Performance-Enhancing Drug Use Among Professional Athletes: A Longitudinal Test of Social Learning Theory

Abstract

This article tests Akers’s social learning theory with self-reported panel data on performance-enhancing drug (PED) use among a sample of professional athletes (n = 510). Using latent growth curve modeling we find that intraindividual differences exist in the developmental growth trajectories of PED use and the social learning process. Specifically, both PED use and the social learning process increase over time, while those who begin with high and low levels of PED consumption and social learning replicate these patterns throughout the sports life cycle. In addition, using structural equation modeling, we find modest to moderate contemporaneous and lagged, direct, indirect, and reciprocal effects between the social learning process and PED use in a manner consistent with the theory.

Keywords

social learning theory criminological theory performance-enhancing drugs

The use of performance-enhancing drugs (PEDs) in sports is not a recent phenomenon, but given the current social climate and increasing calls for public scrutiny and accountability, the topic of PED use has come to the forefront of the world’s attention. There are several recent and celebrated cases, such as Lance Armstrong or the Russian athletes at the 2014 Winter Olympics, who are alleged—or proven—to have been involved in doping. These cases, and others, can lead to a myriad of consequences for the athlete, team, and global community. There is an implicit assumption that when athletes compete, they compete in ethical ways. However, not all athletes abide by the rules of their sport. Use of PEDs, thus, constitutes a form of deviance, and it is important to explore the correlates of doping to better understand its etiology and so that governing bodies of sports may be in a better position to address the problem.

In this context, social learning is one such correlate worthy of empirical scrutiny. As prior research has demonstrated that sports doping is influenced by social processes, particularly interactions with coaches and fellow athletes (e.g., Lucidi et al., 2008; Papadopoulos et al., 2006; Peters et al., 2005), the current study assesses the hypothesis that doping behaviors and attitudes of those with whom athletes interact have an effect on the athlete’s own doping behavior. In fact, one recent study used the central components of social learning theory—differential association, definitions, differential reinforcement, and imitation—to explain PED use, and it found overall support for the theory’s principal assertion (Kabiri et al., 2018b). However, this study was unable to properly consider the causal relationship between social learning and PED use as it did not utilize longitudinal data, and we know of no study to date that has investigated the relationship between social learning and doping over time.

To that end, and in building upon previous research, this study contributes to the literature by testing the causal relationship between the social learning process and PED use among athletes. More specifically, structural equation modeling (SEM) is used to examine whether the social learning process is predictive of PED use and the extent, if any, to which PED use, in turn, has a feedback/reciprocal effect on the social learning process in a three-wave longitudinal study (n = 510) of Iranian athletes.

Literature Review

Social Learning Theory

Deviant behavior and peer relationships are empirically linked. In fact, this association is one of the most prominent and robust in criminology (e.g., Pratt et al., 2010), dating back to a theoretical tradition that began with Sutherland’s (1947) differential association theory. In it, Sutherland postulates that deviance is the result of differential association and exposure to an excess of definitions favorable to crime. Differential association refers to the people that one comes into contact with on a regular basis. These individuals are considered agents of socialization. In principal, they are ones from whom we learn. When people are young, family members are viewed as the primary agents of socialization, and they are generally supplemented/replaced (in terms of influence) by peers as one gets older. Thus, it is not surprising that delinquents, for instance, come from families with delinquent members and are also likely to have delinquent friends. More importantly, these agents can have a significant impact on one’s definitions favorable to deviant behavior (e.g., Matsueda & Anderson, 1998). Definitions are the attitudes and rationalizations that label certain behaviors as right or wrong, prosocial or antisocial. Accordingly, people will be delinquent when they hold an excess of definitions favorable to crime relative to definitions unfavorable to crime.

Building from earlier learning theories, Akers’s (1973, 2009) social learning theory suggests that antisocial behaviors are learned through the same mechanisms as prosocial behaviors, and his theory relies heavily on the psychological principle of operant conditioning. Specifically, social learning theory focuses on four main learning concepts: differential association, definitions, differential reinforcement, and imitation.

Beyond differential association and definitions, differential reinforcement refers to the balance of anticipated or actual rewards and/or punishments following an initial behavior. Through the principles of operant conditioning, behavior can be increased or decreased depending on the rewards or punishments expected. A behavior is likely to continue, for example, if one anticipates receiving a positive stimulus (reward) or if negative stimuli (punishments) are removed/avoided. Conversely, a behavior is not likely to continue if one anticipates receiving a negative stimulus (punishment) or if positive stimuli are removed/lost. These positive and negative stimuli may be both social and nonsocial in nature and it is the net balance of rewards to punishments that determine whether or not a behavior is continued. Finally, imitation refers to behavior that is modeled through the observation of others. Akers suggests that imitation may be the most important of the factors in the initial development (i.e., initiation) of the behavior.

Overall, Akers’s social learning theory asserts that deviant behavior results from the frequency, duration, and intensity of interactions with—and exposure to—significant others (differential associations) who serve as the primary sources of one’s own evaluative judgments regarding the appropriateness of said behavior (definitions). Moreover, those with whom one associates may also serve as role models whose own deviant behavior is modeled (imitation) when it is observed. These associates also serve as the primary source of the actual and anticipated, positive and negative, rewards and punishments (differential reinforcement) for one’s own deviant behavior.

Although many invoke differential association and social learning theory (socialization) versus social control/social bonding theory (selection) in the debate regarding the causal ordering of the delinquent peers-delinquent behavior relationship (e.g., Haynie & Osgood, 2005; Payne & Cornwell, 2007), Akers’s social learning theory explicitly argues that not only does differential association lead to deviant behavior (socialization), but deviant behavior also has feedback/reciprocal effects on the social learning process (Akers, 2009). Involvement in deviant behavior is expected, according to Akers, to influence the frequency, duration, intensity, and priority of the persons with whom one interacts, likely enhancing one’s association with deviant others while reducing their interaction with more conforming others. In turn, involvement in deviant behavior is also expected to influence one’s definitions and to refine one’s expected balance of rewards and punishments (differential reinforcement) as a consequence of involvement; the direction of these feedback effects is a function of the extent to which the consequences of the deviant behavior were differential reinforced. Unfortunately, very few studies have specifically tested for these feedback/reciprocal effects.

