Perfectionism and learning experiences in dance class as risk factors for eating disorders in dancers

Authors

  • Kylie J. Penniment,

    1. School of Psychology and Speech Pathology, Curtin Health Innovation Research Institute, Curtin University, Australia
    Search for more papers by this author
  • Sarah J. Egan

    Corresponding author
    1. School of Psychology and Speech Pathology, Curtin Health Innovation Research Institute, Curtin University, Australia
    • School of Psychology and Speech Pathology, Curtin University, GPO Box U1987, Perth, WA 6845, Australia. Tel.: +61 89266 2367, Fax: +61 89266 3178.
    Search for more papers by this author

Abstract

There is strong evidence that perfectionism is a risk factor for eating disorders. Women who engage in dance training have been reported to be at risk for eating disorders, and it has been hypothesised that expectancies about thinness and restricting food intake are formed partly as a result of exposure to thinness related learning (TRL) experiences in this environment, which may increase their risk. To clarify the relative contribution of perfectionism and learning in accounting for eating disorder symptoms in this group, 142 female ballet dancers completed an online survey. Through structural equation modelling it was found that the association between perfectionism and eating disorder symptoms was partially mediated by learning about thinness and restriction. The results suggest that eating disorder symptoms in dancers are significantly influenced by the interaction of perfectionism and learning, and future research should investigate the efficacy of prevention programs to target these risk factors in female dancers. Copyright © 2011 John Wiley & Sons, Ltd and Eating Disorders Association.

Introduction

Female dancers are a high-risk group for the development of eating disorders (ED) (Anshel, 2004; Ringham, Klump, Kaye, Stone, Libman, & Stowe, 2006). Dancers score higher on measures of eating disturbance (e.g. Abraham, 1996; Ackard, Henderson, & Wonderlich, 2004; Neumarker, Bettle, Bettle, Dudeck, & Neumarker, 1998; Neumarker, Bettle, Neumarker, & Bettle, 2000), body dissatisfaction, (Bettle, Bettle, Neumarker, & Neumarker, 1998; Ravaldi, Vannacci, Zucchi, Mannucci, Cabras, & Boldrini, 2003; Szmukler, Eisler, Gillies, & Hayward, 1985), and have higher disturbances in weight control behaviours and eating attitudes (Dotti, Fioravanti, Balotta, Tozzi, Cannella, & Lazzari, 2002; Pierce, Daleng, & McGowan, 1993) compared to non-dancer controls, for a comprehensive review of literature examining dancers see Solomon and Solomon (1996).

Dancers spend innumerable hours practicing in front of mirrors where their bodies are closely scrutinised by self and others. If they apply high levels of perfectionism to dance and body shape, then the combination of sociocultural pressures for thinness inherent in the dance profession combined with expectations of high performance produce the ideal social climate for development of ED (Garner & Garfinkel, 1980). Druss and Silverman (1979) found that classical ballet students were at risk for developing ED and hypothesised that a prime reason was a drive for perfection because ‘…they practice exercises ritualistically to stop thinking, feeling, to reduce their bodies to the barest minimum of fat and muscle and to deny themselves all other pleasures in order to achieve that goal. They dare to hope for some brief time to become perfect’ (p. 121).

Sociocultural factors have been postulated to be of particular significance in the development of ED in dancers. Low weight is often required of ‘aesthetic performers’ such as dancers where a thinner body is emphasised due to the belief that this enhances performance (Jacobi, Hayward, de Zwaan, Kraemer, & Agras, 2004; LeGrange, Tibbs, & Noakes, 1994). Recently researchers have investigated specific risk factors within the dance subculture that make individuals more susceptible to ED. This higher risk has been attributed mainly to dispositional characteristics of dancers, such as perfectionism and low self-esteem (e.g. Anshel, 2004; Thomas, Keel, & Heatherton, 2005). Some researchers have also proposed that perfectionism may make dancers more vulnerable to the amplified sociocultural pressure for thinness inherent in the dance culture. For example, Thomas et al. (2005) argue that women with higher levels of perfectionism are more likely to dance at increasingly competitive levels, which causes them to become even more exposed to an environment where thinness and dieting is valued. Consequently, it is hypothesised that there is a selective relationship, where individuals high on perfectionism are more likely to either choose participation in more rigorous dance training or are more likely to be successful and thus reach higher levels of success in dance. It is also hypothesised there may be a reactive relationship involved, where individuals with higher perfectionism may react more strongly to TRL experiences in dance class.

