rosenberg self-esteem scale na obesidade mórbida Bifactor analysis of the rosenberg self-esteem scale in morbid obesity

Self-esteem is a set of feelings and thoughts a person has about his/her own worth and competence, which is reflected in a positive or negative attitude toward himself/herself. This study examined the factor structure of the Rosenberg Self-Esteem Scale with scores of 288 obese brazilian women waiting for bariatric surgery. Psychometric studies have found controversial results regarding the factor structure of this measure with samples from different profiles. In this investigation, the unifactor, two-oblique-factor and bifactor models were tested by Confirmatory Factor Analysis. The results clearly support the existence of a single dimension for the scale, but also for the method effects relating to the positively and negatively worded items. Further investigations with the inclusion of obese males are suggested.


INTRODUCTION
Obesity is a multidetermined disease (genetics, diet, physical inactivity, psychological and social) considered a worldwide epidemic. Forecasts show that by 2025 there will be 2.3 billion overweight adults and over 700 million obese people in the world (World Health Organization, 2018). In Brazil, overweight increased by 26% between 2006 and 2016 and obesity increased by 60% in 10 years (Vigitel Brazil, 2017). Obesity has several psychosocial implications for the individual's life, compromising psychological and social health, negatively affecting self-esteem, due to stigma, difficulty accepting body self-image, feeling of failure, inferiority and bullying (Rocha & Costa, 2012).
Many obese people have sought bariatric surgery in order to overcome the problem. This procedure, which involves stomach reduction, is indicated for obese individuals with Body Mass Indexes (BMI) ≥ 40 kg/m² -regardless of associated comorbidities -or, with BMI ≥ 30 kg / m² and associated diseases, who did not have a satisfactory result with other types of treatments (Schakarowski, Padoin, Mottin, & Castro, 2018). Psychology can contribute to preoperative preparation, increasing the success of treatment adherence, improving prognosis and reducing postoperative complications of bariatric surgery (Fagundes, Caregnato, & Silveira, 2016;Pona et al., 2017).

Self-esteem is a set of feelings and thoughts
one has about his/her own value and competence, which is reflected in positive or negative attitudes toward himself/herself. It influences the way a person accepts himself/herself, values the other, sets goals, and projects them into the future. Therefore, it plays an important mediating role in the psychological, social and physical behavior of the individual (Orth, 2017). The concept of self-esteem has been one of the most studied constructs in recent decades, Bleidorn et al. (2016) found 35,000 published articles about it.
One of the most famous measures of selfesteem is the Rosenberg Self-Esteem Scale (RSES), which was developed by Morris Rosenberg (1965). The participants of the study conducted to develop the RSES were university students and adults, of both sexes, from different social environments and ethnic groups in New York.
The original instrument consists of 10 items that represent positive and negative feelings of respect and self-acceptance in a single dimension, 5 of which are positively worded items and 5 negatively worded items (Rosenberg, 1965 More than 50 years of research and hundreds of empirical studies failed to resolve the question concerning the dimensionality of the RSES (Gnambs, Scharl, & Schroeders, 2018). Many studies found a unidimensional structure (Gómez-Lugo et al., 2016;Hutz, 2000;Hutz & Zanon, 2012;Meurer, Luft, Benedetti, & Mazo, 2012). Others supported a structure composed of two correlated factors, sometimes called self-confidence and selfcontempt, sometimes called positive and negative self-esteem (Avanci, Assis, Santos, & Oliveira, 2007;Cárdenas, Cárdenas, Villagrán, & Guzmán, 2015;Ceballos-Ospino et al., 2017;Cogollo, Campo-Arias, & Herazo, 2015;González -Rivera & Pagán-Torres, 2017;Ricardi & Méndez, 2016;Romano, Negreiros, & Martins, 2007;Sbicigo, Bandeira, Dell'Aglio, 2010;Tolentino, Maia, Ant, Sousa, & Melo, 2015;Ventura-Leon, Caycho-Rodriguez, Barboza-Palomino, & Salas, 2018). Another group of researchers also supported a one-factor structure of the RSES, but by controlling method effects In general, method effect refers to the variance that is attributable to the measurement method -in the case of RSES, the negative and positive expression of the items -and not to the construct of interest. The method effect on the results, in terms of sources of variance, may vary in intensity (Maul, 2013). The acquiescence would be a tendency of the participants to provide affirmative answers to items regardless of their content (Valentini, 2017).
However, the presence of these negative items in the instruments also seems to interfere with their latent structure, a type of method effect.
However, other studies have shown that positive and negative items can simultaneously produce effects and, if considered in the tested models, they can result in better fits to the empirical data (Marsh, Scalas, & Nagengast, 2010;Quilty, Oakman, & Risko, 2006). Nowadays, the idea that a single latent factor would be insufficient to adequately describe the responses to the RSES seems to predominate. The scale would show multidimensionality with regard to the wording of items, which could be interpreted as method effects. Although RSES is not strictly unidimensional, secondary dimensions would only have a modest impact on item responses and, therefore, introduce a seemingly small bias in composite scores of the RSES (Gnambs et al., 2018). As a matter of fact, some authors argue that the validity of the global self-esteem factor would hardly be affected even if the wording effects were not controlled (Donnellan et al., 2016).
In this study, the factor structure of the measure was examined using the scores of obese women waiting for bariatric surgery who answered the version of the RSES adapted by Hutz (2000). The analysis observed if there are method effects and if they are linked to the wording of positive items, negative items or both.

