Introduction

Disordered eating behaviors are one of the leading mental health problems among college women (Erskine et al., 2016; Striegel-Moore & Bulik, 2007). The high prevalence of bulimic behaviors is particularly alarming, with approximately 50% of female college students reporting habitual binge eating and 27% engaging in compensatory behaviors (Kelly-Weeder, 2010). Relevant literature has identified several psychosocial and cultural factors associated with bulimic behaviors such as insecure attachment, maladaptive perfectionism, low self-esteem, affect dysregulation, and thin body ideals of Western societies (Fairburn et al., 2003; Jakovina et al., 2018; Striegel-Moore & Bulik, 2007). In recent years, researchers have called for more research to investigate whether particular mechanisms for bulimic behaviors are universal across cultures or culture specific (Erskine et al., 2016; Levinson & Brosof, 2016; Pike & Dunne, 2015).

The purpose of this study is twofold. Guided by adult attachment theory (Bowlby, 1969/1982), trans-diagnostic theory (Fairburn et al., 2003), and available literature, an indirect effect model was proposed, in which attachment insecurity was the independent variable and bulimic behaviors were the dependent variable with three possible mediators. The first purpose of this study was to examine several cognitive-affective indirect-effect pathways through which insecure attachment may affect bulimic behaviors. Second, we explored cultural similarities and differences in the pathways of the model between college women from the U.S. and Korea.

Adult Attachment, Trans-diagnostic Theory, & Bulimic Behaviors

Adult Attachment & Bulimic Behaviors

Attachment theory (Bowlby, 1969/1982) posits that secure affective bonding with an accessible and responsive attachment figure in childhood engenders a sense of felt-security, the development of positive cognitive-affective mental representations (internal working models; IWMs) of self/others, and resilient relationship behavior patterns, which in turn lead to an individual’s greater general wellbeing over the course of life (Wang & Scalise, 2010). In contrast, consistently insecure attachment relationship experiences in childhood endanger self-worth, trust in others, and resilience in relationship, which in turn contribute to various mental health and behavioral problems such as depression, substance abuse, and disordered eating behaviors across the life-span (Han, 2013; Wang & Scalise, 2010). Attachment behavior patterns, although developed in infancy, tend to sustain “from cradle to grave”. Accumulated findings of extensive adult attachment literatures suggest two orthogonal dimensions of insecure attachment—attachment anxiety and attachment avoidance (Brennan et al., 1998). Attachment anxiety, conceptually representing the developed negative internal working models of self, is characterized as possessing low self-worth, a strong fear of rejection or abandonment from others, and an excessive need for approval in relationships (Wang et al., 2012). On the other hand, attachment avoidance represents the developed negative internal working models of others and people score high in this dimension often demonstrate strong discomfort for interpersonal intimacy and a compulsive need for self-reliance in relationship (Wang et al., 2012). In recent decades, some research has applied Bowlby’s attachment theory to non-Western populations, including Koreans (e.g., Lee & Koo, 2015; Zhu et al., 2016). The findings of the limited research body generally support the cross-cultural validity of the two-dimensional adult attachment model with some nuanced cultural differences in adult attachment influences being identified in relevant literature (Wang et al., 2021; Zamudio et al., 2020).

Adult attachment literature has generated noticeable empirical evidence for both the direct and indirect effects of insecure attachment on disordered eating behaviors. For instance, Dakanalis and colleagues reported that both anxious and avoidant adult attachment predicted eating disorder symptoms through maladaptive perfectionism (defined as unrealistically high standards accompanied by critical evaluation) for adult patients with eating disorders (Dakanalis et al., 2014). They found that the mediational effects were consistently significant across three different diagnostic groups of eating disorders—anorexia nervosa, bulimia nervosa (BN), and eating disorders not otherwise specified. The significant indirect effect of insecure attachment on eating disorder symptoms was also found in a study that focused on negative self-concept as the mediator, which indicated that both anxious and avoidant adult attachment styles significantly predicted eating disorder symptoms, including bulimic symptoms through poor self-concept (Demidenko et al., 2010). Another study examining the relationships between insecure attachment and eating disorders found that both attachment anxiety and attachment avoidance were directly and significantly associated with BN, however, only attachment anxiety, but not attachment avoidance, had a significant indirect effect on BN through emotional dysregulation (Jakovina et al., 2018).

Although aforementioned empirical studies all examined the impacts of attachment insecurity on bulimic behaviors, a noticeable common limitation is that they focused only on a single variable that mediated the relationship between attachment insecurity and bulimic behaviors. Since there likely exist multiple indirect effect pathways from insecure attachment to bulimic behaviors through distinct psychosocial mediators, a conceptual model is needed that accounts for more complex psychological mechanisms. To identify a broad range of developmental and maintenance mechanisms for bulimic behaviors, we turned to the trans-diagnostic theory (TDT) of bulimia nervosa proposed by Fairburn and colleagues (Fairburn et al., 1993, 2003).