Thinness and restricting expectancies

Annus and Smith (2009) proposed that participation in dance alone cannot explain ED risk in dancers, as dance training environments vary in the degree to which they emphasise thinness and dieting. In a sample of 500 college women who were asked to retrospectively report their participation in dance class and current ED symptoms on the EDE-Q (Fairburn & Beglin, 1994), it was found that the mere amount of involvement in dance class was unrelated to ED symptomatology; instead reports of learning experiences during dance classes concerning thinness or restricting food predicted adult disordered eating. Annus and Smith (2009) defined thinness and restricting expectancies (TRE) as ‘expectations that being thin or restricting food intake will lead to reinforcement, such as…becoming more attractive’ (p. 50). Thinness and restricting expectancies were measured on the 44 item thinness and restricting expectancy inventory (TREI; Hohlstein, Smith, & Atlas, 1998), an example of an item being ‘if I were thin, I would feel more worthwhile’. Annus and Smith (2009) defined thinness related learning (TRL) as the degree to which the individual is exposed to learning about thinness in a dance class, and measured this via the dance experience questionnaire (DEQ; Annus & Smith, 2009) which they designed for the study, an example item being ‘how often did the instructor emphasise weight or shape in class’ (p. 53). The types of learning experiences that embrace TRL include comments from teachers and peers about the benefits of dieting, social comparison between peers, conducting skinfold tests and weighing dance pupils in class and observational learning of dieting and restriction through teacher or peer modelling.

Annus and Smith (2009) found that TREs mediated the relationship between TRL and ED symptoms. They concluded that learning experiences about thinness are more important in predicting ED symptoms than time spent in dance classes. One problem, however, with the design of their study was that they relied on retrospective recall from adult women in regards to past dance class participation. This is a less accurate way of determining the influence of the learning environment on ED symptomatology than if women who were currently engaged in dance training were examined. Furthermore, no conclusions can be made regarding a causal relationship between learning experiences in dance class and ED symptoms as it was a retrospective design and a longitudinal design would be required to examine if learning experiences to lead to ED symptoms. Nevertheless, the results are interesting as Annus and Smith (2009) suggested that their results indicated that the dance environment is not a risk in itself, rather the risk occurs dependent on the ‘function of the thinness-related learning environment present in a given class’ (p. 57). Furthermore, the importance of considering TREs is shown in one study where manipulation of TRE was found to produce greater declines in thinness expectancies, body dissatisfaction, and purging behaviour than psychoeducation manipulation (Annus, Smith, & Masters, 2008).

Perfectionism

Annus and Smith (2009) proposed perfectionism may affect perceptions of messages regarding thinness and weight in dance class. Dancers with elevated perfectionism may perceive a greater emphasis on thinness within their dance class and expect a more positive result from achieving a very low body weight. Within this high-risk dancer group, it is possible that ED symptomatology varies as a function not only of external, environmental pressure for thinness, but also of an internal, individual imposed level of perfectionism (Annus & Smith, 2009). Therefore, examining not only the role of the learning environment, but also the role perfectionism plays in placing dancers at risk for developing ED is important.

Perfectionism has been clearly demonstrated to be a significant risk and maintaining factor for ED (Egan, Wade, & Shafran, 2010). The most widely accepted definition of perfectionism is that it is a multidimensional construct, and is typically measured by either of the two multidimensional perfectionism scales (FMPS; Frost, Marten, Lahart, & Rosenblate, 1990; HMPS; Hewitt & Flett, 1991). Frost et al. (1990) found a 6-factor solution for the FMPS consisting of; personal standards (PS); setting high personal standards; concern over mistakes (CM); self-criticism for making mistakes in performance; doubts about actions (DA); doubting the quality of one's performance; parental expectations (PE); perception that one's parents held high expectations for the individual; parental criticism (PC); perception that one's parents were critical of performance and organisation (O); orderliness and neatness. The 45-item HMPS (Hewitt & Flett, 1991) has three subscales; self-oriented perfectionism (SOP): meeting one's own standards; other-oriented (OOP): belief that others should be perfect, and criticism directed of others when they fail to achieve these standards and socially-prescribed perfectionism (SPP): perceiving others have unrealistically high standards for the individual. In a comprehensive review, Bardone-Cone and colleagues (2007) demonstrated extensive evidence that perfectionism is higher amongst individuals with AN and bulimia nervosa (BN) compared to controls, and that individuals with AN continue to have elevated perfectionism after recovery. Numerous studies have found that the FMPS subscales of PS, CM, DA and HMPS subscales of SOP and SPP are related to ED pathology (Egan et al., 2010). In their transdiagnostic model, Fairburn, Cooper, and Shafran (2003) assert that clinical perfectionism is one of four core mechanisms that maintain ED pathology. Clinical perfectionism involves basing one's self-worth almost exclusively on the determined pursuit of personally demanding standards (Shafran, Cooper, & Fairburn, 2002). For the purpose of this study we were interested in clinically relevant perfectionism, and the three subscales that have been argued to most closely relate to clinical perfectionism are concern over mistakes, personal standards and doubts about actions (Shafran et al., 2002).

