Schema Therapy (ST) was developed to address long-standing psychological disorders and entrenched personality traits (Young et al., 2003), and a growing number of studies have demonstrated its effectiveness (Masley et al., 2012; Jacob and Arntz, 2013; de Klerk et al., 2016). ST is based on the notion that Early Maladaptive Schemas (EMS) develop as a result of the interaction between temperament and unmet core emotional needs during childhood. EMS consist of patterns of memories, cognitions, emotions and physical reactions that drive coping mechanisms, and become increasingly maladaptive over time. Whereas EMS refer to stable “traits,” schema modes represent the moment-to-moment “states” and coping responses that we all experience. Schema modes are traditionally grouped into four main categories: (1) Innate Child Modes, that represent the emotions experienced in the context of unmet needs; (2) Internalized/Introject [Parent] Modes, which represent the internalized messages from childhood, including parents, teachers, other caregivers, and culture; (3) Coping Modes, which are the survival mechanisms developed during childhood and adolescence to cope with unmet emotional needs, and (4) Adaptive (Healthy) Modes. In ST, the emphasis is on developing healthy coping mechanisms that facilitate the process of developing adaptive ways of meeting one's interpersonal and emotional needs, whilst weakening maladaptive modes and associated EMS (Young et al., 2003).

The (124-item) Schema Mode Inventory (SMI) (Young et al., 2007) was developed to measure schema modes through self-report. A shortened (118 item) version of the original SMI (Young et al., 2007) was validated within a sample of healthy controls, axis I and axis II patients (Lobbestael et al., 2010) and found to have acceptable internal consistency and test-rest reliability. A 14-factor model emerged, consisting of: five child modes, five dysfunctional coping modes, two dysfunctional parent modes and the adaptive Healthy Adult mode. The SMI was primarily developed to measure schema modes in the Borderline and Antisocial PDs, however, it has recently successfully been adapted (SMI-2) to more appropriately measure schema modes as they manifest within Cluster C and paranoid, histrionic and narcissistic PDs (Bamelis et al., 2011). Previous studies have highlighted the need for exploratory research that examines schema modes within specific clinical groups in order to begin to delineate the profiles and new sub-modes that may be identified in these populations (Lobbestael et al., 2010).

A significant proportion of those with Eating Disorders (EDs) do not respond or do not complete treatment (Waller et al., 2009) and go on to develop chronic symptoms (Keller et al., 2006; Castelnuovo et al., 2011). Rigid personality traits are common in this population, with up to 69% of the ED population meeting criteria for a co-morbid personality disorder (PD) (Blinder et al., 2006), a factor that has been shown to adversely affect treatment outcomes (Grilo et al., 2007; Sallet et al., 2010; Farstad et al., 2016). In spite of improved outcomes with Enhanced CBT (CBT-E) (Fairburn et al., 2008; Castelnuovo et al., 2017; Pietrabissa et al., 2017), it remains less effective at addressing rigid co-morbid personality characteristics, such as perfectionism and avoidant traits (Byrne et al., 2011). The Schema Therapy Model for EDs has evolved over the past 10 years (Waller, 2003; Waller et al., 2007; Simpson, 2012), prompted by the need for therapeutic approaches that address deeper level cognitions and entrenched behaviors that do not respond to first-line treatments (Waller, 1997; Waller and Kennerley, 2003; Sorgente et al., 2017).

Preliminary research suggests that Schema Therapy may be highly suited to the needs of those with EDs, particularly those with high levels of complexity or chronicity (Simpson et al., 2010; Simpson and Slowey, 2011; McIntosh et al., 2016). This approach goes beyond maintenance factors to address the core schema-level beliefs that underpin ED psychopathology. Preliminary studies in this area suggest that those with EDs may experience significantly higher levels of maladaptive modes than community samples (Voderholzer et al., 2014; Talbot et al., 2015). Preliminary conceptualizations of the ED population suggest that all three coping styles (avoidance, surrender and overcompensation) play a significant role in driving and perpetuating ED behaviors (Waller and Kennerley, 2003; Simpson, 2012). Recent evidence supports the notion that maladaptive coping modes mediate the relationship between perceived negative parenting and eating disorder symptoms (Sheffield et al., 2009; Brown et al., 2016). In particular, the “Eating Disorder Overcontroller (Coping) Mode (EDO)” has been proposed as highly relevant to the ED population in recent theoretical conceptualizations (Simpson, 2012, 2016), and supported by findings from a recent study (Brown et al., 2016). This mode is theoretically linked to the Unrelenting Standards schema, and functions as a form of overcompensation, whereby perfectionism, achievement, and competitiveness, often focused on the body and eating patterns, are utilized to provide distance from underlying feelings of vulnerability and distress by generating a sense of competence, omnipotence and control.