Trans-diagnostic Theory

TDT is a cognitive-behavioral theory which proposes four core maintaining mechanisms of eating disorders, especially bulimia nervosa—clinical perfectionism, low self-esteem, mood intolerance, and interpersonal difficulties (Fairburn et al., 2003). Clinical perfectionism is the over-evaluation of the need for achieving demanding goals which contributes to self-criticism, fear of failure in weight control, and selective attention to shape and weight. Low self-esteem is a pervasive negative view of oneself largely independent of performance, which is often affected by clinical perfectionism. Mood intolerance is an inability to cope with adverse mood states such as depressive or anxiety symptoms, which are described as proximal triggers for bulimic behaviors. Finally, interpersonal conflicts and adverse social events are interpersonal contexts that often trigger bulimic symptoms and behaviors (Fairburn et al., 2003). Although these four factors are complimentary to one another in developing and maintaining BN, TDT implicates that there is a sequence among four factors in terms of their proximity to bulimic behaviors. Specifically, long-term or chronic interpersonal problems engender clinical perfectionism and low self-esteem, both of which in turn contribute to negative emotions such as depression and anxiety and then trigger binging and vomiting behaviors.

Since adult attachment insecurity is associated with more chronic interpersonal problems (Wang & Scalise, 2010), we believe that the combination of adult attachment theory and TDT offers a strong theoretical framework in understanding and accounting for the development and maintenance mechanisms of several risk factors for bulimic behaviors. Surprisingly, no empirical studies have simultaneously investigated the psychological mechanisms from these two theories.

The Proposed Conceptual Model

To address this gap, we developed a conceptual model to depict the influences of insecure attachment on bulimic behaviors through complex indirect pathways of maladaptive perfectionism, low self-esteem, and depression. Specifically, this model includes three mediators (maladaptive perfectionism, low self-esteem, and depression) in a sequential order linking insecure adult attachment to bulimic behaviors on the basis of the proposition suggested by adult attachment and trans-diagnostic theory (see Fig. 1).

Fig. 1
figure 1

Coefficients of the Direct and Indirect Effect Paths from Insecure Attachment to Bulimia. Note. The coefficients presented on top were from the Korean sample and on the bottom were from the US sample; * p < .05, ** p < .01, *** p < .001; the effects of age, BMI, and social desirability were controlled in all models

Full size image

A sizable body of attachment studies provide empirical evidence for various segments of the indirect pathways embedded in the proposed model. For example, Han and colleagues reported several sequential pathways from poor parental care or childhood attachment trauma, to insecure adult attachment and low self-esteem, to greater psychological distress (Baek & Han, 2020; Han, 2013, 2014). Wei and colleagues revealed significant indirect effects of insecure adult attachment to negative mood through maladaptive perfectionism and poor affect regulation (Wei et al., 2005, 2006). Tasca and colleagues provided evidence supporting the pathways from insecure adult attachment to ineffective affect regulation strategies to more depressive and eating disorder symptoms (Tasca & Balfour, 2014; Tasca et al., 2009). Taken together, these empirical findings support the proposed model with multiple mediators and complex indirect effect pathways from insecure adult attachment to bulimic behaviors.

Cultural Factors for Bulimic Behaviors

The socio-cultural model of eating disorders suggests that the cultural context may have substantial impacts on the development and expression of disordered eating behaviors (Levinson & Brosof, 2016). Some empirical evidence supports this notion. For instance, internalization and pursuit of the thinness ideal of Western cultures has been consistently identified in relevant literature as a risk factor for bulimic behaviors (Anderson-Fye, 2018; Keel & Klump, 2003). Researchers suggest that eating disorders, including bulimia, are not just culture-bound, but rather outcomes of dynamic interactions between the extent of Western influence and level of industrialization and urbanization (Pike & Dunne, 2015). Moreover, the prevalence of eating disorders like bulimia nervosa is found to be associated with age and income (Erskine et al., 2016). In recent years, the prevalence of bulimic behaviors has increased sharply in several Asian countries, including South Korea (Han et al., 2018; Lee et al., 2010). Researchers have attributed the rise of bulimia in Asian countries to a wide range of factors such as globalization of Western ideals of the thin female body, industrialization, urbanization, increased food accessibility, beauty industries, and cultural characteristics of valuing physical appearance for women (Pike & Dunne, 2015). Specifically, disordered eating behaviors dramatically increased for young women in Asia’s high-income families across Japan, Korean, Singapore, and Taiwan (Gordon, 2001; Lee, 2004; Tsai, 2000; Ung, 2003). Particularly, Korean cultures tend to emphasize appearance, beauty, and youthfulness for women and an internalization of the thin ideal is widespread among many college women in South Korea (Han et al., 2018). In fact, some research findings point out that Korean college women show a greater degree of disordered eating symptoms, including bulimic behaviors than those found in the U.S. (Jackson et al., 2006).

