The development and validation of the Adolescent Sport Drug Inventory (ASDI) among athletes from four continents.

A significant barrier to understanding the psychosocial antecedents of doping use among adolescent athletes is the lack of valid measures. In order to address this issue, the first aim of this paper was to develop and validate the Adolescent Sport Drug Inventory (ASDI) among adolescent athletes from Asia, Europe, North America, and Oceania. The second aim was to assess the construct validity of the ASDI. As such, this paper is divided into two parts. Part 1 relates to the development of the ASDI and contains two studies: item development (Study 1) and factorial validity (Study 2). Part 2 contains information on how the psychosocial variables measured in the ASDI are associated with situational temptation, and honesty (Study 3), maturation (Study 4), stress and coping (Study 5), and coaching (Study 6). In devising the ASDI, 19 different models were examined, which culminated in a 9-factor, 43-item ASDI. Coping, mastery-approach goals, and cognitive-social maturity were associated with doping attitudes. Caring motivational climates, strong coach-athlete relationships, and positive coach behaviors were associated with athletes being less susceptible toward doping, which provides construct validity for the ASDI. The ASDI is a valid tool to assess the psychosocial factors associated with doping among adolescent athletes. This questionnaire can be used to identify athletes who are the most at risk of doping, assess how the psychosocial factors associated with doping change over time, and to monitor the impact of antidoping interventions for adolescent athletes. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

another questionnaire that has been used to assess doping attitudes among adolescent athletes 10 (e.g., Madigan, Stoeber, & Passfield, 2016). However, Nicholls, Madigan, and Levy (2017)  11 reported that the PEAS did not exhibit a good model fit for adolescent athletes. At the present 12 time, there is not a valid and reliable questionnaire to assess the psycho-social variables that are 13 associated with doping specifically among adolescent athletes. The lack of theory guided 14 questionnaires to assess the psycho-social doping variables among adolescent athletes, may be 15 due to the lack of theoretical models for adolescent athletes. 16 Only two theoretical models of doping were specifically designed for young people (e.g., developed an integrated model, which included distal (e.g., sportspersonship, past doping, and 19 achievement goals) and proximal (e.g., outcome expectance beliefs and self-efficacy beliefs) 20 predictors that influenced whether a young person intended to dope or not. Lazuras et al. found 21 that 57.2% of variance in doping intentions was predicted by the model. A limitation of this 22 model, however, is that it focused exclusively on psychological predictors of doping intentions, 23 would take them," and "What my coach thinks about PEDs would influence my decision about 1 whether I would take them"). In regards to stress, Nicholls, Perry, et al. (2015) identified stress 2 associated with negative outcomes of matches or competitions, and expectations placed on 3 athletes. As such, we ensured that this was reflective of the stress questions we created (e.g., 4 "There are lots of expectations on me to perform well" and "I feel nervous I will fail"). This 5 exact process of creating questions that reflected each factor was followed for each psycho-social 6 variable from the SDCM-AA (Nicholls, Perry, et al., 2015). This process culminated in questions 7 about threat ("If I took a PED, how likely is it that I would suffer serious health complications," 8 n = 10), benefit ("If I took PEDs, I would get much better at my sport," n = 10), self-esteem ("I 9 am worth being in the team/squads that I am currently involved with," n = 12), cheating ("I 10 would cheat if I thought it would help me win," n = 9), legitimacy ("Samples taken by drug 11 testers are securely looked after," n = 9), reference group opinion ("I wouldn't want my team 12 mates to think that I am a cheat," n = 8), age/maturation ("I am more physically developed than 13 most athletes my age," n = 10), stress ("I usually think that the outcome of matches/competitions 14 will be negative," n = 12), affordability/availability ("I know where to get PEDs from," n = 8), 15 doping susceptibility ("I would be tempted to take PEDs when I have an important competition," 16 n = 7), and "attitudes towards doping (e.g., "Legalising PEDs would benefit sport," n = 13). 17 Though clearly important, item-level analysis is seldom reported in studies. A method 18 that provides appropriate rigor was presented by Waltz and Bausell (1983). Specifically, these 19 authors developed the four-point Content Validity Index (CVI). In this process, a panel of 20 experts judge each item on a scale of one to four for relevance, clarity, simplicity, and ambiguity. 21 A proportion of agreement is then calculated, with scores on the CVI of < .75 generally 22 considered strong. 23