Akers’s social learning theory has been abundantly tested in the criminological literature and it has received considerable empirical support (e.g., Akers & Jensen, 2011). Moreover, the support for the theory has generally held across different samples, cultures, methodologies, and outcomes. For instance, the elements of the theory have been found to be predictive of a wide variety of criminal behavior, including, but not limited to, drug use (Akers et al., 1979), intimate partner violence (Sellers et al., 2003), and cyberbullying perpetration (Shadmanfaat et al., 2019). In fact, a recent meta-analysis from Pratt et al. (2010) demonstrated significant effect sizes for each of the four social learning theory elements.

A primary concern with social learning theory is causal ordering. The theory posits that deviant behavior is the result of learning. Although one’s peers and one’s own deviance have been shown to be very highly correlated, correlation does not equal or imply causation (see, for example, Matsueda & Anderson, 1998). The best way to test the causal processes within social learning theory is to use longitudinal data, and, over the last three decades, several studies have used longitudinal data to do so (e.g., Akers & Lee, 1996; Cochran et al., 2017; Krohn et al., 1985; Powers et al., 2017).

One of the earliest known longitudinal studies of social learning theory is from Krohn et al. (1985). Utilizing data from a 3-year panel study of junior and senior high school students, their path analyses found that the theory did a better job of predicting maintenance of cigarette smoking rather than initiation of cigarette smoking. More consistent with the theory, however, they found that reinforcement mediated the relationship between differential association and smoking. Importantly, they also observed an effect of smoking on their measures of differential reinforcement consistent with the reciprocal effects explicit in Akers’s theory. Research from Akers and Lee (1996), using a 5-year panel data set of 454 seventh to 12th graders in Iowa, demonstrated differential association, definitions, reinforcement, and a combined social learning variable were all predictive of adolescent cigarette smoking and that the effects were stable over time. Furthermore, their path analyses found that the reciprocal effects of smoking on social learning were clearer for differential association than for definitions or reinforcement. Finally, Cochran and colleagues (2017; Powers et al., 2017) used a measure of “past partners’” interpersonal violence to serve as a proxy measure for the reciprocal effects of behavior onto the social learning process. Their results were consistent with the expectations of social learning theory. Although a few of the above studies have tested the relationship between social learning and substance use using longitudinal data, they have primarily focused on adolescent tobacco, alcohol, or marijuana use (Akers & Lee, 1996; Krohn et al., 1985). Moreover, no study to date has examined this relationship with PED use among professional athletes.

PED Use

According to Lazuras et al. (2010), the use of PEDs is defined as the illegal use of drugs among amateur and/or professional athletes, and it is commonly referred to as “doping.” Chan et al. (2015) notes that PED use among athletes is not only a violation against the regulations of the World Anti-Doping Agency, but such use can also lead to a variety of negative consequences such as tarnished reputations, being excluded from competitions, loss of revenue, and moral outrage from fans. Estimates of PED use among professional athletes suggest that its prevalence is approximately 10% to 40%. Dunn and colleagues (2009), for example, noted that 10% of gymnastic athletes used PEDs. Among track and field athletes, Sottas et al. (2011) reported that nearly 15% of such athletes had used PEDs between 2000 and 2010. Similar findings have also been reported for athletes in soccer (Vouillamoz et al., 2009) and cycling (Zorzoli & Rossi, 2010). As it relates specifically to athletes in Iran, from where the current sample is drawn, prior studies have indicated that the prevalence of doping behavior is between 27% and 67% (Angoorani et al., 2012; Manouchehri & Tojari, 2013).

Sports doping is certainly not a new problem in athletics, but the extant literature on the etiology of PED use is quite limited. Recent scholarship, however, has tested several individual, social, environmental, and theoretical factors. Some of these factors include a personal drive for muscularity or thinness (Akindutire & Olanipekun, 2015), infidelity to sport morality (e.g., Melzer et al., 2010), and low self-control (Kabiri et al., 2018a).

With respect, more specifically, to social and environmental factors, Donovan et al. (2002) noted that peer groups—and their socialization effects—play a key role in the formation of PED use. Furthermore, research from Ozdemir et al. (2005) demonstrated that two fifths (41%) of athletes in their sample who had used PEDs reported that social-environmental pressures and the role of friends had significant effects on their own doping tendencies. Other studies have also documented the important role of social and environmental factors (e.g., peer groups, peer attitudes) on PED use (e.g., Lucidi et al., 2008; Peters et al., 2005).

Many athletes compete in team sports, and even those who compete in individual events will typically train with coaches and other athletes in their sport. Thus, the sporting world provides a rich context for socialization to take place. Moreover, much of this socialization is likely to involve a confluence of association, definitions, imitation, and reinforcement. According to research from Peters et al. (2005), the athletes in their sample reported that PEDs were first introduced to them by their coach. Similarly, in a study from Kirby and colleagues (2011), athletes’ PED use is, at least in part, at the request of a fellow team member or trainer. Moreover, findings from Papadopoulos et al. (2006) demonstrated that athletes who knew an existing PED user were 7 times more likely to dope themselves. Accordingly, social learning theory would be a valuable theoretical framework to consider in an attempt to explain PED use.

Given that Akers’s social learning theory purports to explain a wide variety of deviant behaviors, and has been empirically supported as such, it should also be able to predict PED use among athletes. In fact, one recent study on this topic has garnered some initial evidence to support this assertion. Using SEM to analyze their data, Kabiri et al.’s (2018b) study of 589 Iranian athletes found that there were several positive and significant relationships between the four components of social learning theory and past, current, and future PED use. Only two of these 12 failed to uncover a statistically significant effect: (a) definitions favorable to crime and current PED use and (b) differential association and future PED use. One limitation of this study, however, is that it was not able to properly consider the causal relationship between social learning and PED use as it did not utilize longitudinal data. Therefore, the aim of the current study was to go back to these subjects and follow their PED use over two additional waves of data thus permitting us to examine the contemporaneous and lagged direct, indirect, and reciprocal relationships between the social learning process and athletes’ PED behavior (past PED use, current PED use and willingness to use PEDs in the near future).

