Self-Construction, Cognitive Conflicts, and Disordered Eating Attitudes in Young Women

The aim of this study is to identify cognitive variables that predict disordered eating attitudes in a nonclinical sample composed of 50 female university students. Repertory grid technique was used to assess cognitive features of self-construing and cognitive conflicts. Drive for Thinness and Body Dissatisfaction scales from the Eating Disorder Inventory–2 were used as dependent variables, as previous studies suggested that high scores on these scales are associated with the risk of developing or aggravating eating syndromes. Results suggest that drive for thinness can be associated with cognitive conflicts, whereas body dissatisfaction may be higher for those who construct themselves as inadequate and similar to others. In addition, both dependent variables were predicted by being younger and having a higher body mass index.

Eating disturbances are serious health problems that have generated social alarm in recent decades. Epidemiological studies indicate that the incidence of eating disorders (EDs) has progressively increased. In Spain, the estimated prevalence rates range between 1% and 4.5% for bulimia, 0.5% to 1% for anorexia, and 1% to 4.5% for EDs not otherwise specified in the general population (Toro, 2003). Considering only women, prevalence for EDs ranges between 4% and 17.9%. Even if the emergence of symptoms is usually related to adolescence, the percentage of young women who suffer them has increased. Additionally, several studies have shown the occurrence of these problems in women who Self,Cognitive Conflicts,and Eating Attitudes 71 are older than adolescents (e.g., Kugu, Akyuz, Dogan, Ersan, & Izgic, 2006). Considering that young women face many demands related to their entrance to adulthood, EDs pose risks to their quality of life and generate serious social, family, and professional maladjustment. According to one recent study published by the Spanish Ministry of Health and Consumers Affairs (Ministerio de Sanidad y Consumo, 2008), nearly 70% of Spanish women who are of normal weight declare feelings of dissatisfaction with their body.
In addition, many eating behaviors and dysfunctional attitudes have been observed, although they do not strictly fulfill the diagnostic criteria for either anorexia or bulimia (Sancho, Arija, Aosrey, & Canals, 2007). The need to offer proper care to all people suffering from eating problems is clear. Negative attitudes toward one's body and excessive worry about being fat are sources of distress for women, even when they do not reach levels of clinical significance.
Today, health professionals practice preventive intervention in partial, subclinical, and not-specified syndromes, which could evolve in insidious ways from very early stages. Such subclinical eating disturbances have increased alarmingly in the last few decades (12% for restrictive eating behaviors, 6% for overeating behaviors) and affected not only adolescents but also young women (Sancho et al., 2007). This increase has generated the urgent need for early detection and the study of those factors that could predict various types of EDs.
Low self-esteem is one of the characteristics more commonly associated with EDs (Jacobi, Paul, de Zwaan, Nutzinger, & Dahme, 2004;Slade, 1982). In fact, several studies indicate that self-esteem also differentiates people who suffer EDs not only from normal controls but also from people who diet to lose weight (Gismero, 2001;Williams et al., 1993). Watson and Watts (2001) indicated that one's sense of personal value is based on the similarity between the perception of one's "present self" and "ideal self" in terms of the attributes important to the person. Based in Higgins's (1987) theory of self-discrepancies, Strauman, Vookles, Berenstein, Chaiken, and Higgins (1991) studied the relevance of selfdiscrepancy in relation to body dissatisfaction and dysfunctional eating behaviors in college students, finding the discrepancy between the two self-images to be correlated to body dissatisfaction (independently of the participants' body mass index [BMI] and the number of beliefs about the self related to appearance). Other authors found the discrepancy between present and ideal selves was greater in patients with anorexia (Fransella & Crisp, 1979), with bulimia (Feixas, Montebruno, Dada, del Castillo, & Compañ, 2010;Izu, 2007), and in women with subclinical restrictive behaviors (Neimeyer & Khouzam, 1985), compared to women without eating problems.
Besides dissatisfaction with oneself, unpleasant social relationships have been widely related to EDs in both literature and research. Generally, social maladjustment and fear of others' negative evaluation have been linked to a lack of self-confidence, poor self-esteem, and perceived inadequacy of the self (Fairburn, Welch, Doll, Davies, & O'Connor, 1997;Striegel-Moore, Silberstein, & Rodin, 1993). Less attention has been paid to how others are perceived by people affected by EDs (Quiles, Terol, & Quiles, 2003). However, previous studies have shown differences in perceived support from family and peers between ED patients and controls. According to Tiller and colleagues (1997), bulimic patients showed disturbances in perceived adequacy of social relationships, and Grissett and Norvell (1992) have suggested that bulimics perceived less social support from friends and family, and were less socially competent. Cuhna, Relvas, and Soares (2009) found that the group of young women with anorexia nervosa showed more alienation and detachment from mothers, fathers, and peers when compared to a control group.