Data analysis
The scores of negative items of the RSES were inverted to assess self-esteem in the same direction of the positive items (toward the construct).
Descriptive analyzes were performed to verify the normality of data distribution. Then, a series of structural models for the RSES were tested, which have been frequently applied in the literature

RESULTS
Examination of RSES scores revealed a Mardia coefficient of 15.68 (normalized = 8.58), which indicated the multivariate abnormality of their distribution. However, the univariate distribution of scores showed asymmetry <± 1.5 and kurtosis <± 1.5, which is not considered an extreme violation of normality (Finney & DiStefano, 2006).  (Cohen, 1988). Therefore, it would be unlikely to have a possible acquiescence effect in responding to the instrument.

In the framework of Structural Equation
Modeling, CFAs tested the models of RSES used in previous studies (Lindwall, 2012;Marsh et al., 2010;Quilty et al., 2006). These models adopted the Multitrait-Multimethod (MTMM) conceptual framework, which presents two types to separate substantive content (self-esteem) from method effects. One is the correlated trait correlateduniqueness (CTCU) model and the other is the correlated trait (self-esteem), correlated methods model (CTCM). The CTCU inserts correlations among the measurement errors of positively and negatively worded items, and the CTCM includes specific latent method effect factors underlying the items of the same method (ie, positively or negatively) along with a substantive latent factor (self-esteem). In order to examine the method effects, CTCU and CTCM models were compared to models that do not include method effects to establish if they fit the data better (Byrne, 2016b).  (Rosenberg, 1965). The indexes estimated for this model presented fits between bad and poor (Table 3), according to Marôco's classification (2014). Model 2, two oblique factors, positive and negative self-esteem (Figure 1), presented better results, but also did not show a sufficiently adequate fit to the data (Table 1). The analysis of CTCU-based models (Figure 2 (Table 1). Model 5, including correlations between the errors of positively worded items and correlations between errors of the negatively worded items, was generally the best of all models tested, revealing a very good fit to the data (Table 1). with the global self-esteem factor, fit the data very well (Table 1).   (Guo & Hu, 2015;Landwall et al., 2012;Marsh et al. al., 2010;Quilty et al., 2006).

DISCUSSION
According to Schmitt and Stults (1986), Some studies have found an association between female sex, depressive symptoms, low self-esteem, and responses related to the negatively worded items of the RSES (Lindwall et al., 2012;Urban et al., 2014). This method effect has been classified as a response style.
Although these relations were not tested in the current study, it is possible that the tendency of the method effect linked to the negative items represents a response style of obese women.
An important question about the method effect is that it may merely reflect systematic measurement errors or response styles (DiStefano & Motl, 2006;Lindwall, et al, 2012;Quilty et al., 2006). Response style involves a willingness to interpret and endorse items based on a certain tone or valence (DiStefano & Motl, 2009).
Similar to the studies by Lindwall et al. (2012) and Urban et al., (2014) it is possible that the negative mood of these obese women results in a greater tendency to endorse items with negative words from RSES. Anyway, this hypothesis must be tested in future research.
One of the limitations of this study is that its sample consisted of female participants only.
Unfortunately, the percentage of male participation in the groups where the participants were recruited was very small. Thus, it is suggested that future studies attempt to test the invariance of the RSES in a larger sample of obese Brazilian men and women.