Clough and Wilson (1993) found a strong relationship, (r = .69), between scores on the social and psychiatric history form for eating disorders and the multidimensional perfectionism scale (FMPS; Frost et al., 1990) in a sample of ballet and contemporary dance students. However, this study had a small sample and only used the total score on the FMPS rather than subscales. Several researchers have found women who have previously or are currently studying dance at a recreational and elite level score higher on the EDI-perfectionism scale compared with non-dancer controls (Ackard et al., 2004; Anshel, 2004; Neumarker et al., 2000), competitive swimmers (Brooks-Gunn, Burrow, & Warren, 1988) and gymnasts (de Bruin, Bakker, & Oudejans, 2009). Anshel (2004) compared eating attitudes and behaviours on the EDI in 58 full time ballet students to non-dancers in order to determine factors that predisposed ballet dancers to preoccupation with body shape and perfectionism. Dancers scored higher than non-dancers on perfectionism and a significantly strong relationship was found between drive for thinness and perfectionism among dancers and a weak relationship between drive for thinness and perfectionism among non-dancers.

Thomas et al. (2005) found that increased level of dance participation and competitiveness was related to increased level of dieting and self-induced vomiting in dancers involved in dance at a local, regional and national level. Interestingly, however, on the eating disorder inventory (EDI; Garner, Olmsted, & Polivy, 1983), drive for thinness and perfectionism subscales, local students scored significantly higher than regional students. This finding was unexpected considering that local students are generally less technically advanced, have fewer performing opportunities, and ostensibly hold fewer professional aspirations than those at either the national or regional level. Thomas et al. (2005) proposed that drive for thinness varies as a function not only of external, dance school-imposed pressure for thinness, but also of an internal, student-imposed level of perfectionism. The authors concluded that dancers, who exhibit high levels of perfectionism and, perhaps consequently, place themselves in highly competitive environments, may exhibit a significantly increased risk for disordered eating in comparison to dancers who have lower perfectionism.

Rationale

It is clear that perfectionism, TRL and TRE in dance class are important factors in predicting ED symptomatology in dancers. The rationale for the study was to extend on the literature by exploring how individual differences in perfectionism may directly affect perceptions of TRL in dance class in a sample of women who are currently involved in dance training. This is important, as understanding the contribution of these factors may suggest targets for prevention and treatment of ED in dancers. The aim of this study was to determine in a sample of women who were currently involved in dance the relative contribution of perfectionism, TRL and TRE in prediction of ED symptomatology. It was hypothesised that TRL and TRE will partially mediate the relationship between perfectionism and ED symptomatology.

Method

Participants

The participants were 142 female dancers whose primary training was ballet and jazz. On average, participants were 22.33 years old (SD = 3.66, range: 18–30) with a BMI of 20.9 (SD = 3.05, range: 15.6–32.9), BMI was determined via self-report. In addition, 22.5% of the sample may be classified as underweight, with a BMI under18.5 kg/m2. Most participants were Caucasian (93%). A large proportion of participants had a dance related profession, which included being professional dancers and dance teachers (34.5%) and 36.6% were university students. Eighty per cent of participants identified themselves as having an average SES while 8.5% identified with low SES and 11.2% with high SES. Thus the sample did not only involve elite level ballet dancers who are known to have higher rates of ED than controls (Bettle et al., 1998; Dotti et al., 2002; LeGrange et al., 1994; Neumarker et al., 1998, 2000; Pierce & Daleng, 1998; Ringham et al., 2006) as there is also evidence that dysfunctional eating attitudes are found in dancers at a less competitive, non-professional level (Ravaldi et al., 2003).