Method 1
Participants 2 Three sport psychologists and one coach (four males), who were aged between 24 and 55 3 years old, took part in Study 1. The sport psychologists' experience ranged between 2 and 19 4 years and the coach had 18 years' coaching experience from the United Kingdom n = 3) or 5 Australia n = 1). All participants were independent of the research team. 6 Procedure 7 A departmental ethics committee granted ethical approval for this study. Before 8 participating in the research, participants were required to provide written informed consent. 9 Adolescent Sport Doping Inventory 10 The preliminary ASDI was drawn up by the research team and contained 108 items 11 pertaining to the psycho-social variables associated with doping. These items related to attitudes 12 towards doping, threat, benefit, self-esteem, cheating, legitimacy, reference group opinion, 13 age/maturation, stress, doping susceptibility, and affordability/availability. 14 Data Analysis 15 To examine content validity, each psychologist and coach rated items on the 4-point CVI 16 (Waltz & Bausell, 1983). The criteria can be found in Electronic Supplementary Material (ESM) 17 Appendix S1. Each panel member rated each item according to the criteria. CVI was calculated 18 by summing the amount of responses for each item of three or four. This was divided by the total 19 items to be expressed as a fractional proportion. All items with a CVI over .75 were considered 20 to have sufficient content validity. 21

22
Mean CVI scores by item for relevance, clarity, simplicity, and ambiguity are presented 1 in ESM Appendix S2, with item CVI and subscale CVI. In total, seven items presented a CVI 2 below .75. Each of these were reviewed to determine if they could be revised, without replicating 3 an existing item. Four items on the threat scale presented a CVI < .75 and were revised and 4 retained. One item from the attitudes scale and one item from the cheating scale could not be 5 revised without replicating another item and these were therefore removed, yielding a subscale 6 CVI scores of .90 for both of these subscales. One item from the age/maturation scale was 7 revised but to avoid replication, two further items from this scale were removed. CVI scores for 8 all non-revised scales ranged from .89 to .97. 9 Study 2: Factorial validity 10 The purpose of Study 2 was to examine the internal structure of the scale generated in 11 Study 1. Specifically, we tested the factor structure of the preliminary ASDI and refined this 12 through an iterative process, to derive a psychometric assessment with factors that demonstrate 13 relative independence and generate internally consistent scores. 14

Data Analysis 2
Confirmatory factor analysis (CFA) was performed on the initial model. Factors were 3 anticipated to be relatively independent, hence no cross-loadings were specified, creating an 4 Independent Cluster Model (ICM). Scale refinement was an iterative process, examining model 5 fit, standardized parameter estimates (loadings), and modification indices. At the examination of 6 each model, fit indices were assessed by broadly employing Hu and Bentler's (1999) 7 recommendations, the Comparative Fit Index (CFI), the Root Mean Square Error of 8 Approximation (RMSEA), and Tucker-Lewis Index (TLI) of close to .95 were considered as 9 demonstrating good incremental model fit (that is, compared to a null model), and the 10 standardized root mean-square residual (SRMR) and root mean square error of approximation 11 close to .08 and .05 respectively indicate good absolute model fit. To examine the adequacy of 12 factor loadings pertaining to each item, we employed Comrey and Lee's (1992) 13 recommendations of .32 (poor), .45 (fair), .55 (good), .63 (very good), and 0.70 (excellent). 14