Method

Data and Sample

We employed three waves of self-reported panel data from a sample of 510 professional athletes from Rasht, Iran, who have competed for at least 5 years in their sport and were registered in 2016 in the Department of Physical Education for the city of Rasht. Registration in the physical education department is important for it is through such a process that professional athletes are distinguished from amateurs. A random sample was drawn from the registration list, and the sampled athletes were invited to a large meeting area at the sporting complex to complete the questionnaires. As per approved institutional review board (IRB) requirements, the purpose of the study was discussed and voluntary consent was provided by the sampled athletes. Then a self-administered questionnaire was distributed. These procedures were repeated in the fall of 2017 and 2018. Of the 850 athletes asked to participate, 793 participated in the first wave of research. From the 793 athletes who participated in the first wave of research, 620 participated in the second wave of research, and 510 completed both the second and third waves. Our analysis includes the 510 subjects who completed all three waves of research.

Descriptive statistics show that 57.1% of respondents were male and 42.9% were female. Although 41.6% of respondents were younger than 25 years, 39.4% were between 25 and 30 years old, 15.3% were between 30 and 35 years old, and 3.7 % were older than 35 years. Moreover, 70.4% were single and the other 29.6% were married. In addition, 50.4% had, at most, a secondary education, while 27.6% had completed a college or university degree, and 22% had completed a graduate degree. In terms of athletic participation, 6.1% of the respondents participated in weight lifting, 7.3% participated in Taekwondo, 13.5% participated in Karate, 7.1% played handball, 19.4% participated in swimming, 18% played football/soccer, 8.8% played futsal (five-on-five indoor soccer), 5.5% played volleyball, 3.9% played basketball, 5.1% participated in bodybuilding, and 5.3% participated in wrestling.

Dependent Variables

Athletes’ doping behavior (past PED use, current PED use, and future PED willingness) is examined through three measures designed to gain information on athletes’ PED behavior. Participants were asked to report whether they: “currently use a banned substance” and had “previously used a banned substance to enhance their performance.” Response options ranged from 0 (never) to 3 (systematically use banned substances). The subjects were also asked “How strong is your intention to use a banned substance at least once within the next 12 months.” Response options ranged from 0 (not at all strong) to 4 (very strong).

Social Learning Scales

Measures of all four of Akers’s key theoretical constructs are included in the present study. Differential association refers to the extent of one’s interactions and communications with others; these associations vary in frequency, duration, priority, and intensity, and is a two-dimensional construct: behavioral and normative. The normative dimension refers to the quality of beliefs and the attitude of significant others with whom the respondent interacts (i.e., fellow athletes, close friends, reference group members) with regarding doping behavior. The behavioral dimension refers to the behavior of significant others with whom one interacts regarding their own doping behavior.

The behavioral dimension is scaled to represent the elements of frequency, duration, priority, and intensity of relationships based upon responses to the following six items: “How many friends who you’ve known for a long time are using PEDs,” “How many athletes who you’ve known for a long time are using PEDs,” “How many of the friends you associate with the most are using PEDs,” “How many of the athletes you associate with the most are using PEDs,” “How many of your best professional sports friends are using PEDs,” and “How many of the athletes who are your reference models are using PEDs.” The response categories ranged from 1 (none of them) to 5 (all of them).

The normative dimension is represented by a three-item scale and assesses respondents’ perceptions of significant others’ beliefs and attitudes toward PEDs. Participants in the study were presented with the following series of Likert-type statements: “People who are important to you agree with the use of PEDs,” “People who influence you agree with the use of PEDs,” and “People whose opinions you value agree with the use of PEDs.” The response categories ranged from 1 (strongly disagree) to 5 (strongly agree). For each wave of data, these two subscales were used to create a second-order factor score measure of differential association (i.e., an additive scale comprised of these two subscales in which each subscale has been weighted by its factor loading prior to summing).

The differential reinforcement scale was developed to assess both the positive and negative stimuli (consequences), both social and nonsocial in nature, associated with PED use, which can positively or negatively reinforce or punish said behavior. Here, differential reinforcement is measured by means of three subscales, which assess personal gains, social gains, and perceived deterrence. Items for each of these subscales used a Likert-type scale which ranged from 1 (strongly disagree) to 5 (strongly agree).

Personal gains were measured using a three-item scale. Participants in the study were presented with the following statements: “Doping improves my athletic performance,” “PEDs help to overcome boredom during training,” and “PEDs give me the motivation to train and compete at the highest level.” Social gains were also measured using a three-item scale. Participants in the study were asked the following questions: “PED use makes me more accepted by my team members,” “PED use promotes my self-image,” and “PED use promotes my social image in the eyes of coaches, fans, and teammates.” Perceived deterrence was measured by two items: “Do you think there is a great risk of getting caught if you use banned PEDs?” (response options ranged from 1 = very great risk to 5 = no risk at all) and “Do you think you would be in great trouble if you got caught using banned PED?” (responses ranged from 1 = very much trouble to 5 = no trouble at all). For each wave of data, these three subscales were used to create a second-order factor score measure of differential reinforcement.

Definitions refer to ideas, beliefs, attitudes, and values regarding a certain deviant behavior. Definitions can be negative (opposed to the deviant behavior), positive (approving of the deviant behavior), or neutralizing, which justify/rationalize an otherwise negatively defined deviant behavior. This study focused on positive and neutralizing definitions of PED use. Nine items were used to measure the extent to which athletes held a neutralizing definition of doping, and responses ranged from 1 (strongly disagree) to 5 (strongly agree). The items included: “The risks related to doping are exaggerated”; “Athletes should not feel guilty about breaking the rules and taking performance-enhancing drugs”; “Athletes in my sport are pressured to take performance-enhancing drugs”; “Doping is not cheating since everyone does it”; “Health problems related to rigorous training and injuries are just as bad as those from doping”; “In comparison to the damaging effects of alcohol and tobacco, the use of performance-enhancing drugs is not so bad”; “It is not right to criticize athletes who use performance-enhancing drugs to improve their bodies, since many athletes do the same”; “Performance-enhancing drug use is an unavoidable part of the competitive sport”; and “Athletes who use performance-enhancing drugs are not to be blamed, but those people who expect too much from athletes should be blamed instead.” Three items were used to measure the extent to which athletes had a positive definition of doping, and responses ranged from 1 (strongly disagree) to 5 (strongly agree). The items included: “The use of performance-enhancing drugs is an effective way to maximize one’s potential,” “It is OK to use performance-enhancing drugs if this can help one to overcome one’s limits,” and “Using performance-enhancing drugs is no different to using a fiberglass pole, the latest cycling technology or technologically designed swimsuits.” For each wave of data, these two subscales were used to create a second-order factor score measure of definitions.