Measures

Multidimensional perfectionism scale (FMPS; Frost et al., 1990)

The latent variable perfectionism was represented by a mean score of the personal standards (PS), concern over mistakes (CM) and doubts about actions (DA) subscales of the FMPS. These subscales have the strongest association to ED symptomatology (Minarik & Ahrens, 1996), and are closest to the definition of clinical perfectionism (Shafran et al., 2002). PS (7 items): measures striving for high PS; CM (9 items): measures negative reactions to mistakes; and DA (4 items): measures doubting one's own performance. Items are answered on a 5 point Likert-response scale ranging from ‘strongly disagree’ to ‘strongly agree’. The FMPS has been reported to have good reliability and validity. The internal consistency of the subscales used in this study are acceptable; PS, alpha = .83, CM, alpha = .88 and DA, alpha = .77 (Frost et al., 1990). The FMPS has also been shown to have good construct validity having a strong correlation with other perfectionism measures. Internal consistency in the current study was excellent (Cronbach's alpha: PS = .97, CM = .97, DA = .92).

Dance experience questionnaire (DEQ; Annus & Smith, 2009)

The DEQ was used to assess TRL in dance class, dance style (question 1) and level of dance participation (question 7). TRL was calculated by summing questions 7–15 that dance class participants had been involved in. Separate TRL scores were calculated for participants who had experiences in different dance styles and level of dance participation. In the current study the measure was divided into three parcels to represent the latent variable TRL. These parcels were created based on rational clustering of item content to ensure that the parcels were conceptually interchangeable with one another and thus equally representative of the latent construct following Kline's (2005) recommendations. Parcel scores were the mean of the items for each parcel. The TRL subscale has good internal consistency (alpha = .91, Annus & Smith, 2009). Internal consistency in the current study was acceptable (Cronbach's alpha: TRL1 = .79, TRL2 = .96, TRL3 = .91).

Thinness and restricting expectancy inventory (TREI; Hohlstein et al., 1998)

TRE was measured on the 44 item TREI which uses a 7-point Likert scale ranging from 1 (completely disagree) to 5 (completely agree) to assess expectancy for life improvement from thinness and dieting. The TREI is able to differentiate between BN patients, AN patients and normal/psychiatric controls with 94% accuracy (Hohlstein et al., 1998), and its factor structure has been replicated in adolescents (MacBrayer, Smith, McCarthy, Demos, & Simmons, 2001; Simmons, Smith, & Hill, 2002). Internal consistency in the current study was excellent (Cronbach's alpha: TRE1 = .98, TRE2 = .98, TRE3 = .98, TRE4 = .98). The latent variable TRE was represented by four parcels created based on rational clustering of item content to ensure that the parcels were conceptually interchangeable with one another and thus equally representative of the latent construct. Parcel scores were the mean of the items for each parcel.

Eating disorder examination-questionnaire (EDE-Q; Fairburn & Beglin, 1994)

The latent variable eating disorder symptoms were represented by mean scores of the restraint, shape concern, and eating concern subscales of the 38-item EDE-Q, which has evidence of good reliability and validity (Elder et al., 2006). Internal consistency in the current study was excellent (Cronbach's alpha: restraint = .95; shape concern = .96; eating concern = .97).

Procedure

Ethics approval for the study was granted from the Human Research Ethics Committee at Curtin University. On-line recruitment of participants and collection of data provided the opportunity for greater accessibility to participants. Participants were recruited by requesting local dance school teachers, principals and choreographers to forward an advertisement onto appropriate dance students through their email list. The inclusion criteria were being a female aged between 18 and 30 years, currently studying classical and/or jazz ballet. Advertisements were also posted in the Ausdance e-newsletter and via social networking websites. The advertisement included information about the study and invited interested parties to click on a hyperlink that directed them to the participant information sheet, consent form and survey. The survey took approximately 30 minutes to complete.

Results

Dance experience of the sample

The mean age for starting dance tuition was 6.92 (SD = 22, range: 2–20) and mean number of years studying dance was 14.5 (SD = 5.90, range: 2–30). The sample had a mean of 41.35 weeks of tuition undertaken per year and mean of 9.75 hours of dance classes attended per week (SD = 10.82, range: 1–44). Twenty-nine participants (20.4%) reported that they had danced semi professionally at one time, and 34 participants (23.9%) reported that they had danced at a professional level at one time.