15
The first model subjected to CFA was the 11-factor, 104-item scale developed in Study 1. 16 The results indicated substantive misspecification in this model; χ 2 (5192) = 13897.1, CFI = 17 .736, TLI = .728, SRMR = .075, RMSEA (90% CI) = .053 (.052, .054). The subsequent iterative 18 process to refine the model resulted in constructing and testing 19 different models (Table 1). 19 These analyses culminated in the final 9-Factor, 43-item ASDI (Appendix A for ASDI and  20 scoring key), which presented good CFA model fit; χ 2 (5192) = 1440.4, CFI = .954, TLI = .950, 21 SRMR = .039, RMSEA (90% CI) = .035 (.032, .038). Age/Maturation and 22 Availability/Affordability were removed in the model refinement process. The Age/Maturation 23 scale had been reduced to just three items after removing items with a weak factor loading. We 1 considered that there are better mechanisms for measuring maturity than including it as a scale 2 within the ASDI and therefore removed it. Although item loadings on the 3 Availability/Affordability scale were acceptable, we believed that this was a practical 4 consideration, whereas all other scales represented psychological or social constructs. All 5 retained items presented good (i.e., > .55) factor loadings ( Table 2). The largest correlation 6 between all subscales was .57 (Table 3). Item redundancy was inspected using R-square. All 7 retained items made a satisfactory contribution to the overall variance (R 2 > .30). To measure 8 cross-loadings on the refined model, we examined the exploratory structural equation modelling 9 (ESEM) structure, whereby all factors are indicated by all items. This also presented a good 10 model fit; χ 2 (552) = 964.4, CFI = .969, TLI = .950, SRMR = .017, RMSEA (90% CI) = .035 11 (.032, .039) with all items loading substantively on their intended factor and no substantive 12 cross-loadings, further supporting the independence of factors (Table 2). Finally, we examined 13 the internal consistency of the responses to each scale using Cronbach's alpha. All scale scores 14 demonstrated suitably reliable estimates (.78 to .95; see Table 3). 15

PART 2: CONSTRUCT VALIDITY ANALYSIS 16
We assessed the construct validity of the newly created ASDI in Part 2. Campbell and 17 Fiske (1959) described construct validity by referring to its subordinates; convergent and 18 divergent validity. Convergent validity is evidenced by a construct that is positively associated 19 with theoretically related constructs. Conversely, divergent validity is indicated by theoretically 20 independent variables yielding no association. 21

Study 3: Psycho-Social Doping Variables, Temptation, and Honesty 22
We examined the convergent validity of the ASDI by its association with a measure of 1 doping attitudes, situational temptation, and honesty and humility. Firstly, we selected doping 2 attitudes because one of the subscales in the ASDI represents doping attitudes, and the other 3 eight sub-scales have been either theoretically or empirically associated with doping attitudes 4 extent to which associations between doping attitudes and situational temptation, and doping 10 attitudes and honest and humility differed from the associations between these variables and 11 ASDI scales provided an assessment of divergent validity. We constructed a structural equation 12 model (SEM), whereby situational temptation and honesty and humility were predictor variables 13 of ASDI scales and doping attitudes. We also examined the factor structure of the ASDI on a 14 sample independent of participants used in Study 2. 15

Methods 16
Participants 17 A sample of 423 athletes took part in this study. We included a social desirability scale in 18 the questionnaire pack for this study (Petrides, 2009). Thirty athletes scored above the acceptable 19 threshold and therefore their data was removed. As such, the sample analyzed 2009) were used to assess social desirability. Two of the questions were inserted at the end of the 14 PEAS and the Honesty and Humility questions. As these items are not intended to be related to 15 each other, Cronbach's alpha was not calculated. Participants who scored in excess of 20 out of a 16 maximum of 28, were deemed to be supplying socially desirable answers and therefore removed. 17