Imitation refers to the act of modeling one’s own behavior after that of a respected or important figure. Here, imitation was assessed among both primary and secondary groups. Primary group imitation was measured with two items: “There is an athlete who uses performance-enhancing drugs in my family, and I try to imitate his behavior” and “One of my close friends is an athlete who dopes, and I try to imitate his behavior.” The responses ranged from 1 (strongly disagree) to 5 (strongly agree). Secondary group imitation was also measured with two items: “One of my fellow athletes uses performance-enhancing drugs and I try to imitate his behaviors” and “I have an athlete as a role model who uses performance-enhancing drugs and I try to imitate his behaviors.” The responses ranged from 1 (strongly disagree) to 5 (strongly agree). For each wave of data, these two subscales were then used to create a second-order factor score measure of imitation.

Finally, for each wave of data, and following the example of Akers and Lee (1996), the four second-order scales comprising each of the four components of social learning theory (i.e., differential association, differential reinforcement, definitions, and imitation) were used to create a latent variable representing the social learning process as a single construct.

Validity and Reliability of Measurement Instruments

As reported in Table 1, and in accordance to the criteria set forth by Nunally (1978), all scales were found to have high internal consistency (α > .70; composite reliability [CR] > .70). In addition, we measured the scales’ discriminant validity by examining the average amount of variance shared between a construct and its measures (average variance extracted [AVE]). AVE indices were found to be higher than 0.50, as recommend by Fornell and Larcker (1981). In addition, first-order confirmatory factor analysis (CFA) was conducted, and the factor loadings were all significant (factor loadings greater than 0.50). Moreover, the CFAs revealed good fit indices (Kline, 2015).

Table 1.

Validity and Reliability of Measurement Instruments.

	Time 1				Time 2				Time 3
	Factor Loadings	AVE	CR	α	Factor Loadings	AVE	CR	α	Factor Loadings	AVE	CR	α
Variable Name	Min–Max	AVE	CR	α	Min–Max	AVE	CR	α	Min–Max	AVE	CR	α
Social learning constructs				.901				.989				.932
Differential associations				.910				.920				.951
Differential behavior	.652–.854	.577	.890	.889	.694–.833	.586	.894	.890	.735–.892	.690	.930	.927
Differential norm	.864–.871	.749	.899	.899	.795–.913	.757	.903	.902	.904–.943	.845	.942	.943
Differential reinforcements				.855				.800				.893
Personal gain	.740–.812	.616	.828	.827	.489–.792	.506	.710	.706	.678–.862	.575	.801	.794
Social gain	.761–.912	.706	.826	.817	.788–.858	.678	.808	.807	.848–.906	.770	.870	.869
Perceived deterrence	.655–.825	.555	.711	.706	.713–.835	.603	.751	.746	.794–.809	.643	.782	.778
Imitation				.825				.889				.936
Imitation from primary group	.794–.893	.714	.832	.829	.874–.839	.733	.846	.844	.884–.912	.806	.893	.892
Imitation from secondary group	.837–.857	.717	.835	.835	.812–.969	.800	.888	.881	.891–.931	.830	.907	.906
Definitions				.913				.935				.948
Positive definitions	.655–.825	.555	.832	.827	.969–.906	.661	.886	.881	.822–.901	.759	.926	.925
Neutralizing definitions	.593–.824	.515	.895	.888	.671–.834	.617	.928	.927	.674–.878	.642	.934	.933

Note. AVE = average variance extracted; CR = composite reliability; α = Cronbach’s alpha.

Method of Analysis

First, we conducted bivariate correlations to assess the relationship between social learning and PED behavior. Next, we used latent growth curve modeling (LGM). With LGM, two latent factors are specified by factor loadings of repeated measures. The intercept factor represents the level of the construct at Time 0 (i.e., baseline), while the slope factor represents the direction and rate at which the variable changes. In this way, LGM uses longitudinal data to (a) estimate the mean trend or slope of a variable over time, (b) test whether the level or intercept of a variable is related to the rate of change, and (c) examine whether the level and/or rate of change are associated with relevant risk factors or key outcomes (Preacher et al., 2008). Specifically, we examined whether (a) PED behavior increases during the sports life circle, and whether (b) the social learning process changes over time.

Next, SEM in AMOS was used to test the direct and indirect effects of the social learning theory construct on self-reported doping behavior (past use, current use, and willingness to use in the future), as well as the feedback/reciprocal effect of PED use on the social learning process across the three waves of data. The model tested in this study incorporates social learning and PED use variables in 2016 (Time 1), 2017 (Time 2), and 2018 (Time 3).

Results

Table 2 reports the correlations between doping and the social learning constructs across all three waves of data.¹ These correlations permit an initial assessment of the extent to which social learning at time t is associated with both contemporaneous (time t) and future doping (time t + 1 and t + 2) and the extent, if any, to which prior doping behavior (time t − 1 and t − 2) influences the social learning process. Finally, these correlations also permit us to assess the consistency over time in self-reported doping behavior. Especially intriguing is the degree to which willingness to use at time t is positively associated with actual use at time t + 1. The findings indicate that the contemporaneous (time t with time t) bivariate association between social learning and “current” PED use is consistently positive, modest, and statistically significant (r = .224, .284, and .477, respectively, at Waves 1, 2, and 3). The correlations also reveal a significant, lagged effect (i.e., time t on time t + 1 and t + 2) of social learning on “current” PED use (r = .210, .242, and .331). Social learning is also significantly associated with both contemporaneous and lagged measures of respondents’ willingness to use PEDs in the near future (r = .149–.352). With regard to the feedback/reciprocal effect of doping behavior on the social learning process, the correlations are also supportive of Akers’s social learning theory. That is, respondents’ self-reported “past” use of PEDs is associated with both their contemporaneous (r = .224) and lagged (r = .320–.397) social learning. Similarly, “current” PED use is also associated with lagged social learning (r = .268–.414).