Cross-cultural comparison research in bulimic behaviors with multi-national samples is extremely scarce and the limited empirical studies available in the literature tend to focus on comparing the prevalence rate of different types of disordered eating behaviors (e.g., Gunewardene et al., 2001; Jung & Forbes, 2007). Thus, there is a great need for more research to explore the potentially differential effects of various psycho-cognitive-affective variables on bulimic behaviors. To address the gap, this study also explored similarities and differences in the direct/indirect pathways embedded in the model between U.S. and Korean college women.

The Present Study

Guided by adult attachment theory, trans-diagnostic theory, and available literature, a mediational model with three mediators (i.e., maladaptive perfectionism, self-esteem, and depression) was developed to depict the complex indirect effects of attachment insecurity on bulimic behaviors (see Fig. 1). We first explored the cross-cultural differences in all variables of interest between the two samples. Then, we separately examined the direct and indirect effect paths depicted in the model using a sample of female university students recruited from South Korea and another sample of female university students recruited from the U.S. to explore the similarities and differences in these direct/indirect paths of insecure attachment to bulimia between the two country samples. Based on the guiding theories and available literature, it was reasoned that insecure attachment would be positively associated with maladaptive perfectionism and negatively associated with self-esteem, which would in turn result in increased depressive symptoms and greater unhealthy eating behaviors/attitudes. In doing so, two stages of sequential pathways of three mediators were hypothesized based on both the cognitive-behavioral approach and TDT, first, from maladaptive perfectionism to low self-esteem, and then, from low self-esteem to depression.

Specifically, this study hypothesized there would be a significant direct effect as well as seven significant indirect effect paths from insecure attachment variables to bulimic behaviors (see Table 3 for all hypothesized indirect effect paths). The mean comparisons in all measured variables and effect sizes of direct/indirect effects between the two country samples were also explored. Due to the rarity in cross-cultural research for what factors contribute to bulimic behaviors experienced by people from different country, the mean and effect size comparisons were exploratory in nature and we did not have specific hypotheses for them.

Method

Participants & Procedures

Two samples of college women participated in this study; one from the US and another from South Korea. For the US sample, a total of 271 female students were recruited from a large public university in the Southwestern region of the US. Participants’ ages ranged from 18 to 36 years old with a mean of 20.30 (SD = 3.05). The self-reported ethnicity consisted of European American (55.7%), African American (19.9%), Latinx or Hispanic (12.9%), Asian or Asian American (6.6%), and others (5.0%). The reported classification consisted of freshmen (40.6%), sophomores (21.8%), juniors (21.0%), seniors (15.1%), and others (1.5%). Participants were recruited from several undergraduate psychology courses and those who were interested signed up for a time slot using an online appointment making system to come to a particular campus building. Participants completed the research questionnaires independently in a small group format of 1 ~ 15 participants. Most of them took about 30 min to complete the questionnaire. As compensation for their time, each participant received extra credits to be applied to their psychology course.

For the Korean sample, a total of 421 female students were recruited from ten psychology or education courses offered by six large four-year universities in three different regions of South Korea. The Korean sample had a mean of 21.76 (SD = 3.50) in age, ranging from 18 to 34 years old. All Korean participants identified “Korean” as their ethnicity. The reported classification consisted of freshmen (20.2%), sophomores (30.6%), juniors (24.2%), seniors (23.3%), and others (1.0%). Potential participants in our Korean sample were recruited following very similar procedures used for the US group, except for one specific difference. After the announcement of the research opportunity, those who consented to participate were instructed to complete the research questionnaire right away in the classroom, whereas those declining to participate were asked to work on alternative tasks assigned by the instructor. Approximately 90% of students from those ten courses visited by the research assistant completed the questionnaires. All research participants in these courses either received extra credits for the course or were offered a gift coupon of Starbucks coffee.

Instruments

The research questionnaire included several demographic items (e.g., age, height and body weight, race/ethnicity, school years, and majors). BMI (Body Mass Index) was calculated using the formula kg/m2 and used as one of covariates to control its effect. BMI ranged from 15.9 to 49.2 (M = 27.3; SD = 7.0) for the U.S. sample and from 14.9 to 32.1 (M = 20.0; SD = 2.1) for the Korean sample. Korean participants were given a Korean version of the questionnaires. To address issues related to measurement equivalence of instruments, the researchers purposefully selected self-report instruments with an available Korean version that had been applied to Korean samples or validated in previous studies by other researchers.