Descriptive statistics

Means, standard deviations, and correlations can be seen in Table 1. All measures were significantly correlated at a p < .001 level.

Table 1. Correlations, means and standard deviations for thinness related expectancy, thinness related learning and eating disorder inventory subscales
MeasureMSD12345678910111213
  1. Note: TRL, thinness related learning parcels; TRE, thinness/restricting expectancies parcels; RES, restraint; EC, eating concern; SC, shape concern; PS, personal standards; CM, concern over mistakes; DA, doubts about actions. All measures were significantly correlated at a p < .001 level.

1. TRL12.411.40            
2. TRL22.241.340.88           
3. TRL32.641.900.920.83          
4. TRE13.741.990.710.630.72         
5. TRE23.772.090.690.620.690.96        
6. TRE33.802.130.690.610.700.960.96       
7. TRE43.772.100.700.620.700.960.970.97      
8. RES2.302.090.740.700.740.840.810.830.82     
9. EC2.312.130.720.680.740.850.840.860.850.91    
10. SC2.852.110.700.630.740.850.850.870.870.910.92   
11. PS3.201.390.620.590.630.810.770.790.780.750.760.84  
12. CM2.961.330.650.600.650.880.850.870.850.790.800.840.88 
13. DA2.901.350.670.630.670.860.850.850.850.790.810.840.860.91

Data analysis

Structural equation modelling (SEM) with latent variables was used to test the proposed model. LISREL 8.8 (Joreskog & Sorbom, 2007) was used to estimate parameters for the measurement models (via confirmatory factor analyses) and for the later simultaneous estimation of measurement and structural models. Parcelling was used to represent latent variables as this method has several advantages in the modelling of latent variables relative to the use of individual items. Parcels are likely to have a stronger relationship to the latent variable, are less likely to be affected by method effects, and are more likely to meet assumptions of normality (Marsh, Hau, Balla, & Grayson, 1998). A missing values analysis determined that less than 5% of the data was missing at random within the sample and was replaced with mean substitution. Data screening indicated data non-normality, both at the univariate and the multivariate level, therefore, data was normalised with PRELIS. Covariance matrices were analysed. Maximum likelihood was the estimation method used. Several fit indices were used to evaluate the models: chi-square test and other fit indexes less sensitive to sample size, including chi-square: degrees-of-freedom ratio, with values of 2.0 or less indicating acceptable fit (Kine, 2005); the comparative fit index (CFI) and goodness-of-fit index (GFI), with values of .90 or above indicating acceptable fit (Benet-Martnez & Karakitapoglu-Aygun, 2003; Marsh, Hau, & Wen, 2004) and the root-mean-square error of approximation (RMSEA), with values of .06 or below indicating acceptable fit (Benet-Martnez & Karakitapoglu-Aygun, 2003). Chi-square difference tests were used to compare nested models.

Measurement model

The measurement model for latent constructs with multiple indicators (the TRL, TRE, ED symptoms and perfectionism) was evaluated with a single group, four-factor confirmatory factor analysis (n = 142). The model was specified so that indicators loaded uniquely on the latent factor they represented. The four factors were permitted to correlate with one another. Factor metrics in these analyses were set by constraining all factor variances to unity (Anderson & Gerbing, 1988). Although the model was rejected (χ2 [59, N = 142] = 95.43, p = .001), other fit indices less influenced by sample size indicated adequate model fit (RMSEA = .06, CFI = .99, GFI = .99 χ2/df = 1.62). The viability of the measurement model is further confirmed by the fact that all indicators significantly (p < .001) loaded highly on their respective latent factors, with completely standardised factor loadings ranging from .89 to .97 for TRL, .98 for the TRE, .94 to .96 for ED symptoms and .91 to .96 for perfectionism.

Structural equation model

The structural model consisted of one exogenous (i.e. independent) factor (perfectionism) and three endogenous (i.e. dependent) factors (TRL, TRE and ED symptoms). The hypothesised causal paths between exogenous and endogenous variables were estimated as path coefficients. Five nested models were compared through model trimming. This begins with an assessment of a saturated structural model (see Figure 1 and Table 2) and then simplifies it by eliminating paths. This is done by specifying that at least one path previously free estimated is now constrained to equal 0. As a model is trimmed, its overall fit to the data typically becomes poorer (e.g. χ2 increases). However, if the χ2 goodness of fit statistics for the two models are not significantly different (χ2 difference is not statistically significant), then we would conclude that the two models fit the data equally well and opt for the more parsimonious model, however, if they are statistical significant the path should be retained in the model. The goal of model trimming is to find a parsimonious model that still fits the data reasonably well. Fit statistics appear in Table 2.