Table 2.

Zero-Order Correlations Between Social Learning and PED Use Indicators Over Time (N = 510).

Variable Name	M	SD	1	2	3	4	5	6	7	8	9
Past PED use	0.84	1.06	1.000
Current PED use T1	0.53	0.89	.307**	1.000
Current PED use T2	0.83	1.01	.344**	.379**	1.000
Current PED use T3	1.17	1.20	.449**	.430**	.551**	1.000
PED willingness T1	0.65	0.95	.063	.007	.187**	.105**	1.000
PED willingness T2	1.01	1.05	.220**	.231**	.261**	.341**	.193**	1.000
PED willingness T3	1.33	1.24	.306**	.219**	.275**	.321**	.143**	.266**	1.000
Social learning T1	65.88	17.96	.224**	.224**	.210**	.242**	.169**	.149**	.156**	1.000
Social learning T2	76.37	19.82	.320**	.268**	.284**	.331**	.113**	.219**	.204**	.418**	1.000
Social learning T3	87.09	23.94	.397**	.414**	.393**	.477**	.141**	.233**	.352**	.344**	.457**

Note. T1 = Time 1, T2 = Time 2, T3 = Time 3. PED = performance-enhancing drug.

p < .05. **p < .01.

In terms of temporal consistency of both social learning and PED use, the correlations reveal several interesting findings. For instance, while the social learning construct at time t is significantly and positively associated with itself at times t + 1 and t + 2, these associations are modest to moderate in strength (r = .344–.457). Suggesting, perhaps, that the social learning process is, indeed, a process. In fact, it appears to be a very dynamic process such that the behavioral and normative positions of one’s significant others regarding PED use is changing over time: so too are respondents’ definitions regarding doping, the extent to which they model their doping behavior after others, and their anticipated balance of rewards and punishments associated with PED use. Finally, respondents’ time t reported willingness to use PEDs in the near future is predictive of the actual use (i.e., “current” use at time t + 1 and t + 2), for which the bivariate associations were observed modest to moderate in strength (r = .007–.341).

Before testing the social learning model of PED use, nonconditional growth curve analyses are performed to assess if there are intraindividual differences in the developmental growth trajectories of PED use and the social learning process over three time points (i.e., intercept and slope for each repeated measure); otherwise, there is no need to further test the associations among these parameters (Karney & Bradbury, 1995). For each repeated measure, a nonconditional linear growth curve model was estimated. Fit indices including chi-square value, comparative fit index (CFI), Tucker–Lewis index (TLI), and root mean square error of approximation (RMSEA) were used to evaluate model fit (Hu & Bentler, 1999). Table 3 provides results for the nonconditional growth curve models for all repeated measures. In sum, we find that intraindividual differences exist in the developmental growth trajectories of PED use and the social learning process. More specifically, both PED use and the social learning process increase over time. In addition, those who begin with high and low levels of PED consumption and social learning replicate these patterns throughout the sports life cycle.

Table 3.

Univariate Latent Growth Curve Model of PED Use and Social Learning Components.

Variable	Intercept (initial level)		Linear slope		Covariance (intercept and slope)	Model fit
Variable	M	Variance	M	Variance	Covariance (intercept and slope)	CMIN	df	p	CMIN/df	CFI	TLI	RMSEA
Current PED use	0.52** (0.04)	0.24** (0.05)	0.32** (0.02)	0.06* (0.03)	0.10** (0.03)	0.484	3	.922	0.161	1.000	1.008	.001
PED willingness	0.66** (0.04)	0.12* (0.07)	0.34** (0.03)	0.11* (0.04)	0.02 (0.04)	5.538	3	.136	1.846	.957	.957	.041
Differential associations	18.16** (0.33)	14.44** (4.46)	2.38** (0.25)	6.78* (2.57)	2.52 (2.68)	6.118	3	.106	2.039	.974	.974	.045
Differential reinforcement	14.71** (0.22)	3.72* (2.10)	2.17** (0.17)	2.17* (1.26)	1.75 (1.30)	5.600	3	.133	1.867	.968	.968	.041
Imitation	9.55** (0.18)	4.48* (1.30)	0.81** (0.13)	1.49* (0.72)	0.48 (0.77)	9.860	3	.020	3.287	.941	.941	.067
Definitions	23.51** (0.38)	11.30** (6.16)	5.25** (0.30)	10.44* (3.87)	6.61* (3.85)	5.067	3	.167	1.689	.981	.981	.037
Social learning	65.94** (0.77)	124.03** (21.75)	10.61** (0.54)	44.39** (11.47)	10.06 (12.22)	6.41	3	.093	2.136	.986	.986	.047

Note. Standard error in parentheses. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; PED = performance-enhancing drug.

p < .05. **p < .01.

Figure 1 presents the results of our SEM analysis of the contemporaneous and lagged direct, indirect, and feedback effects between the social learning construct and our measures of doping behavior across the three waves of our panel data. The full results of this SEM are presented in the appendix. For the fitted AMOS model, the summary statistics—that is, CMIN/df (1.576), CFI (.957), standardized root mean residual (SRMR = .041) and RMSEA (.034)—are better than critical values and represent the goodness-of-fit for the proposed model.

Figure 1.

Structural equation model of the contemporaneous and lagged direct, indirect, and reciprocal effects of social learning and PED behavior (i.e., past use, current use, and willingness to use) over time (N = 510).

With regard to the effects of social learning on doping behavior, Figure 1 reveals modest contemporaneous effects of learning on “current” PED use consistently across each wave of data (B = .21, .25, and .32 at Waves 1, 2, and 3, respectively). Social learning is also associated with an inclination to use PEDs in the near future (b = .20, .39, and .49, respectively, for Waves 1, 2, and 3). Finally, social learning at time t has a lagged indirect effect on “current” PED use at time t + 1 mediated by respondents’ time t reported willingness to use PEDs in the near future.

With regard to the feedback/reciprocal effects of PED use on social learning, the results in Figure 1 reveal that “past” PED use is significantly and modestly associated with social learning measured during the three waves of data collection (b = .28, .28, and .16, respectively for Waves 1, 2, and 3). Likewise, the results in Figure 1 indicate that time t indicators of “current” PED use have a lagged association with social learning at time t + 1 (b = .17 and .16 for Waves 1–2 and 2–3, respectively). Combined, these results indicate both that doping behavior is a function of a social learning process and that, in turn, this social earning process is influenced by respondent’s PED use.