Adult Attachment Anxiety & Avoidance

The Experiences in Close Relationships-Revised Scale was used to measure adult attachment variables (ECR-R; Fraley et al., 2000). Participants are asked to rate descriptions of how they generally experience relationships on a 7-point Likert-type scale ranging from 1 (disagree strongly) to 7 (agree strongly). The ECR-R consists of 36-items that were selected from 323 items based on responses provided by more than 1,000 university students in the U.S. using item response theory analysis (Fraley et al., 2000). The two subscales reflect the two underlying factors of insecure attachment, anxiety and avoidance, with higher scores indicating greater attachment anxiety or attachment avoidance. The anxiety subscale measures fear of abandonment and rejection (18 items; e.g., “I rarely worry about others leaving me”) whereas the avoidance subscale assesses discomfort with intimacy and dependence (18 items; e.g., “prefer not to be too close to others”).

Following the method of back translation, the ECR-R was translated into Korean by Kim et al. (2011). The Korean version of ECR-R was administered to over 1,300 Korean college students and went through both confirmatory factor analyses and item response theory-based differential item functioning analyses. To account for attachment insecurity, as well as positively and negatively-worded factors, a bifactor model was used to validate the two subscales of the ECR-R Korean version (Kim et al., 2011). Both scales have demonstrated good internal reliability with U.S. samples (α = 0.93 for anxiety, α = 0.94 for avoidance; Fraley et al., 2000) and in Korean samples (α = 0.93 for anxiety, α = 0.90 for avoidance; Kim et al., 2011). In our study, Cronbach’s alpha was 0.92 for the scores on the anxiety subscale and 0.94 for the scores on the avoidance subscale for the U.S. sample, and 0.90 for anxiety and 0.92 for avoidance for the scores in the Korean sample.

Maladaptive Perfectionism

The maladaptive perfectionism variable was measured by the Perfectionism subscale of the Dysfunctional Attitude Scale (DAS; Weissman & Beck, 1978). The DAS includes two subscales, based on a sample of university students to measure dysfunctional beliefs associated with vulnerability for depression. The Korean version of the DAS was translated and validated by Kwon (1994) and further verified by Noh and Kim (2016) in a recent study with Korean college students. The Perfectionism subscale includes 15 items assessing high personal standards or maladaptive evaluation concerns. Example items include “It is difficult to be happy unless one is good looking, intelligent, rich, and creative” and “People will probably think less of me if I make a mistake.” Participants were asked to rate the degree of agreement on each statement using a 7-point Likert-type scale ranging from 1 (totally agree) to 7 (totally disagree). The higher the score, the greater the level of perfectionism. The Perfectionism subscale was reported to have a Cronbach’s alpha of 0.91 for a sample of the U.S. college students (Jacobs et al., 2009) and Cronbach’s alpha of the Korean version was found to be 0.91 in a sample of Korean college students (Noh & Kim, 2016). Internal consistency for the current study was α = 0.89 for the U.S. sample and α = 0.85 for the Korean sample.

Self-Esteem

Self-esteem was measured using the Rosenberg Self-Esteem Scale (RSES), which consists of 10 items (Rosenberg, 1965). Each item is rated on a 4-point Likert-type scale from 1 (strongly agree) to 4 (strongly disagree), and responses are summed to produce a total self-esteem score, with higher scores indicating more positive self-esteem. An example item includes “I take a positive attitude toward myself.” The RSES was originally validated with a sample of adolescents (Rosenberg, 1965) but has been widely employed by a large number of studies focused on college student samples with adequate reliability. For instance, a Cronbach alpha coefficient of 0.92 was reported by Wang and Castaneda-Sound (2008) in a study with 367 college students. The Korean version of the RSES was developed by Jeon (1974) and has been widely used with Korean college student samples yielding evidence for the scale’s reliability (α ranging from 0.75 to 0.88) and validity (Lee et al., 2009). The internal consistency for our US sample was 0.91 and 0.88 for the Korean sample.

Depressive Symptoms

The Center for Epidemiologic Studies- Depression Scale (CES-D; Radloff, 1977) was used to assess depressive symptoms. Respondents are asked to respond to the 20 items based on the frequency of occurrence of the depressive symptoms in the past week using a 4-point Likert-type scale ranging from 1 (rarely or none of the time) to 4 (most or all of the time). An example item includes “I thought my life had been a failure”. Higher scores indicate greater depressive symptomology. The CES-D has been widely used by a large number of empirical studies with samples of community residents and college students and has been translated into several different languages. Adequate internal consistency reliability (α = 0.85) and test–retest reliability of 2 to 8 weeks (r = 0.51 to 0.67) have been reported by the scale’s developer (Radloff, 1977). More recently, Chin and colleagues reported internal consistency reliability (α = 0.86) and test–retest reliability of 2 weeks (r = 0.91) on a Chinese sample (Chin et al., 2015). The Korean version of the CES-D was developed and validated by Chon et al. (2001) and has been widely used in Korea research with acceptable psychometric properties (Noh & Kim, 2016). Internal consistency was α = 0.90 for both the U.S. and Korean sample in this study.