The EDO has some parallels with the “Perfectionistic Overcontroller” mode that has been identified as highly relevant within Obsessive-Compulsive Personality Disorder (Bamelis et al., 2011), which has high comorbidity with EDs (Halmi et al., 2005). However, in eating disorders, it is proposed that the EDO has a more substantial focus on controlling the body, attaining a state of purity, and an emotional “high” associated with overcoming basic human needs (Simpson, 2016). Although the EDO is tapping into a similar construct to that of other clinical perfectionism measures, it is focused on specific characteristics and behaviors of the ED population, and focused on a broader concept than that measured by these other scales. In addition, the Helpless Surrenderer has been proposed as a mode which has particular relevance to those with EDs. In this mode the patient avoids expressing vulnerability or emotional needs directly, but instead seeks help passively, such as through withdrawal, complaint, or seeking “quick-fix” solutions. This mode may be mistaken for the Vulnerable Child mode due to what appears to be overt distress. It is hypothesized that this mode may be closely linked with Dependence, Emotional Deprivation, and Subjugation schemas (Simpson, 2016).

The purpose of this study was to develop a psychometrically sound adapted version of the Schema Mode Inventory in order to facilitate more precise measurement of mode states within a population with self-reported disordered eating behaviors. The result, a new Schema Mode Inventory for Eating Disorders (SMI-ED) would contribute to the ST conceptualization of schema modes in this sample. More specifically, this study aimed to conduct a Confirmatory Factorial Analysis (CFA) in order to examine the structure and psychometric properties of the SMI-ED, and to determine the internal reliability of its subscales. The relationship between eating disorder symptoms (restraint, binge-eating, and purging) and schema modes was also explored. In addition, we investigated whether specific EDE-Q subscales discriminated schema modes.

Item analysis revealed a non-perfect normal distribution, with Kolmogorov-Smirnov and Shapiro-Wilk tests being significant (p < 0.001). Skewness ranged between −1.14 and 3.87 (mean_sk = 0.23, SD_sk = 0.78; mean_sk/std.err = 2.24; SD_sk/std.err = 7.61), and kurtosis ranged between −1.47 and 18.1 (mean_k = −0.22, SD_k = 2.10; mean_k/std.err = −1.05; SD_k/std.err = 10.32).

Results from the CFA suggest an acceptable sixteen-related-factors solution for the SMI-ED, despite not all the fix indexes achieving the desired value (Hu and Bentler, 1999). Indeed, the chi-square model for fit was statistically significant [χ(17645)2 = 38210.465; p < 0.001] whilst the Comparative Fit Index value did not meet the threshold suggesting ideal fit [CFI > 0.90; (Hu and Bentler, 1999): CFI = 0.767]. However, the RMSEA showed a good approximation fit of the model to the data [RMSEA = 0.045 (90% CI: from 0.044 to 0.046), p (RMSEA < 0.05)= 1]. Moreover, by dividing the χ² for the degrees of freedom (df) of the model (Jöreskog, 1969; Hu and Bentler, 1999), the proposed model is potentially acceptable (χ²/df = 2.16; <3) (Hoyle, 2012). Finally, the SRMR further suggested the goodness of model fit [SRMR = 0.077 (Hu and Bentler, 1999)].

As shown in Table 2 each item loaded significantly on the factor associated with it (p < 0.001), mean_loadings = 0.727; SD_loadings = 0.110; ranging from 0.350 (item #117) to 0.905 (item #85). In addition, correlations between the sixteen factors ranged from 0.082 to 0.838; mean_r−factors = 0.258; SD_r−factors = 0.404 (Table 3).

The scales had acceptable internal consistency, with coefficients ranging from 0.807 to 0.976; mean_α−factors = 0.914; SD_α−factors = 0.048. Specifically, the highest value was registered for the Punitive Mode subscale (0.976), followed by the Vulnerable Child (0.956), Impulsive Child (0.950), Angry Child (0.948), Demanding Mode (0.947), Eating Disorder Overcontroller (0.946), Happy Child (0.933), Detached Protector (0.932), Compliant Surrender (0.926), Enraged Child (0.915), Undisciplined Child and Healthy Adult (0.908), Self Aggrandizer and Bully and Attack (0.883), Helpless Surrenderer (0.808), and Detached Self-Soother (0.807).