Finally, there are moderate levels of consistency in social learning over time (b = .58 and .51 between Waves 1 and 2 and between Waves 2 and 3, respectively). Doping behaviors are also moderately associated with one another over time. The intrawave associations between “past” and “current” PED use range from b = .13 to b = 25. Likewise, respondents’ reported willingness to use PEDs at Time 1 are only modestly associated with their actual/“Current” PED use at time t + 1 (b = .13 and .11, respectively from Waves 1 to 2 and from Waves 2 to 3).

Discussion

The purpose of this study was to test a social learning model of doping behavior (PED use) with three waves of panel data from professional athletes residing in Iran. In general, the social learning model of doping was confirmed. This overall finding is consistent with prior research on general deviance (e.g., Pratt et al., 2010), as well as with prior research on sports doping (e.g., Donovan et al., 2002; Lucidi et al., 2008). In addition, the results indicate that the behavioral process of doping remains relatively stable through time. The social learning process has modest to moderate contemporaneous and lagged, direct and indirect effects on doping behavior and, importantly, doping behavior, in turn, has modest to moderate, direct and lagged reciprocal influences on the social learning process.

Akers (2009) explicitly states that not only does differential association lead to deviant behavior (socialization), but deviant behavior also has feedback/reciprocal effects on the social learning process. Involvement in PED use likely increases one’s association with those who use, or support the use of, PEDs, while reducing their interaction with more conforming others. In turn, involvement in PED use also influences one’s definitions and refines one’s expected balance of rewards and punishments (differential reinforcement). The direction of these feedback effects is a function of the extent to which the consequences of PED use were differentially reinforced. Consequently, the reciprocal effects we found in the current analyses are consistent with both theory (Akers, 2009) and prior longitudinal research of social learning theory (e.g., Akers & Lee, 1996; Cochran et al., 2017; Powers et al., 2017).

Policy Implications

Social learning theory is based on the important assumption that deviant behaviors, like all social behaviors, are formed in a social context. From this perspective, according to the findings of this research, professional athletes, as persons embedded into the sporting world as one of their primary social milieus, who are more exposed to the behavioral and normative patterns of deviant people (differential associations with people who use PEDs and/or who consider doping as good/appropriate, who serve are role models, and who are a primary source of the rewards and costs associated with PED use) are, themselves, likely to use PEDs. These athletes will also calculate the potential costs and benefits of their own use of PEDs and will form their own attitudes toward PED use. If the potential benefits of doping are greater than possible costs, then they are more likely to define PED use positively, or at least are better able to neutralize negative definitions of doping, and will, in turn, be more inclined to use PEDs.

Although policies aimed at educating athletes of the various adverse effects associated with PED use could prove effective, it has been shown that anabolic steroid users demonstrate sophisticated pharmacologic knowledge and are aware of the associated health risks (Perry et al., 1990). Increasing the certainty and/or severity of punishment could be used as an alternative to curtail PED use among professional athletes. Coaches and teammates would be less likely to suggest PED use if sports clubs were systematically fined or penalized for such usage. Perhaps a combination of increased education and increased accountability would decrease acceptance of PEDs among athletes (see, for example, Kondric et al., 2011). Success with such programming could have positive trickle-down effects for future professional athletes. If young athletes do not idolize those who use PEDs, they too may be less likely to use them.

Study Strengths and Limitations

Although Akers’s (2009) social learning theory argues for the existence of a reciprocal relationship between the social learning process and deviant behavior, few studies empirically test this hypothesized relationship. The few true tests of social learning theory that have assessed causality and reciprocal/feedback effects have found general support (e.g., Akers & Lee, 1996; Cochran et al., 2017; Krohn et al., 1985; Powers et al., 2017); however, they were all conducted using samples of American students. The current study was the first to examine the hypothesized reciprocal effects of the social learning process and deviant behavior using a non-American sample. In addition, this study is among the first longitudinal examinations of PED use that was specifically designed to assess the efficacy of theoretically derived variables. The study was based on a random sample of professional athletes, which makes it even more intriguing. In addition, the sample comes from Iran, a country rarely examined by social scientists, which gives our findings important comparative utility.

Conversely, notable limitations exist. First, some attrition occurred between the first round of data collection and subsequent rounds, which may affect the validity of the results if the retained participants are fundamentally different from those who discontinued (Jennings & Reingle, 2014). Second, due to regional differences, the findings may not be generalizable to non-Iranian athletes. Third, we asked our respondents about past, present, and future behaviors. It is possible the athletes were unable to recall past use or unwilling to disclose current use due to fear of sanction. Thus, the self-report nature of data may be limited by social desirability bias (Jennings & Reingle, 2014). To combat this effect, participants were informed numerous times as to the voluntary and anonymous nature of participation. Fourth, and relatedly, the measures were based on survey data collected from a single source—the athlete. This can contribute to common method variance, and future research should address this by collecting data from multiple sources (e.g., Lindell & Whitney, 2001).

Footnotes

Appendix

Structural Equation Model of the Contemporaneous and Lagged Direct, Indirect, and Reciprocal Relationships between Social Learning and PED Use Over Time (N = 510).