Bulimic Behaviors

Disordered eating behavior and attitudes related to bulimic behaviors were assessed using the subscale of Bulimia and Food Preoccupation (BFP) of the Eating Attitudes Test (EAT-26; Garner et al., 1982). The BFP subscale is often called Bulimia subscale because research suggests that the BFP measures a valid risk factor for BN (Orbitello et al., 2006). The BFP subscale includes 6 items and respondents are asked to rate each item using a scale ranging from 1 (never) to 6 (always), with higher scores indicating greater disordered eating symptoms. Example BEF items include: “I feel that food controls my life” and “I have the impulse to vomit after meals.” Gleaves et al. (2014) reported a mean internal consistency reliability score of 0.67 (ranging from 0.60-0.79) from 23 studies with samples of non-clinical adolescents and college students that used the BFP subscale of the EAT-26. The test–retest reliability over a 4–5-week period of the BFP was reported to be 0.87 in a previous study (Banasiak et al., 2001). Ko and Cohen (1998) developed and validated the Korean version of the EAT-26 through a process of translation and back-translation with Korean women. The Korean version of the scale was found to have a Cronbach’s α of 0.81 in a large nation-wide sample of Korean adult women (Rhee et al., 1998). The Cronbach alpha of the BFP was 0.86 and 0.80 for the U.S. and Korean sample, respectively in this study.

Data Analysis Plan

The data analysis plan generally included three steps. First, the preliminary analyses began with conducting data cleaning processes to examine the pattern of missing data and check for the normality of the data. Then, a series of t tests were used to examine the differences in all variables of interest between the Korean and American sample. The last step was to examine the hypothesized mediational effects using PROCESS, which is based on the multiple regression approach with the bias-corrected bootstrapping method (Hayes, 2013). Research has suggested that bootstrapping is a robust test in examining indirect effect (Shrout & Bolger, 2002). Additionally, findings from the a priori power analysis conducted through the G*Power program (parameters: Power = 0.9, α = 0.05, predictor # = 8, and small to medium effect size = 0.10) resulted in a minimal sample of 199 participants which suggested that both country samples had sufficient participants.

Results

Preliminary Analyses & Between Country Differences

The SPSS version 26 was used for data analyses. Missing data were first examined by Little’s χ2, and the results suggested Missing Completely at Random. Since all variables contained less than 5% of missing data, the Expectation–Maximization Algorithm was used to handle the missing data as suggested by Schlomer et al. (2010). Univariate and multivariate outliers were checked using Boxplot and Mahalanobis distance statistic which resulted in removal of a small numbers of outliers (n = 6). Skewness statistics of variables were checked and the findings suggested that all measured variables were normally distributed except depressive symptoms and bulimia, both of which showed positive skewedness. Since the positively skewed trend in the two psychological symptoms variables of both country samples was believed to reflect the characteristics of the non-clinical samples (i.e., university students), a decision was made not to perform data transformation (Tabachnick & Fidell, 2019). The means, standard deviations, and bivariate correlations of all measured variables as well as age, social desirability, and BMI from the Korean and American sample were computed and findings are displayed in Table 1. The results indicated that BMI, social desirability, and age were significantly correlated with different variables of interest. Therefore, these three variables were entered in the multiple regression analyses as covariates to control for their effects.

Table 1 Bivariate Correlations, Means, Standard Deviations, and Between-Group Comparisons
Full size table

To examine the differences between the two country samples in the means of all measured variables, a series of t tests were conducted, and the results are shown in Table 1. The findings indicated that compared to the American sample, Korean participants reported significantly higher scores in attachment avoidance (t = -3.20, p = 0.004) and maladaptive perfectionism (t = -5.72, p < 0.001) but significantly lower scores in attachment anxiety (t = 2.10, p = 0.037), self-esteem (t = 2.69, p = 0.007), BMI (t = 20.13, p < 0.001), and social desirability (t = 3.46, p = 0.001).

Mediational Analyses for Korean Participants

Two multiple regression analyses were conducted using PROCESS Model 6 (Hayes, 2013) to examine the hypotheses with one of the insecure attachment variables as the independent variable (IV) for each analysis. Specifically, the following variables were entered into the PROCESS conceptual model: X = attachment anxiety (or attachment avoidance), M1 = maladaptive perfectionism, M2 = self-esteem, M3 = depressive symptoms, Y = bulimia and food preoccupation, and covariates = social desirability, BMI, age, and the insecure attachment variable that was not entered as the IV.