Data Analysis 18
Data were screened for completeness, outliers, univariate normality, and social 19 desirability. We examined internal consistency by estimating omega point estimates and 20 confidence intervals in addition to coefficient alpha, as omega holds fewer assumptions than 21 alpha (Dunn, Baguley, & Brunsden, 2013). As we had large variations in length of scale, we also 22 calculated mean inter-item correlation (MIIC). 23 We examined the factor structure of the ASDI, using CFA and ESEM (Asparouhov & 1 Muthén, 2009). The main analyses comprised of testing a SEM positing situational temptation, 2 honesty and humility, and exogenous predictor variables of ASDI scales. Doping attitude was 3 co-varied with all ASDI factors. 4 Results 5

Preliminary Analyses 6
There were no missing data or outliers identified (see ESM Appendix S3 for descriptive 7 statistics). Univariate skewness was < 2 in all variables with the exception of the attitudes scale 8 of the ASDI, which was slightly positively skewed, with a large proportion of participants 9 scoring the minimum on this scale. Standardized parameter estimates for all factor loadings are presented in Table 4. The loadings 1 clearly support the factor structure of the ASDI in the ICM. The ESEM model with geomin 2 rotation allowed all items to load on all subscales. Model fit was again good; χ 2 (552) = 1079.89, 3 p < .001, CFI = .948, TLI = .915, SRMR = .019, RMSEA = .049 (90% CI = .045, .054). The 4 priority however, was to check that all items loaded onto their intended scale sufficiently and that 5 cross-loadings were not substantive. The factor loadings indicated that all items load 6 substantively onto their own factors and no cross-loadings on any factor were greater than .25. 7 This supports the factor structure and the independence of each scale within the ASDI. 8

Study 4: Psycho-Social Doping Variables and Maturation 16
Adolescence is associated with dramatic biological and psychological changes (Lazarus, 17 1999 to their lack of maturity. As such, it is likely that maturation levels be related to attitudes and 22 susceptibility. 23 The aim of Study 4 was to examine the convergent validity of the ASDI, by exploring the 1 relationship between psycho-social constructs associated with doping and maturity. We predicted 2 a negative relationship between biological maturity, cognitive-social maturity, and emotion 3 maturity with doping attitudes and susceptibility ( Data from all measures was screened for outliers, missing data, and univariate normality. 7 We assessed internal consistency using omega point estimates and bootstrapped confidence 8 intervals. We then ran a hierarchical multiple linear regression to determine the statistically 9 predictive capabilities of maturity and ASDI variables on doping susceptibility. 10

Results 11
Less than 1% of cells contained missing data and there were no outliers. Descriptive 12 statistics, normality estimates, and omega point estimates are presented in ESM Appendix S5. 13 There were no issues with skewness (all scales < 2). All subscale scores comfortably exceeded 14 the generally acceptable level of ω > .70 for estimates of internal consistency. Indeed, all ASDI 15 exceeded .80. 16 Assigning z scores for biological maturity, we found that of the 204 whom provided 17 sufficient data to calculate this variable, 115 (56.37%) were early in their maturation, 21 18 (10.29%) were on time, and 68 (33.33%) were late. A one-way ANOVA to determine whether 19 there were differences among the groups, yielded no significant differences. 20 To gain initial insight of variable associations, we examined the Pearson bivariate 21 correlations with 1,000 bootstrapped samples of ASDI scales with biological, emotional, and 22 social cognitive maturity (see ESM Appendix S6). Correlations were interpreted following the 23 recommendations of Li, Peng, Zhang, and Zhu (2012) of < .20 = no correlation, .20-.39 = low 1 correlation, .40-.59 = moderate correlation, .60-.79 = moderately high correlation, and > .80 = 2 high correlation. Biological maturity was unrelated to doping constructs, while emotional and 3 social cognitive maturity was negatively associated with doping susceptibility. 4 Next, we conducted a hierarchical multiple linear regression to determine the extent to 5 which variance in doping susceptibility was account for by maturity and the remaining ASDI 6 variables. First, we entered demographic variables of gender, ethnicity, skill level, and years' 7 experience in Model 1, then maturity variables in Model 2, and finally, the eight remaining ASDI 8 subscales in Model 2. Confidence intervals were obtained from 1,000 bootstrapped samples. The 9 results from this analysis are presented in Table 5. Model one (demographics) was not 10 statistically significant (ΔR 2 = .030, F (4,301) = 2.335, p = .056). Model two explained a 11 substantive amount of variance (ΔR 2 = .213, F (7,258) = 13.673, p < .001). This was a 12 cumulative effect of the three maturity variables, however, as none of them presented statistically 13 significant coefficients. ASDI variables were then entered in model three. Overall, 66.4% of 14 doping susceptibility variance was accounted for, as Model 3 also substantively increased R 2 15 (ΔR 2 = .421, F (15,290) = 38.197, p < .001). Three ASDI scales significantly contributed to the 16 increased variance in doping susceptibility explained; benefit (β = 14, p < .01), cheating (β = 35, 17 p < .01), and reference group (β = 38, p < .01). 18