Independent Variable		Dependent Variable	Direct effect	Indirect effect	R ²
Past PED use	→	Current PED Use T3	.15**(.13)	.36**(.32)
Social learning T1	→	Current PED Use T3	—	.12**(.24)
Current PED use T1	→	Current PED Use T3	.13*(.09)	.25**(.19)
PED willingness T1	→	Current PED Use T3	—	.06**(.04)
Social learning T2	→	Current PED Use T3		.19**(.29)
Current PED use T2	→	Current PED Use T3	.36**(.28)	.06**(.05)
PED willingness T2	→	Current PED Use T3	.13**(.11)	—
Social learning T3	→	Current PED Use T3	.17**(.32)	—
R² of current PED use T3					.481
Past PED use	→	PED Willingness T3	—	.31**(.26)
Social learning T1	→	PED Willingness T3	—	.10**(.19)
Current PED use T1	→	PED Willingness T3	—	.22**(.16)
PED willingness T1	→	PED Willingness T3	—	.01**(.01)
Social learning T2	→	PED Willingness T3	—	.18**(.27)
Current PED use T2	→	PED Willingness T3	—	.10*(.08)
Social learning T3	→	PED Willingness T3	.27**(.49)
R² of PED willingness T3					.238
Past PED use	→	Social Learning T3	.35**(.16)	.80**(.37)
Social learning T1	→	Social Learning T3	—	.39**(.39)
Current PED use T1	→	Social Learning T3	.50**(.19)	.34**(.13)
PED willingness T1	→	Social Learning T3	—	.05*(.02)
Social learning T2	→	Social Learning T3	.63**(.51)	.05**(.04)
Current PED use T2	→	Social Learning T3	.36**(.16)	—
PED willingness T2	→	Social Learning T3
R² of social learning T3					.657
Past PED use	→	PED Willingness T2	—	.19**(.19)
Social learning T1	→	PED Willingness T2	—	.11**(.24)
Current PED use T1	→	PED Willingness T2	—	.08*(.07)
Social learning T2	→	PED Willingness T2	.22**(.39)	—
R² of PED willingness T2					.149
Past PED use	→	Current PED Use T2	.13**(.14)	.20**(.20)
Social learning T1	→	Current PED Use T2	—	.10**(.23)
Current PED use T1	→	Current PED Use T2	.26**(.23)	.05*(.04)
PED willingness T1	→	Current PED Use T2	.14**(.13)
Social learning T2	→	Current PED Use T2	.14**(.25)
R² of current PED use T2					.271
Past PED use	→	Social Learning T2	.50**(.28)	.38**(.22)
Social learning T1	→	Social Learning T2	.47**(.58)	.03**(.04)
Current PED use T1	→	Social Learning T2	.36**(.17)
R² of social learning T2					.621
Past PED use	→	PED Willingness T1	—	.05**(.06)
Social learning T1	→	PED Willingness T1	.08**(.20)	—
R² of PED willingness T1					.041
Past PED use	→	Current PED Use T1	.21**(.25)	.05**(.06)
Social learning T1	→	Current PED Use T1	.08**(.21)	—
R² of current PED use T1					.135
Past PED use	→	Social Learning T1	.61** (.28)	—
R² of social learning T1					.079

Note. T1 = Time 1, T2 = Time 2, T3 = Time 3. PED = performance-enhancing drug.

p < .05. **p < .01.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

C. Jordan Howell

Notes

Author Biographies

Saeed Kabiri earned his master’s degree in sociology at the University of Guilan (2012) and his PhD in Social Problems of Iran at the University of Mazandaran (2017). He has published several papers about the sociology and criminology of sports.

Seyyedeh Masoomeh (Shamila) Shadmanfaat earned her master’s degree in sociology at the University of Guilan (2016) and has published several papers about the sociology and criminology of sports with a focus on gender differences.

C. Jordan Howell is the research coordinator for the Evidence-Based Cybersecurity Research group at Georgia State University, and a doctoral candidate in criminology at the University of South Florida. He has published several papers in theoretical criminology and his primary research interest is in cybercrime.

Christopher Donner is an assistant professor of criminal justice and criminology; he earned his doctorate in criminology in 2013 from the University of South Florida. His research interests are in policing and criminological theory.

John K. Cochran is a professor of criminology at the University of South Florida. He earned his doctorate in sociology in 1987 at the University of Florida. He has more than 100 peer-reviewed publications, and his interests are criminological theory and capital punishment.

References

Akers

R. L.

(1973). Deviant behavior: A social learning approach. Wadsworth.

Akers

R. L.

(2009). Social learning and social structure: A general theory of crime and deviance. Routledge.

Akers

R. L.

Jensen

G. F.

(2011). The empirical status of social learning theory of crime and deviance: The past, present, and future. In Cullen

F. T.

Wright

J. P.

Blevins

K. R.

(Eds.), Taking stock: The status of criminological theory. Transaction Press.

Akers

R. L.

Krohn

M. D.

Lanza-Kaduce

Radosevich

(1979). Social learning and deviant behavior: A specific test of a general theory. American Sociological Review, 44, 636–655.

Akers

R. L.

Lee

(1996). A longitudinal test of social learning theory: Adolescent smoking. Journal of Drug Issues, 26(2), 317–343.

Akindutire

I. O.

Olanipekun

J. A.

(2015). Gender consideration of doping behavior of athletes in Southwest Nigeria. International Journal of Innovative Science, Engineering, & Technology, 2, 48–53.

Angoorani

Narenjiha

Tayyebi

Ghassabian

Ahmadi

Assari

(2012). Amphetamine use and its associated factors in body builders: A study from Tehran, Iran. Archives of Medical Science, 8(2), Article 362.

Chan

D. K.

Lentillon-Kaestner

Dimmock

J. A.

Donovan

R. J.

Keatley

D. A.

Hardcastle

S. J.

Hagger

M. S.

(2015). Self-control, self-regulation, and doping in sport: A test of the strength-energy model. Journal of Sport and Exercise Psychology, 37(2), 199–206.

Cochran

J. K.

Maskaly

Jones

Sellers

C. S.

(2017). Using structural equations to model Akers’ social learning theory with data on intimate partner violence. Crime and Delinquency, 63(1), 39–60.

10.

Donovan

R. J.

Egger

Kapernick

Mendoza

(2002). A conceptual framework for achieving performance enhancing drug compliance in sport. Sport Medicine, 32, 269–284.

11.

Dunn

Mazanov

Sitharthan

(2009). Predicting future anabolic-androgenic steroid use intentions with current substance use: Findings from an Internet-based survey. Clinical Journal of Sports Medicine, 19(3), 222–227.

12.

Fornell

Larcker

D. F.

(1981). Structural equation models with unobservable variables and measurement error: Algebra and statistics. Journal of Marketing Research, 18(3), 382–388.

13.

Haynie

D. L.

Osgood

D. W.

(2005). Reconsidering peers and delinquency: How do peers matter? Social Forces, 84(2), 1109–1130.

14.

L.-t.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55.

15.