Results indicated that both attachment anxiety (b = 0.33, t = 8.87, p < 0.001) and attachment avoidance (b = 0.22, t = 6.47, p < 0.001) were significant predictors for maladaptive perfectionism while controlling the effects of the covariates. The model accounted for 39% of variance in maladaptive perfectionism (R2 = 0.39). In the model where self-esteem was the DV, attachment anxiety (b = -0.09, t = -3.34, p = 0.001), attachment avoidance (b = -0.09, t = -3.91, p < 0.001), and maladaptive perfectionism (b = -0.17, t = -5.59, p < 0.001) were significant predictors while controlling the effects of covariates. The total effect of all variables included in this model accounted for approximately 30% of variance in self-esteem (R2 = 0.30). Additionally, attachment anxiety (b = 0.09, t = 3.53, p < 0.001), maladaptive perfectionism (b = 0.09, t = 2.95, p = 0.003), and self-esteem (b = -0.35, t = -7.12, p < 0.001) were significant predictors for depressive symptoms. The model explained 34% of variance in depressive symptoms (R2 = 0.34). In the final model with attachment anxiety as the IV and all mediators and covariates, both maladaptive perfectionism (b = 0.22, t = 3.80, p < 0.001) and depression (b = 0.29, t = 3.29, p = 0.001) were significant predictors for preoccupation with food but attachment anxiety (b = 0.07, t = 1.41, p = 0.159), attachment avoidance (b = -0.06, t = -1.48, p = 0.139), and self-esteem (b = 0.13, t = 1.38, p = 0.168) were not. Together, all variables included in the final model accounted for 16% of variance in bulimic symptoms (R2 = 0.16) which is considered slightly above a medium effect size (Cohen, 1992). Regression coefficients of the predictors for different DVs are shown in Table 2.

Table 2 Multiple Regression Coefficients of All Models
Full size table

Findings from the 5,000-sample, bias-corrected bootstrap analyses indicated significant indirect effects of attachment anxiety on bulimia through paths 1, 3, 5, 6, and 7 as indicated by their 95% confident intervals (CI) not including zero (Hayes, 2013). However, the direct effect from attachment anxiety to bulimia was not significant (95% CI [-0.024, 0.161]). When attachment avoidance was the IV, significant indirect effects of attachment avoidance on bulimia were found on the same paths: 1, 3, 5, 6, and 7, but the direct effect from attachment avoidance to bulimia was not significant (95% CI [-0.144, 0.020]). Taken together, the results indicated that attachment insecurity demonstrated significant impacts on bulimic behaviors for Korean participants through the indirect-effect paths of 1, 3, 5, 6, and 7, but that the direct effect path and the indirect-effect paths of 2 and 4 were not significant. Table 3 displays the β product values (by of multiplying the standardized regression coefficient value of each individual path included in the indirect effect pathway) and 95% CI ranges of the bias-corrected bootstrap analyses of all indirect effect paths. All of the β product values are considered in small effect size (Cohen, 1992). Figure 1 displays the standardized path coefficients of all individual paths in the indirect effect model.

Table 3 Significance of Direct/Indirect Effect Paths from Insecure Attachment to Bulimia by Participant Country
Full size table

Mediational Analyses for American Participants

The same analyses used for Korean participants were conducted again to examine the indirect effect models with the American sample. Results indicated that both attachment anxiety (b = 0.26, t = 5.34, p < 0.001) and attachment avoidance (b = 0.11, t = 2.66, p = 0.008) significantly predicted maladaptive perfectionism. The model accounted for 24% of variance in maladaptive perfectionism (R2 = 0.24). Additionally, attachment anxiety (b = -0.15, t = -5.27, p < 0.001) and maladaptive perfectionism (b = -0.19, t = -5.31, p < 0.001) were significant predictors for self-esteem but not attachment avoidance (b = -0.03, t = -1.06, p = 0.288). Together, all variables in included in the model explained approximately 36% of variance in self-esteem (R2 = 0.36). When depression was the DV, attachment anxiety (b = 0.08, t = 3.37, p = 0.001), attachment avoidance (b = 0.07, t = 3.58, p < 0.001), and self-esteem (b = -0.31, t = -6.21, p < 0.001) were significant predictors. The total effect of all variables in the model accounted for 41% (R2 = 0.24) of variance in depressive symptoms. In the final model with bulimia as the DV, maladaptive perfectionism (b = 0.15, t = 2.55, p = 0.011), self-esteem (b = -0.24, t = -2.27, p = 0.024), and depression (b = 0.36, t = 2.90, p = 0.004) were significant predictors for preoccupation with food. The final model accounted for 28% of variables in preoccupation with food (R2 = 0.28), which is considered large effect size (Cohen, 1992). Regression coefficients of the predictors for different DVs are shown in Table 2.