Study 5: Psycho-Social Doping Variables, Psychological Stress, Achievement Goals, 19
Emotions, and Coping 20 Nicholls, Perry, et al (2015) identified stress as a key factor that may influence attitudes 21 towards doping among adolescent athletes. However, little is known about how stress may be 22 associated with the psycho-social constructs linked to doping. The aim of Study 5 was to further 23 examine the convergent validity of the ASDI, by exploring the relationship between psycho-1 social constructs associated with doping and stress appraisals, achievement goals, and coping. 2 Based on Nicholls, Perry, et al (2015), we hypothesized that threat appraisals would correlate 3 positively, whereas as challenge appraisals would correlate negatively with attitudes to doping. 4 We also predicted that there will be positive relationships between performance-approach and 5 performance-avoidance goals with doping attitudes and doping susceptibility, but negative 6 relationships between attitudes to doping with mastery-approach and mastery-avoidance goals. 7 Finally, task-oriented coping strategies would correlate negatively with favorable attitudes 8 towards doping, whereas distraction-oriented and disengagement-oriented coping would 9 correlate positively with doping attitudes. This was because athletes using distraction-and 10 disengagement-oriented coping are less likely to be successful with such strategies ( professional (n = 2), county or state (n = 22), national (n = 9), or international (n = 5) levels. 20 Three athletes failed to report their skill level. 21

Data Analyses 10
Data from all measures was screened for outliers, missing data, and normality. Given the 11 complexity of model required to assess the associations between variables, we tried to limit the 12 number of parameters to be estimated in order to achieve Bentler and Chou's (1987) 13 recommendation of a ratio of five cases per free parameter. For the main analyses, we tested a 14 series of path models whereby ASDI subscales were posited as exogenous variables. These were 15 predictors of achievement goal variables, which in turn were predictors of stress appraisal and 16 finally, these were posited as predictors of coping. 17

18
There were no concerns regarding missing data (< 1%) or outliers. Descriptive statistics, 19 normality estimates, and omega point estimates are presented in ESM Appendix S7. All 20 subscales presented normal distribution and subscale data comfortably exceeded the generally 21 acceptable level of ω > .70 for internal consistency. Indeed, all scale responses exceeded .80, 22 with the exception of disengagement-oriented coping (ω = .70, 95% CI = .63, 75). 23 We examined the correlations of ASDI scales with achievement goals, stress appraisal, 1 and coping strategies. Correlations were generally low (see ESM Appendix S8), although esteem 2 and stress appeared to have the strongest relationship with other variables. Next, we conducted a 3 hierarchical multiple linear regression to determine the extent to which doping susceptibility was 4 predicted by the remaining ASDI variables. First, we entered demographic variables of gender, 5 ethnicity, skill level, and years of playing experience in Model 1, before entering the eight 6 remaining ASDI subscales in Model 2. The results from this analysis are presented in ESM 7 Appendix S9. Model 1 (demographics) revealed minimal effect (ΔR 2 = .035, F (4,356) = 3.20, p 8 = .013). Overall, 65.5% of doping susceptibility variance was accounted for, as Model 2 9 substantively increased R 2 (ΔR 2 = .620, F (12,348) = 55.06, p < .001). Four ASDI scales 10 significantly contributed to the increased variance in doping susceptibility (e.g., attitude, benefit, 11 cheating, and reference group opinion). 12