Within personal construct theory (PCT), the construction of self is intrinsically related to the construction of others (Jackson & Bannister, 1985). Greater differentiation between self and others, taken as an index of perceived social isolation, has been associated with other psychological problems, such as depression (Neimeyer, 1985), and with the negative consequences of sexual abuse (Harter, Neimeyer, & Alexander, 1988). Feixas et al. (2010) suggested that women with longer histories of bulimia nervosa show greater perceived social isolation.
Other studies with ED patients showed that their system of constructs was more restricted-or, in other words, less multidimensional-compared to nonclinical samples (Button, 1983(Button, , 1993. This suggests that these patients use just a few constructs for interpreting their interpersonal relationships. Inspired by PCT, our research group has promoted a line of research focused on the view of some psychological symptoms as personal dilemmas. Methodologically, it is based on the study of cognitive conflicts using the repertory grid technique (RGT; see Feixas, Saúl, &Ávila-Espada, 2009), which has already been applied to several clinical problems (Benasayag, Feixas, Mearin, Saúl, & Laso, 2004;Carretero, Feixas, Pellugrini, & Saúl, 2001;Compañ et al., 2011;Feixas, Cipriano, & Varlotta-Domínguez, 2007;Feixas et al., 2010;Feixas et al., 2009;Montesano, Feixas, & Varlotta, 2009) and has led us to the elaboration of a treatment manual (Feixas & Saúl, 2005;Fernandes, Senra, & Feixas, 2009;Senra, Feixas, & Fernandes, 2005). In our proposal we consider two types of cognitive conflicts that might reflect personal dilemma: dilemmatic constructs and implicative dilemmas. For the former, we take Kelly's (1955) view of constructs as unique and personal distinctions derived from experience. On these bipolar dimensions of meaning, the person often gives different values to each pole of the construct and orients his or her behavior by choosing the desirable alternative. Using the RGT, the desirable pole of each construct can be identified by looking at the scores given to the "ideal self." According to Feixas and Saul (2004), a conflict could arise when both poles of the construct are deemed undesirable by the respondent, and neither offers a clear-cut course of action. These constructs could be considered dilemmatic if they promote ambivalence because both poles offer both advantages and disadvantages. Although it is common for people to have a few of these constructs, a large proportion of dilemmatic constructs in a grid may indicate that the person's constructions are not useful for choosing which direction to take. This situation may result in feelings of insecurity and hesitation, and action could be blocked as a tentative avoidance of the disadvantages perceived in both of the alternatives. However, as Winter, Bell, and Watson (2010) suggested, the placement of the ideal self in a midpoint of a construct might have different interpretations, some of them not necessarily related to conflict.
Another type of conflict arises due to a peculiar association between constructs, according to which acquiring the desirable characteristics would imply giving up those that the person would like to keep. An implicative dilemma can be understood as a strong entailment between a construct in which the person wishes to change (discrepant construct) and another construct in which changing is not desirable (congruent construct). Because of this particular association, giving in to the undesired characteristics could imply a threat for the system and for identity .
For example, if a person changes position in the problematic (discrepant) construct, moving from the actual pole ("shy") to the desired pole ("sociable"), this change would imply, according to his or her point of view, a similar change in the congruent construct ("good person"); thus, to avoid an undesired change (becoming "selfish"), the person also avoids the desired change (see Figure 1). Therefore, renouncing the symptom could carry negative implications for the person's self-construction. Feixas et al. (2010) explored the cognitive indicators derived from the RGT in a sample of young women with bulimia nervosa (BN) compared to a group of asymptomatic, age-matched women. The authors found that the presence of implicative dilemmas was more frequent in BN patients' RGT than in controls' RGT. These data are confirmed by the results obtained by Izu (2007) with another sample of bulimic women of the same age. As far as we know, there are no previous studies focusing on the role of cognitive conflicts in terms of dilemmatic construing and its relation to disordered eating attitudes in nonclinical samples.
The aim of this study is to explore the relationship between patterns of construing self and others (including cognitive conflicts) and eating problems in women from a nonclinical sample,

Good person Selfish Shy Sociable
Present self /Ideal self Present self Ideal self r = .65 r = .65

FIGURE 1
Example of implicative dilemma.

