan (1990) followed current (self-administering) AAS users and non-users over time during periods of use and non-use. If AAS cause aggression, users should report increased aggression during use compared to non-use periods and non-users. Although a significant user/non-user by use/non-use period interaction for aggression was reported, neither user status nor time period had a reliable effect on aggression. AAS users reported more hostility than non-users, regardless of drug phase. This finding, rather than supporting the AAS/aggression causal hypothesis, suggests that a third unmeasured variable is likely influencing both AAS use and aggressive behavior. This finding replicates previous correlational findings - users and non-users differ on measures of aggression (e.g., Galligani, Renck, & Hansen, 1996; Moss, Panzak, & Tarter, 1992), suggesting potential pre-existing dispositions and/or current psychological differences.

Su et al., (1993) assigned male non-athletes with no AAS use history to a within subject design and examined mood changes over four phases: placebo, low (40 mg/day) and high dose (240 mg/day) Methyltestosterone and placebo withdrawal. The reliable difference in self-reported hostility between placebo and high dose periods was one of approximately 80 comparisons. Hence, the significance level of p < .05 potentially should have been corrected for multiple comparisons. Exact p levels were not given so the potential effect of such a correction is unknown. Su et al., nonetheless acknowledge that "Symptomatic differences did not, however, reflect differences in plasma anabolic steroid levels (p. 2763)" presaging both Riem & Hursey (1995) and Dabbs (1996). If aggression evidences no association with T levels (Dabbs, 1996) or plasma levels of AAS (Riem & Hersey, 1995; Su et al., 1993) over time, then the mechanisms of change in perceived aggression require further examination. Although subjects reported more aggression, the lack of association between plasma AAS levels and self-report hardly justifies the assertion that AAS cause aggression. Changes were apparently related to some other unmeasured variable.

These findings also suggest enhanced placebo effects at higher doses. A large dose of AAS might cause some non-specific stimulation, reinforcing the belief that an active substance has been taken. It has been suggested that the use of inert placebos (e.g., oil injections in the AAS studies) might be inadequate. Drugs have both specific (in this case, androgen receptor binding) and nonspecific (e.g., CNS stimulation) effects. If only the specific effects are hypothesized to cause certain changes or adaptations, then a placebo must not only control for administration of treatment (e.g., injection) , but also for nonspecific (or undesirable) effects. A recent meta-analysis of studies of antidepressant medication (Kirsch & Saperstein, 1998) found that approximately 50% of drug effects result from this "active" placebo effect. Hence a more effective choice would also mimic the effects of the drug’s that are not related to the desired outcome. An effective placebo for high dose testosterone should mimic all potential stimulatory or undesirable effects while not being androgenic or anabolic. The potential for these effects to enhance the expression of aggression will be discussed in part III of this series.

Kouri, Lukas, Pope, & Oliva, (1995) administered increasing doses of testosterone cypionate (TC) or placebo in a within subject design to eight males, including 3 former AAS users. Increased aggressive responding in response to provocation was reported following TC administration as compared to placebo and baseline and higher self-report scores were reported for post TC as compared to baseline. The measure of aggression was somewhat contrived (see Part I) and several participants who failed to believe the manipulation were excluded. The makeup of the final sample was not reported. Given previous findings, including prior users might also complicate interpretation. Unfortunately this sample size would be too small to analyze for prior use effects. In addition, the measure of aggression (number of button pushes to subtract points in retaliation) is not likely to generalize to real world behavior, telling us little about AAS and aggression as it is typically exhibited or experienced.

Swanson (1989) examined differences in aggression between current AAS users (self-administering), non-using athletes, and non-using non-athletes on a sham reaction time task. There were no between group differences reported. Self-selection bias is a problem in this study, as is the nature of the measure of aggression. Based on previous findings, this study might have been biased toward finding group differences and the lack of differences fails to replicate the empirical findings. This may reflect issues related to both the manipulation and the measure.

True experiments exhibit increased internal validity and are the optimal test of the AAS and aggression hypothesis. Randomly assignment controls for selection biases or predisposing characteristics, hence testing the hypothesis independent of their influence. By definition, such variables are equally represented in all treatments as a function of random assignment. Such studies have limited external validity, or the ability to generalize to self-initiated and maintained AAS use in naturalistic settings as compared to empirical studies, due to the strengths noted above.

Bjorkqvist, Nygren, Bjorklund, and Bjorkqvist (1994) found no effect of AAS treatment for either self-reported and observer rated mood, which does not support a causal role for AAS. Reliable increases in various self-report measures and observer ratings occurred exclusively in the placebo group, suggesting that psychological factors might mediate the AAS and aggression relationship. The lack of a placebo effect in the testosterone group requires further investigation.

Both Tricker et al. (1996) and Bhasin et al., (1996) reported no effect of testosterone administration on mood or behavior, regardless of assessment method. Administration of AAS and placebo to randomly assigned participants found no reliable differences in self-reported or observer rated aggression between treatment groups.

Yates, Perry, MacIndoe, Holman, & Ellingrod (1999) randomly assigned males to receive various doses of TC. A small number of participants dropped out, but there was no difference in attrition between groups and no dropout was predicated on increased aggression. No reliable differences on self-report measures or observer ratings of aggression were found. These findings suggest that, when stringent inclusion/exclusion criteria are used, AAS are minimally associated with any psychological disturbance.

The above experimental studies suggest that when, by design, factors that might pre-dispose individuals to aggression are represented equally across treatment groups, the association between AAS use and aggression is greatly reduced or non-existent.

Summary

This examination of the research designs and methods represented in the AAS literature suggests that support for the hypothesis that AAS use causes aggression is limited and the association varies with the design used. Studies with higher levels of external validity but less control over extraneous factors, suggest an association between AAS use and increased aggression in real world use and naturalistic settings. Those who use AAS, for any number of reasons, appear more likely to exhibit aggression. In fact, prospective studies (e.g., Su et al., 1993) suggest that users may be more likely to exhibit aggressive behavior during periods of non-use, as well. It seems therefore that another factor, either in addition to or independent of AAS use exerts influence over behavior concurrent with AAS use. An understanding of such issues is even more important when the potential for conclusions regarding this relationship to influence the availability of testosterone and nor-testosterone precursors is considered (see Yesalis, 1999).

Experimental studies that control for such variables are less supportive of a causal relationship. Individuals, randomly assigned to receive AAS (as opposed to self administering) and therefore having an equal chance of receiving or not receiving an active treatment, do not exhibit increased aggressive behavior, regardless of the method of assessment.

In order to understand the apparent discrepancies in this evidence, it is useful to reframe the question. In fact, a restatement of hypotheses is frequently needed in such cases. There are actually two different questions being asked and answered in the literature. The first does not address causation, but more precisely asks "Are those who use AAS more likely to show increased aggression?" The empirical data suggest that the answer to this question is, for the most part, yes. In general, observational data find that, in natural application, those who use AAS report or are reported to exhibit increased aggression. While this answer justifies the further study of the relationship, some details remain unanswered, such as in whom, when, in what circumstances, and through what process. Aggression is not a universal consequence of AAS use nor is it always associated with physiological measures of AAS use.

The follow-up question then is "Does the use of AAS cause this aggression?" The observation of AAS users in their natural environment cannot answer this question and experimental studies suggest that the mere use of AAS cannot conclusively be inferred to cause the observed behavior.

Part III of this series will discuss the potential pre-existing conditions and psychological conditions that might explain the discrepancy between the empirical evidence for an AAS/aggression relationship and the experimental evidence opposing it. It will offer one potential answer to the questions of for whom and by what process.

Please send us your feedback on this article.

References