This study tested the psychometric properties of the SMI-ED, an adapted version of the original SMI, and investigated its psychometric properties within a population characterized by disordered eating patterns. Findings indicated an adequate fit for a 16-factor model, with moderate intercorrelations between subscales, and high internal consistency within subscales.

Although the chi-square model for fit and the CFI values resulted in a less than ideal fit, it must be highlighted that fit indices are affected by the sample size [i.e., chi-square (Jöreskog, 1969; Tuker and Lewis, 1973; Bentler and Bonett, 1980; James and Mulaik, 1982; Kline, 2015)] and the number of considered indicators respectively [i.e., CFI (Hu and Bentler, 1998; Russell, 2002; Kenny and McCoach, 2003; Fan and Sivo, 2005; Iacobucci, 2010; Kline, 2015)]. Indeed, as suggested by Kenny and McCoach (2003), testing well-fitting models with too many indicators, may cause malfunctioning in the CFI index, but it should not be a cause for concern if the SRMR and RMSEA reveal a good approximation fit of the model to the data (Hu and Bentler, 1998, 1999; Russell, 2002; Kenny and McCoach, 2003; Fan and Sivo, 2005; Iacobucci, 2010; Kline, 2015).

The final SMI-ED comprised 190 items that were found to best represent and distinguish 16 schema modes within this sample (the Helpless Surrenderer-HS and the ED-Overcontroller—EDO modes were specifically added for adapting the tool to the EDs sample). The final 16 factors indicated 5 maladaptive child modes; 2 maladaptive internalized/introject modes; 7 maladaptive coping modes; and 2 healthy factors, and included 73 new items. This supports our suggestion that the original SMI may not adequately capture the mode presentations for EDs, and that they may be better represented by the inclusion of items more directly related to the experiences reported by this specific population. The potential utility of measuring schema modes using separate subscales is therefore confirmed. The addition of two new coping modes, as well as additional items added to existing mode scales within the SMI-ED are likely to be of clinical utility in facilitating identification of ED-specific behaviors that characterize specific modes. Although mode scale scores may remain relatively unaffected by the new items, the qualitative examination of specific high-scoring items is likely to be of clinical value.

In support of Lobbestael et al.'s (2010) original findings (Lobbestael et al., 2010), Vulnerable, Angry, Enraged, Undisciplined, Impulsive, and Contented Child modes and the two “parent” (or introject) modes retained some items from the original SMI, with the addition of some new items. The ED-Overcontroller Mode (EDO) was found to be differentiated in the SMI-ED, supporting the notion that what was previously identified as a clinically important concept forms a measurable and coherent mode (Simpson, 2012, 2016) and upholding previous findings (Brown et al., 2016). All of these were new items. This mode is overcontrolling of the self and/or others and uses over-control of the body and perfectionism, to overcompensate for an underlying sense of powerlessness, shame or guilt. This overcompensatory coping mode can lead to feelings of mastery, powerfulness and pseudo-control (Arntz, 2010; Simpson, 2012, 2016). Results indicated that the one new item and one adapted item in Self Aggrandizer mode clustered with the 9 items that were retained from this mode within the original SMI. This supports our clinical observations that this mode may take on a slightly different “flavour” to the Self Aggrandizer in the original SMI, as the emphasis is on extreme high standards and a sense of pride or even superiority, often with an excessive emphasis on shape, weight and thinness which is perceived to provide protection from further shame or criticism (Simpson, 2016). This is consistent with findings for those with Obsessive-Compulsive Personality Disorder, who in this mode experience themselves as superior to others who are perceived as less principled and scrupulous (Beck et al., 2004; Weertman et al., 2008; Bamelis et al., 2015). Two of the three new items proposed for the Bully and Attack mode (“I feel invincible” and “If I'm not constantly on guard people will take advantage of me”) did not have a high loading on this scale, and will therefore be removed from the shorter scale in the second study. This may be because they are both likely to function as passive mechanisms for defending themselves against possible attack from others, rather than overtly bullying others.

The Helpless (or “Aggrieved”) Surrenderer Mode included 5 new items, and all but one had high loadings on this factor. The lower loading on the item “I am very possessive and hold on to the people that are important to me” may be indicative of the pattern of passivity and resignation connected with this mode, whereby the person feels helpless to find direct ways of asking for support and connection from others. This mode is characterized by a pattern whereby others are expected to intuitively know what they feel and need, with a tendency to feel frustrated when this does not materialize (Simpson, 2016). This mode has some overlap with the Complaining Protector mode (Bernstein and van den Broek, 2009) and also perhaps the Self-Pity/Victim Mode (Edwards, 2012, 2015) in the sense that all of these modes are represented by a passive-aggressive style of communication and a tendency to externalize responsibility for getting one's needs met. In this mode, self-deprivation, and physical frailty become an indirect means of communicating a need for emotional support (e.g., “if I am physically unwell, they will see how much I need them”). The remaining coping modes (Compliant Surrenderer, Detached Protector, Detached Self-Soother) and Healthy Adult mode were parallel to those described in the original validation study (Lobbestael et al., 2010) with the addition of some new items.