75
and to study if those variables are useful for predicting eating attitudes, such as drive for thinness and body dissatisfaction. Previous studies have suggested that these attitude scales are associated with the risk of developing or worsening partial eating syndromes, or even full EDs (Cooley & Toray, 2001;Killen et al., 1996;Leon, Fulkerson, Perry, & Early-Zald, 1995).

Participants
The sample was composed of 50 women, undergraduate or graduate students, between the ages of 18 and 45 years (M = 25.14, SD = 5.58).

Instruments and Measures
The EDI-2 (Garner, 1991; Spanish adaptation by Corral, González, Pereña, & Seisdedos, 1998) was used for the evaluation of characteristics and symptoms considered relevant in EDs. It consists of 91 items forming 11 scales. Only the scales evaluating attitudes related to food, weight, and figure (Drive for Thinness [DT] and Body Dissatisfaction [BD]) were used for the purposes of this study. This inventory also collects age, weight, and height data. REPERTORY GRID TECHNIQUE (RGT) In the grid format used in this study, the interviewer elicited elements by asking for significant others: parents (or other significant relatives), siblings, present partner (and past significant partners), and close friends. Participants varied in the number of elements provided under these roles. Also, persona non grata, self, and ideal self elements were included in the grid. The number of elements ranged between 12 and 25 (M = 16.02, SD = 3.32) across participants. These elements were considered in dyads to elicit constructs by describing similarities and differences between them. For each similarity or difference described, an opposite construct pole was also elicited. This process continued until the "saturation point," when participants were unable to generate additional constructs (i.e., when they began to repeat, did not find new constructs, or showed obvious signs of fatigue). The number of constructs elicited from each participant ranged from 17 to 50 (M = 33.34, SD = 8.87). Following elicitation of constructs, the interviewer asked participants to rate all of the elements on each construct using a Likert scale format: 1 means very of the emergent pole (i.e., very brave) and 7 means very of the implicit pole (i.e., very cowardly). A series of indexes and measures were computed for each grid.
Self-Ideal discrepancy. This measure is obtained from the Chisquared distance between the present and ideal self elements. High distances are usually related to low self-esteem.
Perceived social isolation. This index is calculated using the product-moment correlation between the scores of the present self and the mean scores of the other elements included in the grid. It reflects the degree of similarity to others perceived by the respondent. A z transformation was performed in order to convert the nonlinear correlation into a linear variable.
Percentage of implicative dilemmas (PID). The identification of implicative dilemmas (Feixas, Saúl, & Sánchez, 2000;Feixas et al., 2009) is based on the relations between congruent constructs (those in which there are no differences between the present and ideal self elements) and discrepant constructs (those in which there are relevant differences between the two). Implicative dilemmas were considered to exist if the Pearson's correlation between the scores of congruent and discrepant constructs is positive and equal to or bigger than 0.35. That is, the desirable pole of the discrepant construct is correlated to the undesirable pole of the congruent construct. We use the PID in order to take into account the size of the grid; it is calculated by dividing the number of ID by the total number of possible combinations of constructs if taken two at a time, and then multiplied by 100 .

Percentage of dilemmatic constructs (PDC).
Those constructs for which the ideal self received a midpoint rating (4) were taken as dilemmatic constructs, following Feixas and Saul's (2004) suggestion. PDC is calculated simply by dividing the number of dilemmatic constructs by the total number of constructs in the grid, and then multiplying by 100.