Figure 1.

Saturated structural equation model to explain eating disorder risk in dancers. Large circles are latent constructs, rectangles are observed or measured variables, and arrows pointing to observed variables are residual or error variances. Parameter estimates are standardised. All parameter estimates are significant (p < .05). Note: PS, personal standards; CM, concern over mistakes; DA, doubts about actions; P, perfectionism; TRL, thinness related learning; TRE, thinness/restricting expectancies; ED, eating disorder symptoms; RES, restraint; EC, eating concern; SC, shape concern

Table 2. Goodness of fit indices and chi-squared difference values for structural equation models
Modelχ2dfpχ2/dfGFINFIRMSEACFIAIC
  1. Note: TRL, thinness related learning; TRE, thinness/restricting expectancies; ED, eating disorder symptoms; P, perfectionism.

Saturated model95.4359.0011.62.910.99.06.99182.00
1a. TRL to TRE path trimmed104.4360.001.73.90.99.07.99164.81
χ2 difference91.00      
1b. P to TRE path trimmed219.6360.003.66.84.97.12.98238.27
χ2 difference124.21.00      
2a. TRE to ED path trimmed113.8260.001.90.89.98.08.99173.27
χ2 difference18.391.00      
2b. TRL to ED path trimmed110.3760.001.84.89.98.08.99172.63
χ2 difference14.941.00      
3. P to TRE path trimmed219.6360.003.66.84.97.12.98238.27
χ2 difference124.21.00      

Firstly, TRL was tested as a partial mediator of the relationship between perfectionism and TRE by eliminating (constrained to zero) the path between TRL and TRE. The resulting chi-square was significant thereby indicating the path between TRL and TRE should be included in the model and confirming that TRL is a partial mediator (see Table 2, model 1a). The direct path between perfectionism and TRE was eliminated (constrained to zero) in favour of an indirect association (fully mediated model) between perfectionism and TRE through TRL. Chi-square tests of differences between models resulted in a significant decrease in model fit (p < .001) when moving from the partially to full mediated model (see Table 2, model 1b). This result further confirms our hypothesis that a direct relationship between perfectionism and TRE is required in the model therefore further confirming TRL only partially mediates the relationship between perfectionism and TRE.

Secondly, TRE was tested as a partial mediator of the relationship between TRL and ED symptoms by eliminating (constrained to zero) the path between TRE and ED. The resulting chi-square was significant thereby indicating the path between TRE and ED should be included in the model and confirming that TRL is a partial mediator (see Table 2, model 2a). The direct path between TRL and ED was eliminated (constrained to zero) in favour of an indirect association (fully mediated model) between TRL and ED through TRE. Chi-square tests of differences between models resulted in a significant decrease in model fit (p < .001) when moving from the partially to full mediated model (see Table 2, model 2b). This result further confirms our hypothesis that a direct relationship between TRL and ED is required in the model therefore further confirming TRE only partially mediates the relationship between TRL and ED.

Finally, both TRL and TRE were tested as a partial mediator of the relationship between perfectionism and ED by eliminating (constrained to zero) the path between perfectionism and ED. The resulting chi-square was significant thereby indicating the path between TRE and ED should be included in the model and confirming that TRL and TRE are partial mediators of the relationship between perfectionism and ED (see Table 2, model 3). The results indicate that the all paths in the saturated model will be retained (see Figure 1) to explain the ED symptomatology in dancers.