Path Analyses 13
The first path model constructed was a mediation model, whereby coping strategies were 14 regressed on stress appraisals, which were regressed on achievement goals, which were regressed 15 on ASDI scales. Mastery-approach was covaried with performance approach and mastery-16 avoidance was covaried with performance-avoidance to better represent the relationship between 17 these variables. This model required the estimation of 78 parameters, presenting a ratio to predictor of all endogenous variables, achievement goals could be predictors of stress appraisals 1 and coping strategies, but not of ASDI scales. Stress appraisals could predict coping strategies, 2 but coping strategies, as the final variables in the model, could not act as predictor variables. 3 Specifically, we added paths so that task-oriented coping was predicted by esteem and 4 legitimacy, challenge appraisal was predicted by esteem, legitimacy, and stress, and threat .001, 95% CI = .09, .36) goals. Attitude was negatively predictive of both mastery-approach and 16 mastery-avoidance. Stress presented a positive path to mastery avoidance and performance 17 avoidance. Notably, stress was also a significant predictor of threat appraisals (β = .51, p < .001, 18 95% CI = .39, .62). Finally, we examined indirect effects throughout the model. The results of 19 this analysis are presented in ESM Appendix S10. The most significant indirect effect was stress 20 via threat appraisals leading to disengagement coping (γ = .28, p < .001, 95% CI = .20, .36). 21

Study 6: Psycho-Social Doping Variables and Coaching Factors 22
The sporting environment that a coach creates is associated with attitudes among athletes 1 (Christodoulidis, Papaioannou, & Digelidis, 2001). It is therefore plausible that the motivational 2 climate, the coach-athlete relationship, and coaching behavior may be linked to doping attitudes, 3 because coaches can exert a strong influence on young athletes (Wrobble et al., 2002). Indeed, 4 Terney and McLain (1990) reported that 2% of athletes said a coach had recommended anabolic 5 androgenic steroids (AAS). The aim of Study 6 was to examine the construct validity of the ASDI, 6 by exploring the relationship between psycho-social constructs associated with doping and the 7 motivational climate, coach-athlete relationship, and coach behavior. 8 We predicted that attitudes to doping would be negatively associated with an empowering 9 motivational climate, but positively associated with a disempowering motivational climate.

Data Analyses 14
All data were screened for outliers, missing data, normality, and internal consistency. We 15 ran a hierarchical multiple linear regression to determine the predictive capabilities of 16 environmental and ASDI variables on doping susceptibility. 17