Jennings

W. G.

Reingle

J. M.

(2014). Criminological and criminal justice research methods. Wolters Kluwer.

16.

Kabiri

Cochran

J. K.

Severson

Shadmanfaat

S. M.

Rahmati

M. M.

Sharepour

(2018a). Social and personal controls and performance enhancing drug use: Toward an explanation of doping activity among professional athletes in Rasht, Iran. Deviant Behavior, 39(11), 1483–1496.

17.

Kabiri

Cochran

J. K.

Stewart

B. J.

Sharepour

Rahmati

M. M.

Shadmanfaat

S. M

. (2018b). Doping among professional athletes in Iran: A test of Akers’s social learning theory. International Journal of Offender Therapy and Comparative Criminology, 62(5), 1384-1410.

18.

Karney

Bradbury

T. N.

(1995). Assessing longitudinal change in marriage: An introduction to the analysis of growth curves. Journal of Marriage and Family, 57(4), 1091–1108.

19.

Kirby

Moran

Guerin

(2011). A qualitative analysis of the experiences of elite athletes who have admitted to doping for performance enhancement. International Journal of Sport Policy and Politics, 3, 205–224.

20.

Kline

R. B.

(2015). Principles and practice of structural equation modeling. Guilford Press.

21.

Kondric

Sekulic

Petroczi

Ostojic

Rodek

Ostojic

(2011). Is there a danger for myopia in anti-doping education? Comparative analysis of substance use and misuse in Olympic racket sports calls for a broader approach. Substance Abuse Treatment, Prevention, and Policy, 6, 27–39.

22.

Krohn

M. D.

Skinner

W. F.

Massey

J. L.

Akers

R. L.

(1985). Social learning theory and adolescent cigarette smoking: A longitudinal study. Social Problems, 32(5), 455–473.

23.

Lazuras

Barkoukis

Rodafinos

Tzorbatzoudis

(2010). Predictors of doping intentions in elite-level athletes: A social cognition approach. Journal of Sport and Exercise Psychology, 32(5), 694–710.

24.

Lindell

M. K.

Whitney

D. J.

(2001). Accounting for common method variance in cross-sectional research designs. Journal of Applied Psychology, 86, 114–121.

25.

Lucidi

Zelli

Mallia

Grano

Russo

P. M.

Violani

(2008). The social-cognitive mechanisms regulating adolescents’ use of doping substances. Journal of Sports Science, 26, 447–456.

26.

Manouchehri

Tojari

(2013). Examining the conceptual model: Relationships between sport orientation, doping attitude and doping behavior in Iranian elite martial artists. European Journal of Experimental Biology, 3(2), 175–182.

27.

Matsueda

R. L.

Anderson

(1998). The dynamics of delinquent peers and delinquent behavior. Criminology, 36(2), 269–308.

28.

Melzer

Elbe

Brand

(2010). Moral and ethical decision-making: A chance for doping prevention in sports? Nordic Journal of Applied Ethics, 4, 69–85.

29.

Nunally

(1978). Psychometric theory. McGraw-Hill.

30.

Ozdemir

Nur

Bagcivan

Bulut

Sumer

Tezeren

(2005). Doping and performance enhancing drug use in athletes living in Sivas, Mid-Anatolia: A brief report. Journal of Sports Science & Medicine, 4(3), 248–252.

31.

Papadopoulos

F. C.

Skalkidis

Parkkar

Petridou

(2006). Doping use among tertiary education students in six developed countries. European Journal of Epidemiology, 21, 307–313.

32.

Payne

D. C.

Cornwell

(2007). Reconsidering peer influences on delinquency: Do less proximate contacts matter? Journal of Quantitative Criminology, 23(2), 127–149.

33.

Perry

P. J.

Andersen

K. H.

Yates

W. R.

(1990). Illicit anabolic steroid use in athletes: A case series analysis. The American Journal of Sports Medicine, 18(4), 422–428.

34.

Peters

R. J.

Jr. Adams

L. F.

Barnes

J. B.

Hines

L. A.

Jones

D. E.

Krebs

K. M. A.

Kelder

S. H.

(2005). Beliefs and social norms about ephedra onset and perceived addiction among college male and female athletes. Substance Use and Misuse, 40, 125–135.

35.

Powers

R. A.

Cochran

J. K.

Maskaly

Sellers

C. S.

(2017). Social learning theory, gender, and intimate partner violent victimization: A structural equations approach. Journal of Interpersonal Violence. Advance online publication. https://doi.org/10.1177/0886260517710486

36.

Pratt

T. C.

Cullen

F. T.

Sellers

C. S.

Winfree

L. T.

Jr. Madensen

T. D.

Daigle

L. E.

Gau

J. M.

(2010). The empirical status of social learning theory: A meta-analysis. Justice Quarterly, 27(6), 765–802.

37.

Preacher

K. J.

Wichman

A. L.

Briggs

N. E.

MacCallum

R. C.

(2008). Latent growth curve modeling. SAGE.

38.

Sellers

C. S.

Cochran

J. K.

Winfree

L. T.

Jr. (2003). Social learning theory and courtship violence: An empirical test. In Akers

R. L.

Jensen

G. F.

(Eds.), Social learning theory and the explanation of crime (pp. 109–127). Transaction Press.

39.

Shadmanfaat

S. M.

Howell

C. J.

Muniz

C. N.

Cochran

J. K.

Kabiri

Fontaine

E. M.

(2019). Cyberbullying perpetration: An empirical test of social learning theory in Iran. Deviant Behavior. Advance online publication. https://doi.org/10.1080/01639625.2019.1565513

40.

Sottas

P. E.

Robinson

Fischetto

Dollé

Alonso

J. M.

Saugy

(2011). Prevalence of blood doping in samples collected from elite track and field athletes. Clinical Chemistry, 57(5), 762–769.

41.

Sutherland

E. H.

(1947). Principles of criminology. Lippincott.

42.

Vouillamoz

Thom

Grisdale

Saugy

Giraud

Robinson

Geisendorfer

(2009). Anti-doping testing at the 2008 European football championship. Drug Testing and Analysis, 1(11–12), 485–493.

43.

Zorzoli

Rossi

(2010). Implementation of the biological passport: The experience of the international cycling union. Drug Testing and Analysis, 2(11–12), 542–547.