Introduction

The causes and maintenance of eating disorders tend to be heterogeneous and multifactorial [1, 2]. Body dissatisfaction, negative affect, perfectionism, impulsivity, substance use, childhood maltreatment, unhelpful family interaction that results in high expressed emotion are some of the risk factors that contribute to and maintain eating pathology in adolescence or early adulthood [2, 3]. One of the most studied precursors of eating disorders is dieting [4, 5]. Weight and body image are common concerns among adolescent girls, and dieting is becoming more prevalent in younger age groups [6]. Although moderate dieting could be helpful in preventing childhood obesity, research suggests that dieting in childhood and adolescence may not be a benign practice [7]. Dissatisfaction of body shape and weight has been found to underlie dieting motivation of girls as young as 9 years of age [8]. Longitudinal studies have found that unhealthy weight control behaviors in adolescence predicted greater risk for obesity and eating disorders [9–12]. Early onset of dieting has been associated with negative health outcomes including eating disorders, obesity, and alcohol use at the 10-year follow-up [13]. However, it is unclear why dieting at an early age has an effect on eating disorders in adulthood. Disordered eating is highly comorbid with anxiety disorder and depression, and these psychological problems may put early dieters at greater risk for developing eating disorder later in life [14–16]. It is also possible that prolonged and chronic dieting could change personality traits to the commonly identified personality features of eating disorder patients including perfectionism, obsessive–compulsiveness, neuroticism, negative emotionality, harm avoidance, low self-directedness, and low cooperativeness [17, 18]. However, personality features of eating disorder patients tend to be heterogeneous [17, 19], and clustering of personality pathology within eating disorder patients have been identified [20].

This study directly compared how women with eating disorders who have started dieting early in their childhood and adolescence are different in their personality features, psychopathology, and eating disorder symptoms compared to those who have started dieting in their adulthood. We hypothesized that women who start dieting in childhood to adolescence (EARLYdieting group) would have higher MMPI-2 profiles and EDI-2 scores than women who start dieting in adulthood (ADULTdieting group). Those who start dieting early in life are more likely to have suffered eating disorder longer compared to those who start dieting in their adulthood. Thus, it is possible that the difference of the MMPI-2 and EDI-2 scores may be partially affected by how long the participants had suffered eating disorders. We hypothesized that after adjusting for the duration since the onset of eating disorder, the difference of MMPI-2 and EDI-2 between the two groups would be preserved. Finally, we hypothesized that the EARLYdieting group would use more serious and varied means of dieting.

Materials and methods

Participants

The participants were 205 women who sought treatment at an out-patient eating clinic specialized for eating disorders between 2007 and 2012 (Table 1). Participants were aged 20–34 years (mean = 24.42, SD = 3.2), 97.1 % were single and 2.9 % had never married. All patients had both binging and purging history and were diagnosed by a psychiatrist according to the criteria in DSM-IV-TR [21]. 37 patients were diagnosed with anorexia nervosa binge eating/purging type, 142 with bulimia nervosa purging type, and 26 with eating disorder not otherwise specified. Our study was carried out under the guidelines for the use of human subjects established by the Institutional Review Board at the Severance Mental Health Hospital.

Table 1 Demographic and clinical variables of participants by the onset of dieting
Full size table

Measures

Self-report measures of dieting history inquired about when the participant first started dieting (elementary school, middle school, high school; adulthood). Participants reported on using any or all of seven dieting methods: food restriction, exercise, use of laxative, use of diet pills and products, vomiting, chewing and spitting, and other. The onset of eating disorder was defined as the age when both binging and purging became present. Duration since the onset of eating disorder was calculated by subtracting the age of onset of eating disorder from the age when the patient was diagnosed at our clinic.

Minnesota Multiphasic Personality Inventory (MMPI-2) is a self-report personality test, which consists of 8 validity scales, 10 clinical scales, 15 content scales, and 16 supplementary scales [22]. The MMPI-2 was first translated into Korean by Han (1993) following a translation/back-translation procedure. The reliability and validity of the Korean version of MMPI-2 are well established [23]. Eating Disorder Inventory (EDI-2) is a self-report test with 11 subscales that measure symptoms and psychological correlates associated with eating disorders [24]. The Cronbach’s alpha for the Korean EDI-2 is 0.77–0.93 [25].