General Discussion 23
The ASDI is a valid tool to assess the psycho-social factors associated with doping 1 among adolescent athletes, which has been tested with independent samples. Indeed, the findings 2 from Study 3 in Part 2, successfully replicated the factor structure of the ASDI created in Part 1. 3 Study 3 utilized an independent sample, and the ASDI demonstrated robust internal consistency 4 in responses for the second time. Second, the findings from Study 3 also provide additional 5 support for the convergent validity of the ASDI, and equivocal support for divergent validity. To 6 further examine convergent validity, we conducted studies identifying relationships with 7 maturation (Study 4), stress, emotions, and coping (Study 5), and coaching factors (Study 6), 8 which make unique contributions to the doping literature by identifying other factors that are 9 associated with doping attitudes and susceptibility. 10 We found partial support for our hypothesis that maturation was associated with doping 11 attitudes. Although biological maturity was not associated with doping attitudes, attitudes 12 towards doping correlated significantly with emotional maturity and the three subscales of 13 cognitive-social maturity. It should be noted, however, that the correlations were low. Nicholls, 14 PEDs may be used to help athletes manage negative emotions associated with their own 20 performance or insecurities about their appearance. This contention is supported by the finding 21 that stress levels were negatively associated with emotional maturity. For example, an athlete 22 may be angry or anxious about poor performance, and thus taking PEDs could eradicate such 23 negative emotions, because the athlete is likely to believe that his or her performance will 1 improve if PEDs are consumed. As such, doping may be a form of coping that allows athletes to 2 regulate their internal responses to stress. In regards to cognitive social maturity, all three 3 subscales correlated negatively with doping attitudes, which was expected. It unsurprising that 4 conscientiousness was negatively associated with doping attitudes. These findings imply young 5 people with high levels of conscientiousness see PEDs as bad and would therefore be less likely 6 to dope. Similarly, those who are less influenced by their peers are also more likely to have an 7 unfavorable view of doping. Peers may be a key factor in influencing whether a young person 8 will dope or not, because Wroble et al. (2002) found that 18% of young people that took AAS 9 did so because of peer pressure. 10 We found partial support for our hypotheses relating to stress, coping and emotions. 11 Study 5 represents one of the first attempts to examine the relationship between stress appraisals 12 and doping attitudes. The coaches who were interviewed by Nicholls, Perry, et al (2015)  13 suggested that stress may be a key factor in influencing whether athletes will dope. Although we 14 did not examine doping prevalence, doping attitudes predict doping among young people (Zelli 15 et al., 2010). The athletes who used disengagement-oriented coping and had a positive attitude 16 towards doping, may have believed that they could not be successful in their sport without taking 17 PEDs, which is may be why they gave up trying to achieve their goals. Only one form of 18 achievement goal, mastery-approach goals, was associated with doping attitudes. The direction 19 of the correlation was expected, but our finding suggests that goals are less related to doping 20 attitudes than other constructs within the cognitive-motivational-relational theory of emotions 21 (Lazarus, 1999). 22 Controlling coaching behavior was the only coach factor that was significantly associated 23 with doping attitudes. Another factor that predicts doping prevalence among young people is 1 susceptibility (Barkoukis et al., 2014). We found that susceptibility was associated with the 2 motivational climate, the coach-athlete relationship, and coach behaviors. That is, athletes who 3 were susceptible towards doping were in a controlling and uncaring environment, had a poor 4 relationship with their coach, and were coached with controlling behaviors. Cheating, stress, and 5 in particular, reference group opinion appeared to be the strongest predictors of doping 6 susceptibility. It could be argued that doping susceptibility may be influenced by a combination 7 of personality and individual differences (i.e., cheating), social factors (i.e., reference group 8 opinion), and states (i.e., stress levels). This is one of the first studies to identify factors that may 9 predict doping susceptibility. Indeed, Part 2 of this research lends support for the notion that 10 maturation, stress variables, and coaching are all related to the psycho-social variables that 11 predict doping. However, it should be noted that the effect sizes were relatively small for some 12 variables. Although this could be viewed as a limitation, it could be argued that doping is 13 predicted by many different variables that all make a small contribution. Another limitation of 14 this research is that we did not measure doping prevalence, and have inferred a relationship 15 based on previous findings with young athletes (Zelli et al., 2010). An additional limitation is 16 that our path analysis implies mediation in a cross-sectional design. Maxwell and Cole (2007)  17 pointed out that true mediation consists of causal processes and therefore such designs typical 18 create biased estimates. 19 Another potential limitation of this research relates to the use of self-reported 20 questionnaires. Although the ASDI may be affected by social desirability, scholars such as 21 Ntoumanis, Barkoukis, Gucciardi, and King Chun Chan (2017) argued that self-report 22 questionnaires are the most realistic way of capturing the psycho-social variables associated with 23 doping. Further, they may even provide a more accurate representation of actual doping use, 1 given that WADA (2018) reported 2% of samples contained banned substances, whereas 10% of 2 athletes admitted doping offences in a self-reported study (Lazuras et al., 2010). It should be 3 noted that all questionnaires throughout the six studies were completed anonymously. 4 In conclusion, the ASDI improves on existing measures by Bloodworth et al.