Correlations between modes were generally high. As expected, correlations between Punitive and Demanding Parent modes and Vulnerable and Angry Child modes were high, reflecting the toxic effect of internalized critical messages in generating emotional distress. Detached Protector and ED-Overcontroller were also surprisingly highly correlated with Vulnerable Child mode, which may be an indication of mode-flipping, whereby one closely follows the other. Indeed, coping modes are generally utilized to keep Vulnerable child mode (and attendant distress) outside conscious awareness. There was a significant difference between disordered eating behaviors (EDE-Q subscales) on the majority of the SMI-ED subscales, with most indicating higher levels of restraint and eating concern compared with shape and weight concern (Table 5). Both “Restraint” and “Eating Concern” subscales of the EDE-Q discriminated the modes most representative of eating disorder populations. In the case of the “Parent” (Introject) modes “Restraint” discriminated these more significantly than “Eating Concern” subscale. The higher mean scores on the “Restraint” and “Eating Concern” subscales compared with the “Shape-”and “Weight-Concerns” subscales may be due to the high number of participants with high levels of restrictive behaviors in this sample (as indicated by endorsement of: restraint over eating, avoidance of eating, food avoidance, dietary rules and desire for an empty stomach). Indeed, “Restraint” remained significantly higher than “Eating Concern” in spite of the fact that it comprised less than half of the number of participants (compared with “Eating Concern”), further corroborating this result. “Restraint” represents an attitude and behaviors that are highly relevant across all eating disorders. It is the main symptom that perpetuates AN, precedes binges in bulimic disorders, and also functions as a compensatory behavior in BN. In addition, the mean BMI in the “Restraint” category was low (18.18), indicating the majority of participants were characterized by restrictive eating patterns.

As expected, in this clinical sample the majority of eating disorder symptoms were positively correlated with Internalized/Introject and Coping modes and negatively correlated with the Healthy modes. Three of the coping modes, Bully-Attack, Self-Aggrandizer and Helpless Surrenderer only showed low correlations with eating disorder symptoms. It may be that these modes are more relevant to maintenance cycles or to other co-morbid symptoms, rather than directly influencing specific ED symptoms. Alternatively, it seems plausible that as the majority of these modes are described for a wide range of disorders, they may have a stronger association with general psychopathology than specific symptoms associated with eating disorders. Future studies may explore the role of these modes further within the ED population, and investigate whether they are more relevant to specific ED diagnostic groups. However, there were moderate to high correlations between the majority of coping modes and restraint and eating/weight/shape concerns. This is consistent with previous studies which have highlighted the link between avoidant coping, numbness, feelings of “lightness” and even euphoria, and restrictive eating/starvation syndrome (Spranger et al., 2001; Wildes et al., 2010; Kaye et al., 2013; Brown et al., 2016). The fact that these coping modes emerged as significant factors is consistent with findings from previous studies (Talbot et al., 2015; Brown et al., 2016). Low correlations were identified between other specific ED symptoms (e.g., bingeing, vomiting, exercise) and schema modes, however correlations between global eating disorder ratings and schema modes were mostly moderate to high.

This study represents the first step in creating a psychometrically sound instrument for the assessment of schema modes in EDs, and provides greater insight for the conceptualization and treatment of this population. This measure was comprised of a hybrid of 117 items from the original SMI with 73 new eating-disorder-specific items relating to schema modes. Confirmatory Factor Analysis confirmed the sixteen-related-factors solution for the 190-item SMI-ED, thus providing preliminary evidence for the adequate reliability of the new measure. The adaptation of the SMI for those with disordered eating represents the first step in developing a tailor-made measure of schema modes for the ED population. The adapted SMI-ED provides a tool for identifying and exploring a wide range of mechanisms through which modes are manifested in this population. Further studies are warranted to further refine and shorten the SMI-ED, in order to improve its utility as a research and clinical tool with this population.

SS conceived the study and wrote the manuscript with support from GP and AR. GP also contributed to data interpretation, and AR performed the analysis. GM helped supervise the project. TS contributed to data collection and preparation. CM and JN revised the advanced versions of the manuscript. GC approved the final version of the manuscript.