Discussion

The purpose of the research was to test a model that evaluated the involvement of perfectionism, TRL and TRE with ED symptomatology in dancers. Results supported the hypothesis that the relation between perfectionism and ED symptoms was partially mediated by TRL. Dancers who reported higher perfectionism perceived greater TRL in their dance classes compared to dancers with lower perfectionism. It may be that individuals with higher perfectionism perceive greater TRL in their dance classes irrespective of the actual amount of learning in regards to thinness in their environment. In converse, dancers may develop higher perfectionism as a result of dance environments that emphasise greater thinness. Another potential hypothesis is that individuals higher on perfectionism are more likely to be successful in dance classes and obtain higher levels of achievement in dance, and participation in more rigorous dance training may lead to individuals being exposed to more learning regarding the importance of thinness. It was not possible in this study to identify the exact classes and environments that the dancers had studied in, thus they had studied in a range of different environments. A useful question for future research would be to examine students in the same dance school to determine if greater levels of perfectionism are related to perceptions of more messages regarding the importance of thinness within the same dance classes, as it may be assumed that individuals in the same dance class are exposed to the same messages regarding thinness.

Dancers who experienced greater TRL were also more likely to have higher TREs. Dancers higher on perfectionism were also more likely to directly experience greater TREs. When all of these relations are considered as a whole through the use of structural equations analyses, the partial mediation role of TRL was supported.

The hypothesis that the relation between TRL and ED symptoms were partially mediated by TREs was also supported. Dancers who reported higher TRL in their dance classes experienced more ED symptoms. This finding is consistent with ED expectancy theory, that dance class related learning experiences about the benefits of thinness and dieting contribute to the formation of reinforcement expectancies regarding thinness and dieting, and the expectancies are the proximal influence on symptomatic behaviour (Annus & Smith, 2009).

A direct relationship between perfectionism and ED symptoms was found where dancers with higher perfectionism experienced more ED symptoms, which supports the very well established link between perfectionism and ED (Bardone-Cone et al., 2007). This association between perfectionism and ED has been consistently found (e.g. Ackard et al., 2004; Anshel, 2004; Neumarker et al., 2000; Thomas et al., 2005). Given the importance of perfectionism as a risk factor for ED, it would be useful for future research to examine the efficacy of targeting young women involved in dance with elevated levels of perfectionism for prevention programs, as preliminary findings are positive for the ability of programs targeting perfectionism to reduce ED risk in female adolescents (Wilksch, Durbridge, & Wade, 2008). This is particularly important, as it appears that women with higher levels of perfectionism are more susceptible to developing expectancies for the benefit of thinness and restricting based on learning in regards to thinness in dance class. In summary, the findings suggest that women with elevated perfectionism are at a particular risk when participating in dance class in regards to developing expectations of the benefits of restriction of food as a result of learning in regards to thinness that is emphasised in these environments.

There are limitations to the study. The data were cross-sectional, thus longitudinal research needs to be conducted as no causal relationships can be concluded from the present results. Although the current model fit the data well, there are multiple other risk factors for ED and it is very likely that this risk is determined by multiple factors and is not simply a function of the emphasis on thinness in the dance environment, or perfectionism. Future research is required to examine other models that incorporate additional risk factors. Furthermore, our sample was too small to split into professional and non-professional dancers to examine if the model was the same in these groups, and it would be useful to examine this in future research as it is possible that factors could differ across these groups. Another useful direction for future research also would be to consider exploring and observing the behaviour of educators and the possible effect of actual learning experiences in a dance class. A final limitation of the study is that our view of perfectionism was aligned with the construct of clinical perfectionism, and as a result we examined a selected subset of scales from the FMPS and did not use relevant scales from the HMPS. Future research may consider doing this to examine the model further to see if there is a different relationship with learning experiences and the other subscales of perfectionism.

Notwithstanding these limitations, the findings offer support for previous research that has considered perfectionism to be a critical antecedent of ED risk in dancers. The current study extended on previous studies as it included a wider and ostensibly more representative cross-section of the population of dancers as most of the literature to date consisted of single-school studies with relatively small samples sizes. Furthermore, we used a sample of women who are currently involved in dance, thus our study did not suffer the problems inherent with other studies such as Annus and Smith (2009) who relied on retrospective recall. Our findings suggest that future research should examine if delivering programs to reduce perfectionism in young women involved in dance with is useful in preventing ED, as they are more likely to be susceptible to negative learning experiences regarding thinness. In regards to practical implications for clinicians, our results suggest that it would be useful for clinicians to consider directly targeting perfectionism through CBT programs for perfectionism (e.g. Shafran, Egan, & Wade, 2010) in young women involved in dance classes as it is likely to be a factor that puts them at higher risk of developing an ED within this environment. The results are a direction for future research to investigate ways to ameliorate the salient factor of perfectionism in this high risk population.

Ancillary