Procedure
Students were recruited after asking for their voluntary participation in their classes. A date was fixed for an individual assessment session, which usually lasted about 2 hr. The interviewers were doctoral students with extensive training in repertory grid administration. The RGT was administered using a pencil-and-paper format and analyzed using the GRIDCOR 4.0 software (Feixas & Cornejo, 2002). SPSS version 15.0 was used to perform statistical analyses. Pearson's correlations were calculated between EDI-2 scales, age, BMI, and RGT variables. The sample was divided into three groups, based on percentile scores in the EDI-2 attitude scales (DT and BT), as standards for Spanish adult women without EDs were not available. Values beneath the second quartile were considered low (34 participants for DT and 28 for BD), whereas those above the third quartile (13 participants for DT and 16 for BD) were taken as high values. To focus on the comparison between subjects of a potential risk group (high values in the scales of the EDI-2) and those without dysfunctional eating attitudes, participants falling within the second and third quartiles (three participants for DT and 6 for BD) were excluded from further analyses.
A logistic regression with forward variable selection was performed to predict high values on DT and BD, introducing BMI, age, self-ideal discrepancy, perceived social isolation, number of constructs, proportion of dilemmatic constructs, and proportion of implicative dilemmas as the independent variables. In order to avoid the effect derived from size differences between both groups (given the low prevalence in the group with a high level of symptoms), the probability cutoff points were adjusted, maximizing the false positive to false negative ratio (see below). Finally, receiver operating characteristic curves were constructed to test the accuracy of the regression models in correctly predicting high values on DT and BD scales.

Sample Description
The BMI of the participants oscillated between 16.69 and 30.38, with a mean of 21.7 (SD = 3.18). According to the international cutoff points indicated by the World Health Organization (1995,2000), normal BMI was displayed by 82% of the sample, and no cases of severe thinness (BMI < 16.00) or obesity class II/III (BMI > 35.00) were found.
The mean scores in EDI-2 scales were low and similar to those expected for female graduate students, according to American standards. However, as can be observed in Table 1, we found 4% of scores on DT and 8% of scores on BD above the mean from the Spanish clinical sample. Descriptive statistics for the grid measures are shown in Table 2.

Logistic Regression for Drive for Thinness
The overall regression model for DT (see Table 3), contained four variables: proportion of dilemmatic constructs, percentage of implicative dilemmas, BMI, and age. The full model was statistically significant, c 2 (4) = 25.78, p < .001, indicating that it was able to distinguish between participants presenting low or high scores on this scale. Age and BMI were the first two variables that entered into the equation by the logistic regression procedure used. Results indicate that younger and heavier women were more likely to show  higher scores on DT. Nevertheless, considering only those variables as predictors, 38.5% of false negatives were obtained (see Table 4). The whole model resulting at Step 4 included proportion of implicative dilemmas and percentage of dilemmatic constructs. This model showed pseudo R squared indexes of 42.2% (Cox and Snell R squared) and 61% (Nagelkerke R squared) for presenting low or high DT values, and correctly classified 91.5% of participants. False negatives were reduced from 38.5% in Step 1 to 7.7% in Step 4.

Logistic Regression for Body Dissatisfaction
The overall regression model for BD (see Table 5) also contained four variables: self-ideal discrepancy, perceived social isolation, BMI, and age. The full model was statistically significant, c 2 (4) = 29.64, p < .001, indicating that it was able to distinguish between subjects presenting low or high scores in BD. As for DT, age and BMI were the first variables to be introduced in the regression model, indicating that the probability of obtaining high scores on BD increased as BMI was higher and age was lower (see Table 5).
Using only these two variables as predictors, 100% of participants with high values were correctly classified, but the model was imprecise when classifying participants with low BD. In this sample, 53.6% of false positives were found when using only BMI and age as predictors (see Table 6). The whole model showed that an increase of BMI, self-ideal discrepancy, self-others correlation, and younger age increase the probability of showing high BD values. This model showed pseudo R squared indexes between 49% (Cox and Snell R squared) and 67.1% (Nagelkerke R squared), and correctly classified 81.8% of participants (71.4% of low scores and 100% of high scores).

Regression Model Accuracy
For DT and BD, the areas under the curve (AUC) equaled .93 (SE = .04, p < .001), with 95% confidence interval = .85-1.00, p < .001. The AUCs indicated that both regression models were accurate, and that prediction was significantly different from 82 G. Dada et al.