Beck Depression Inventory (BDI) is a self-report questionnaire that evaluates the intensity of depression [26]. The Cronbach’s alpha of the Korean version of BDI is 0.85 for the Korean population [27]. Beck Anxiety Inventory (BAI) is a self-report questionnaire that measures the severity of anxiety symptoms [28]. The reliability and validity of the Korean version of BAI is well established [29].

Statistical analysis

Since MMPI-2 scales are generally significantly correlated with one another, we used multivariate analysis of variance (MANOVA) for the 8 validity scales and 10 clinical scales to first find significant difference of mean scores of MMPI-2 between the two groups, with Wilk’s λ as the measure of significance. We performed univariate analysis of variance (ANOVA) with Bonferroni correction for error on each scale to determine which scale is contributing to the significant group difference. Then, we controlled for the duration since the onset of eating disorder using multivariate analysis of covariance (MANCOVA) and univariate analysis of covariance (ANCOVA). We performed the same analysis for the 11 subscales of EDI-2 with the same logic. We performed a Chi square analysis of seven dieting methods. For our post hoc power analysis, we used the GPOWER 3.1 program [30].

Results

101 women who had started dieting in their childhood to adolescence (EARLYdieting group) were compared to the 104 women who had started dieting in their adulthood (ADULTdieting group). Both MANOVA and MANCOVA (duration since onset of eating disorder as covariate is evaluated at 3.56; see Table 2) revealed significant difference in the MMPI-2 validity and clinical scales in relations to the onset of dieting, F(18, 186.00) = 1.74, p = .04, Wilk’s λ = .86, partial η 2 = .14; F(18, 185.00) = 1.70, p = .04, Wilk’s λ = .86, partial η 2 = .14. Univariate F tests revealed that EARLYdieting group scored higher on the clinical scales 1 and 3 (hypochondriasis, mean T score difference = 3.27, standard error = 1.27, p = .01; hysteria, mean T score difference = 3.61, standard error = 1.17, p < .01). When the covariate duration since the onset of eating disorder was controlled, difference was maintained in both scales 1 and 3 (hypochondriasis, mean T score difference = 2.80, standard error = 1.31, p = .03; hysteria, mean T score difference = 3.19, standard error = 1.21, p < .01), and ANCOVA additionally revealed significant difference in clinical scale 0 (social introversion, mean T score difference = −3.97, standard error = 1.90, p = .04). With a sample size of 205 and an alpha level at .05, the MANOVA analysis of the MMPI-2 would detect the observed between groups comparison effect size in the present study (partial η 2 = .14) (i.e., a large effect, according to Cohen’s, 1988, effect size conventions [31]) with 97 % power. The MMPI-2 validity and clinical scale profiles of the two groups prior to covariate inclusion is presented in Figs. 1 and 2.

Table 2 Means and standard deviations of T scores of MMPI-2 scales for EARLY and ADULT dieting groups and the results of group comparisons
Full size table
Fig. 1
figure 1

Minnesota Multiphasic Personality Inventory-2 (MMPI-2) validity scale mean T score profile for the two groups before applying covariate (duration since onset of eating disorder). VRIN variable response inconsistency, TRIN true response inconsistency, F infrequency scale, FB F-back scale, FP infrequency-psychopathology scale, L lie scale, K subtle defensiveness scale, S superlative self-presentation scale

Full size image
Fig. 2
figure 2

Minnesota Multiphasic Personality Inventory-2 (MMPI-2) clinical scale mean T score profile for the two groups before applying covariate (duration since onset of eating disorder). Scale 1 hypochondriasis, Scale 2 depression, Scale 3 hysteria, Scale 4 psychopathic deviate, Scale 5 masculinity–femininity, Scale 6 paranoia, Scale 7 psychasthenia, Scale 8 schizophrenia, Scale 9 hypomania, Scale 0 social introversion

Full size image

Both MANOVA and MANCOVA (duration since the onset of eating disorder as covariate is evaluated at 3.56; see Table 3) revealed significant difference in the EDI-2 in relations to the onset of dieting, F(11, 193.00) = 2.27, p = .01, Wilk’s λ = .89, partial η 2 = .11; F(11, 192.00) = 2.03, p = .03, Wilk’s λ = .90 partial η 2 = .09. Univariate F tests both before and after covariate inclusion revealed that the EARLYdieting group scored consistently higher than the ADULTdieting group on the bulimia scale (mean T score difference = 1.50, standard error = .56, p < .01; mean T score difference = 1.49, standard error = .58, p = .01). With a sample size of 205 and an alpha level at .05, the MANOVA analysis of the EDI-2 would detect the observed between groups comparison effect size in the present study (partial η 2 = .11) (i.e., a medium to large effect [31]) with 95 % power.

