Emotional expression and eating in overweight and obesity

ABSTRACT Objectives: To explore psychological factors associated with emotional eating and obesity in a sample of overweight and obese adults attending a weight management programme. Design: A cross-sectional quantitative research design. Methods: Participants (n = 97) completed the Emotional Eating Subscale of the Dutch Eating Behaviour Questionnaire, the Attitude towards Emotional Expression (AEE) scale and the mindful awareness observe subscale of the Kentucky Inventory of Mindfulness Skills scale. Clinical measures of body mass index (BMI) were also recorded. Results: Regression analyses revealed that AEE was a significant predictor of emotional eating (β = 0.59, p = .000). Control, the belief that emotions should be controlled (β = 0.39, p = .026) and the response to eat to diffuse emotion (β = 0.37, p = .045) were statistically significant predictors of BMI. Mediation analyses revealed that mindful awareness skills had a significant indirect effect on the relationship between AEE and emotional eating. Conclusions: Findings highlight the influence of AEE on emotional eating and body weight, thereby helping to validate recent developments in an affect phobia model of emotional eating. The authors highlight the prevalence of emotional eating in overweight and obese adults. The potential preventative role of mindful awareness skills may be limited. Validation of the model may be a useful framework for the development and implementation of future weight management interventions.

may contribute to the development of negative attitudes to emotional expression, which may disrupt the cognitive processing of emotional information leading to impaired ability to regulate emotion (Joseph, Williams, Irwing, & Cammock, 1994). Finnegan et al. (2014) recently found that validation of a child's emotional needs influenced the development of attitudes towards emotional expression in adulthood and that negative attitudes to emotional expression modulated emotional eating in a large non-clinical adult population. Other research has reported similar findings (Espeset, Gulliksen, Nordbø, Skårderud, & Holte, 2012;Ford, Waller, & Mountford, 2011;Haslam et al., 2012).

Emotional eating and body weight
Eating in response to positive emotional states is not regarded to be an 'obese' eating style ( van Strien, Donker, & Ouwens, 2016). However, emotional eating in response to negative emotional states has been associated with an increased consumption of energy-dense foods (Nguyen-Michel, Unger, & Spruijt-Metz, 2007;Torres & Nowson, 2007). However, it remains unclear if emotional eating directly impacts body weight. Previous studies have contended that emotional eating does not always result in weight gain (Allison & Heshka, 1993), and indeed, emotional eating is prevalent among populations with a healthy body mass index (BMI) (Fischer et al., 2007;Geliebter & Aversa, 2003). However, the consistency of these findings has been debated (Goldbacher, La Grotte, Komaroff, Vander Veur, & Foster, 2016); Nolan, Halperin, & Geliebter, 2010). A recent large-scale longitudinal study reported that emotional eating may play an important role in body weight (Keller & Siegrist, 2015) whereby high levels of emotional eating are associated with higher BMI. A large body of evidence has suggested that this behaviour may contribute to weight gain, and may prevent successful weight maintenance (Koenders & van Strien, 2011;Robbins & Fray, 1980;Torres & Nowson, 2007; Van Strien, Herman, & Verheijden, 2009). Further research is needed to determine whether or not there is a causal relationship between emotional eating and body weight.
Emotional eating has been associated with poorer weight loss outcomes and attrition in weight management interventions (Byrne, Cooper, & Fairburn, 2003;Canetti, Berry, & Elizur, 2009;Keränen et al., 2009). A number of studies have reported that behavioural weight management programmes are challenging and largely unsuccessful for individuals who engage in emotional eating (Cox, Zunker, Wingo, Jefferson, & Ard, 2011;Neve, Collins, & Morgan, 2010; Van Strien, Herman, & Verheijden, 2012). Such findings indicate that emotional eating may be a barrier to weight management success. A recent study found that three out of four overweight and obese adults enrolled in a weight management intervention self-reported emotional eating (Braden et al., 2016). However, few interventions assess the impact of emotional eating. This presents challenges to clinical practice.
Emotional eating and mindful awareness Weight management interventions need to provide strategies to increase awareness of behavioural patterns and decrease emotional eating (Foster, Makris, & Bailer, 2005). A recent randomised controlled trial found that decreased emotional eating was associated with greater weight loss (Braden et al., 2016).
Previous research has explored the relationship between trait mindfulness and body weight. Findings indicate that higher trait mindfulness may be inversely associated with weight gain, in particular, the observation component of mindfulness (Camilleri, Méjean, Bellisle, Hercberg & Péneau, 2015;Liebman et al., 2003;Mantzios & Wilson, 2014;Mantzios, Wilson, Linnell, & Morris, 2015). Finnegan et al. (2014) reported that the mindful awareness, observe subscale of the Kentucky Inventory of Mindfulness Skills (KIMS) scale (Baer, Smith, & Allen, 2004) highly moderated levels of emotional eating and BMI. Although participants are more likely to over-report height and underreport weight when self-report data are used to compute BMI (Gorber, Tremblay, Moher, & Gorber, 2007), most research has employed self-report measures of BMI. These studies require replication using clinical measures of body weight.