Discussion
Eating symptoms have been commonly associated with adolescence, and it seems clear that heavier people tend to be more worried about their body shape; therefore, it is not surprising to find strong negative correlations between age and attitude scales, in addition to positive correlations between BD and BMI. Moreover, lower age and higher BMI were included as predictors of high values on the EDI-2 Attitude scales at the first steps of the analyses. Nevertheless, regression models including only BMI and age as predictors were less accurate than the complete models including cognitive conflicts (for DT) or self-construction (for BD) variables. Specifically, if the variables regarding cognitive and self construction were not taken into account, almost 40% of women from this sample seeking thinness would be wrongly classified as if they showed low values in DT (38.5% of false negatives), and more than 50% of women who are actually satisfied with their bodies would be classified as dissatisfied (53.6% of false positives).
Previous research within PCT (Button, 1983(Button, , 1993 has shown that ED patients used fewer constructs than controls. In this study DT was correlated with providing fewer constructs, suggesting that a lower number of dimensions in the construct system might facilitate concentration on the idea of thinness even for women without an ED diagnosis. Besides this finding, RGT indexes did not show other significant linear correlations with DT or BD when taken one by one. However, by differentiating participants with potential risk from those with healthy eating attitudes, more sophisticated and complex interrelationships among variables were found when performing logistic regressions. The resultant regression models showed that, besides higher BMI and lower age, a larger proportion of dilemmatic constructs and a higher percentage of implicative dilemmas were good predictors for DT. While implicative dilemmas have been formerly associated with several psychological problems , and specifically with eating disorders (Feixas et al., 2010;Izu, 2007), the assumption that the so-called dilemmatic constructs are indicative of cognitive conflict has not been consistently supported by previous research. The use of midpoint ratings for the ideal self element has received diverse interpretations in the personal construct literature (e.g., Winter et al., 2010). Its association to psychopathology or any other forms of malfunctioning is not clear, and it is controversial. The results of this study, however, might provide some support for such an association because they indicate that a higher proportion of midpoint ratings on the ideal self is useful for predicting some risky eating attitudes (DT). Furthermore, in order to illustrate this result, PDC values from both groups were compared, showing that this proportion for the high DT group was almost twice as high (M = 12.5, SD = 7.8) as for the low DT group (M = 7.4, SD = 8.1), a significant difference (U = 133, z = 2.11, p = .035, r = .31). Nevertheless, the support for the association between a high proportion of midpoint ratings for the ideal self obtained for DT was not replicated for BD. Therefore, the debate around the meaning and clinical implications of dilemmatic constructs remains open, and further research should clarify the implications of our findings.
Being younger and heavier, considering oneself as inadequate, and perceiving others as similar were predictors of BD. These findings are consistent with research showing that a negative self-esteem is related to dysfunctional eating attitudes. Compared with traditional measures of self-esteem, the RGT allows for the assessment of the satisfaction with oneself in terms of those attributes that are meaningful for the person (i.e., personal constructs). In addition to the construction of self, it is possible to assess the construction of others and the perceived similarities or differences between self and others. Our results suggest that perceived similarity with others predicts greater BD. The above may seem inconsistent with previous research relating the differentiation between self and others to psychological suffering. But this finding could be understood if the views of those with high BD have a negative view of others, so that this negative view also includes the self, an issue that also deserves further research.

Conclusions
Being younger and having a higher BMI seem to be related to the possibility of developing disordered eating attitudes, but those variables are not accurate enough for distinguishing women highly worried about their body shape from those who are not. This study has shown that cognitive and self-construing variables that can be assessed using the RGT are useful for the prediction of these negative attitudes. Specifically, the impulse to seek thinness (or the fear of being fat) can be related to more dilemmatic ways of construing (in terms of both implicative dilemmas and dilem-matic constructs) and to a system with fewer constructs than that of those without DT. Driving for thinness can thus be associated with a construct system presenting various forms of conflicts in construing self and others so that it does not produce anticipations useful for orientating behavior in social interactions. In this context, thinness would be pursued because it could be perceived as a safe route to success in social relationships.
Previous studies provided strong support for the relation between low self-esteem and disordered eating attitudes. Our findings suggest that being dissatisfied with one's body can be associated with the negative perception not only of one's self but also of others.
There are some limitations that need to be acknowledged and taken into account when considering this study and its contributions. We had a relatively small sample size for a complex subject with multiple variables; additionally, all women included were undergraduate students. Therefore, even though the results were statistically significant, they should be viewed as preliminary. Moreover, the data collected for this study were examined at just one point in time, and, as such, provided only a description of the relationships between variables. Further research with longitudinal studies is necessary to explore whether those associations observed in the present study are maintained over time.
Although requiring further exploration in a large-scale investigation, the findings of this study suggest that personal construction is relevant for the understanding of disordered eating attitudes.