Table 3 Means and standard deviations of EDI-2 scales for EARLY- and ADULTdieting groups and the results of group comparisons
Full size table

92.3 % of the ADULTdieting group engaged in exercise as their dieting method compared to 80.2 % of the EARLYdieting group, χ 2(1, N = 205) = 6.37, p = .01. Individual Chi square analysis of other dieting methods did not reveal statistical difference. 25.7 % of the EARLYdieting group used 6 or more dieting methods compared to 11.5 % of the ADULTdieting group, χ 2(1, N = 205) = 6.85, p = .01. The two groups did not differ significantly in the BDI, t(203) = 1.21, p > .05, and BAI, t(203) = 1.93, p > .05. Chi square analysis revealed that the two groups did not differ by diagnostic subtypes, χ 2(2, N = 205) = .45, p = .80.

Discussion

Our hypothesis that women who started dieting in their childhood and adolescence would report more psychopathology was confirmed by the elevated scores of their MMPI-2 clinical scales 1, 3 (and 0 when adjusted for the duration of illness). They also reported more eating disorder related symptoms, as supported by the higher scores in EDI-2 bulimia scale. These differences between the two groups are independent of the diagnostic subtype of eating disorder and the levels of depression and anxiety as measured by the BDI and BAI. When we adjusted for the duration since the onset of eating disorder, the significant differences of both MMPI-2 and EDI-2 between the two groups were maintained. In other words, dieting during childhood and adolescence, independent of how long the participants suffered from eating disorders, is related to more psychopathology and severe eating disorder symptoms in adulthood among women diagnosed with eating disorder. One explanation for this difference may be the chronic starvation-state produced by early dieting onset. It has been found that prolonged caloric restriction alone can produce changes in personality as measured by increases in the MMPI scale 1, 2, and 3 [18].

A number of studies suggest that using multiple purging methods is associated with more severe eating disorder and greater general psychopathology [32–34]. We found that women who start dieting early were more likely to try six or more dieting methods than those who start dieting later in life. The fact that we did not find group difference for each dieting method (except for exercising, which ADULTdieting group was more likely to use) is consistent with the findings of Haedt et al. [33] that purging behaviors may be interchangeable but the use of multiple purging methods is associated with greater eating pathology at 10 year follow-up.

It should be noted that individuals with eating disorders have poor introspective skills and tend to distort and deny symptoms in self-reports [35]. This may have also affected the accuracy of the self-reported time line of dieting onset and eating disorder onset in our study. Moreover, our findings are based on retrospective correlates, and it is unsure if the observed differences in the MMPI-2 are consequent to, exacerbated by, and/or risk factors for eating disorders and dieting behaviors [36]. We did not have scores for the MMPI-2 supplemental scales and clinical subscales which could have offered richer information on the participants in terms of other risks and maintaining factors of eating disorders. Lastly, our sample consisted of Korean women with eating disorders. Although epidemiological studies report similar levels of eating disorder symptoms among college students in Korea and the US [37], sociocultural factors such as differences in dietary practice, beauty ideals, and obesity rates can influence the relationship between dieting and eating disorders. Furthermore, studies found that eating disorders in men are increasing and a significant numbers of males report unhealthy weight control practices [38, 39]. Future studies should assess whether the current findings are similar across gender.

Dieting has been accepted as a risk factor for eating disorder, but there has not been any study on how early onset of dieting can be different from later onset. In the context of previous research, our findings support that dieting is not a benign practice in childhood and adolescence even if it does not develop into a diagnosable eating disorder at the time. In addition, women who start dieting early took longer to seek treatment since the onset of both binging and purging. In other words, they have significantly longer duration of illness compared to women who start dieting later in life. Despite reporting higher psychopathology and eating disorder symptoms, early dieters who develop eating disorders may delay treatment seeking because as adolescents, they will be initially dependent on their parents’ assessment for treatment. If they are in denial and hiding symptoms from their parents, their social network would not know what to do and this will cause a delay [40]. Therefore, families, schools, and communities need to pay special attention to girls who start dieting during childhood through adolescence to recognize unhealthy dieting practices. There is a need for timely intervention even when the eating problems do not fully meet the diagnostic criteria of eating disorders. Future prospective research can track how prolonged dieting changes traits and how the change in personality exacerbate and/or maintain eating disorder symptoms.