Findings from the 5,000-sample, bias-corrected bootstrap analyses indicated significant indirect effects of attachment anxiety on bulimia through all indirect-effect paths except for path 5. When attachment avoidance was the IV, results indicated significant indirect effects of attachment avoidance through four indirect-effect paths of 1, 3, 4, and 7 for US participants. Neither the direct effect of attachment anxiety (95% CI [-0.045, 0.151]) nor of attachment avoidance (95% CI [-0.136, 0.025]) on bulimia was significant for the US sample. Table 3 displays the β product values and 95% CI ranges of the bias-corrected bootstrap analyses of all indirect effect paths. All of the β product values are considered in small effect size (Cohen, 1992).

Discussion

Guided by both adult attachment and transdiagnostic theory of bulimia, this cross-cultural study was conducted to understand the interrelationships among adult attachment, maladaptive perfectionism, self-esteem, depression, and bulimia among college women from the US and South Korea. Overall, our results showed that insecure attachment was associated with bulimic behaviors through three-serial mediational functions of maladaptive perfectionism, lowered self-esteem, and heightened depression both in US and Korean women. Additionally, findings suggest that the predictors included in the models explained a larger amount of variance in bulimic behaviors for US participants than the amount for the Korean sample. Specifically, the total effect of all variables included in the final model accounted for 16% of variance in bulimia for Korean participants, which is considered above a medium effect size (f2 = 0.19), and explained 28% of variance in bulimia for our American sample, which falls in a large effect (f2 = 0.39) suggested by Cohen (1992). Furthermore, this study advances the literature by identifying several culturally specific nuances in the specific indirect-effect pathways through which adult attachment insecurity affects bulimic behaviors, including that adult attachment is indirectly associated with bulimia via maladaptive perfectionism and self-esteem only in the US group.

Our finding indicated that maladaptive perfectionism mediated the association between insecure attachment and bulimia, which is consistent with previous research (e.g., Dakanalis et al., 2014). The present study provides further support for the cross-cultural similarity of this indirect effect path (i.e., attachment ➔ maladaptive perfectionism ➔ bulimia). Adult attachment research has found that individuals high in attachment anxiety and avoidance are more likely to show maladaptive perfectionism and that maladaptive perfectionism is reasoned to be resultant from unmet needs in early attachment relationships (Dakanalis et al., 2014; Wei et al., 2006). Notably, women with attachment anxiety may have internalized disappointment from their attachment figures to strive for perfection, as well as gain acceptance and approval from others (Wei et al., 2006). On the other hand, women with attachment avoidance tend to obtain perfection to avoid other’s rejection, as well as protect their hidden sense of imperfections (Dakanalis et al., 2014; Wei et al., 2006). Applying the attachment framework to disordered eating, the insecurity within self and others increases the possibilities of setting unrealistic high standards to overly evaluate shape, weight, eating, and body thinness (Dakanalis et al., 2014), leading to the development and maintenance of bulimia (Fairburn et al., 2003).

From a cross-cultural comparison perspective, another important finding of the current study is the difference in the examination of maladaptive perfectionism and self-esteem as indirect effect pathways accounting for the associations between adult attachment and bulimia. Our study indicated that insecurely attachment women in the US were more likely to experience maladaptive perfectionism and diminished self-esteem, which in turn, increased the likelihood of experiencing bulimia. Maladaptive perfectionism causes one to create unattainable and unrealistic goals, which inevitably enlarges the discrepancy between one’s standards and actual performance (Enns & Cox, 2002). This discrepancy can then create self-doubt in oneself and heighten self-criticism, resulting in diminished self-esteem (Pelletier Brochu et al., 2018). Eventually, lack of self-esteem can become a risk factor for bulimia (Cheng, 2014). Perhaps, individuals with low self-esteem are more likely to monitor and evaluate self-worth through achieving socially valued domains (i.e., body image, weight, and shape control) to gain societal acceptance (Fairburn et al., 2003). In short, the study presents that maladaptive perfectionism and self-esteem explain the link between adult attachment and bulimia among women in the US.

This indirect effect pathway, however, is not significant among Korean women, suggesting culturally specific nuances in the model. Previous studies suggest that eating disorders are heavily influenced by the intersections of sociocultural factors and cultural context (Jung & Forbes, 2007). For instance, studies highlight that lookism, the belief that a good appearance play a critical, if not the most important, role for one’s future and success, in Korean society knowingly and unknowingly can have a pervasive impact on a variety of psychosocial functioning, ranging from interpersonal relationships to job satisfaction/insecurity and psychological wellness (Han et al., 2018). Although tentative, the link between self-esteem to bulimia in the proposed model may be moderated by culturally specific factors such as lookism in South Korea. Further studies can examine if culturally-specific factors can interact with country of membership to moderate the aforementioned link.

Our findings provide cross-cultural evidence supporting the interrelationships among adult attachment, maladaptive perfectionism, self-esteem, depression, and bulimia. Specifically, the results suggest that insecure attachment is associated with maladaptive perfectionism and diminished self-esteem, which in turn, increases the possibilities of depression and bulimia. Binge eating, an outlet to cope with perceived inadequacy and an attempt to control depressive mood, can bring temporary emotional relief (Smyth et al., 2007); however, this pattern functions as a negative reinforcement that can maintain bulimia in the long run (Fairburn et al., 2003; Pelletier Brochu et al., 2018). In short, this particular indirect effect mechanism (adult attachment➔ maladaptive perfectionism➔ self-esteem➔ depression➔ bulimia) is cross-culturally applicable among the US and Korean women groups.

Eating disorders are traditionally characterized as disorders that occur to young, White, middle to high social economic status (SES) women (Han et al., 2018). The perception that eating disorders are a “western cultural-bound” syndrome has led to some biases that Asians are immune to disordered eating such as bulimia and less likely to have disordered eating due to a “petite body size” (Cheng, 2014). Our research advances literature by providing cross-cultural applicability of an attachment-based model explaining the complex effects of a number of psychological factors for bulimic behaviors reported by Korean women.

Limitations and Future Directions

This study has several limitations. First, due to its cross-sectional and correlational nature, the directionality and causality of the results are not warranted. Future research should consider experimental and/or longitudinal research designs to examine the replicability of this study. Second, the study was conducted based on self-reported measures. Although we controlled the confounding variables such as social desirability and BMI, objective measures (e.g., observational data, clinical assessment) can provide critical information. Third, conclusions of the current study are exclusively applied to bulimia, and the mechanisms might not be generalized to other eating disorder subtypes. Future studies may take into consideration other subtypes of eating disorders. Lastly, although this study advanced existing literature to examine the cross-cultural applicability of our proposed model, we did not explore what specific cultural factors differentiated the attachment influences on bulimia between these two samples. Future research may examine the moderator effects of specific sociocultural factors (e.g., media internalization, lookism, gender equality, collectivism/individualism) on the pathways embedded in the model to identify what factors have contributed to the significant differences.

Counseling Implications

Despite these limitations, this cross-cultural study has several implications for counseling services. Our study unfolds several mechanisms by applying both adult attachment and the trans-diagnostic theory of bulimia to identify three mediators: maladaptive perfectionism, self-esteem, and depression. Subsequently, our results suggest that the treatment of bulimia may target one’s insecure attachment style, dysfunctional schemas of self-evaluation, such as maladaptive perfectionism and/or core low self-esteem, and depressive mood. When working with college women with bulimic symptoms, counselors need to be aware that bulimic symptoms often arise from dynamic interactions of perfectionistic self-criticism, poor self-worth, and depressive symptoms (Fairburn et al., 2003). Interventions focusing on identifying and disputing perfectionistic, overtly self-critical voices to increase clients’ awareness may be helpful in addressing bulimic symptoms (Agras et al., 2000). Counselors may also help clients with bulimic symptoms reduce the discrepancy between their internalized ideal and realistic self, as well as practice self-compassion with the goal of enhancing their self-esteem (Wonderlich et al., 2015).

Since the findings support that anxious and avoidant attachments are the upstream sources underlying perfectionistic and self-critical voices, negative image of self, and depressive symptoms, mental health counselors and other professionals should help their clients address where their fear of rejection and abandonment, as well as excessive needs for approval or compulsive independence, stem from (Wang et al., 2012). By doing so, clients may be better able to put more effort into not only changing maladaptive thoughts, negative self-image, and depressive symptoms but also transforming their insecure attachment. In counseling sessions, counselors may assist clients in reflecting on how their current relationship patterns relate to their childhood relationship experiences with attachment figures (e.g., mother, father). Interventions, such as two-/empty chair dialogue may be helpful in facilitating considered clients’ honest and genuine conversations with an attachment figure (Wonderlich et al., 2015). The counselors should strive to serve as a secure base for their clients. With adequate emotional support and empathy, clients may have a cathartic experience and gain insight into how painful early experiences in close relationships contributed not only to their current relationship patterns, but also to their maladaptive perfectionist thoughts, low self-worth, depressive mood, and even bingeing and purging behaviors.