The present study
Efforts to increase our understanding of eating behaviour in obese and overweight populations are both pertinent and timely, given the increasing rates of obesity internationally. The aim of this present study is to build upon the work of Finnegan et al. (2014) and to assess the relationship between attitudes to emotional expression, mindful awareness skills, emotional eating and body weight among clinically overweight and obese adults attending a weight management programme. Using a cross-sectional quantitative approach, the hypotheses were as follows: (a) emotional eating is directly linked and positively associated with BMI; (b) healthy attitudes towards emotional expression are directly and inversely associated with emotional eating, and mediate the relationship between emotional eating and BMI; (c) Observe mindful awareness skills will directly moderate both emotional eating and BMI.

Participants
Participants were 101 adults attending a weight management programme at weight management clinics in Ireland. Adults clinically assessed as overweight or obese were recruited on a voluntary basis. The mean age of participants was 47.01 (SD = 14.65) with a range of 18-72 years. Gender distribution was 29 males (29.7%) and 68 females (70.3%). No participants identified themselves as transgender. BMI calculations indicated that 4% of participants in the sample were normal weight, 50.5% were overweight and 45.5% participants were obese. Healthy weight participants who did not meet the inclusion criteria were excluded from analysis resulting in a final sample of 97 participants. Of the remaining 97 participants, no participant was excluded in the case of missing data.

Sample size calculation
The sample size was in line with recommendations, which stipulate that a sample should include an empirical estimate of 89 participants for power values in mediation analyses (Fritz & MacKinnon, 2007), and an empirical estimate of 74 for in multiple regression analyses (Green, 1991;Tabachnick & Fidell, 2013).

Design
The researchers utilised a cross-sectional quantitative design.

Clinical measures
Clinical measures of participant body weight were taken by researchers.

BMI
BMI was used to operationalise body weight. BMI was calculated using the Tanita Body Composition Analyser, a non-invasive procedure. The recommended categorisation system was used to classify participants into body weight categories based on BMI measurements: healthy weight = 18.5-24.9 kg/m 2 ; overweight = 25-29.9 kg/m 2 ; obesity ≥ 30 kg/m 2 .

Self-report measures
Participants were asked to complete a questionnaire comprised of a number of validated measures and questions on demographics. The questionnaire was designed to be brief, as research has indicated that short questionnaires improve response rate (Edwards et al., 2002).

Emotional eating subscale of the Dutch eating behaviour questionnaire
The Emotional Eating Subscale of the Dutch Eating Behaviour Questionnaire (DEBQ-em) is a 13-item scale which uses a five-point Likert scale to operationalise emotional eating ( Van Strien, Frijters, Bergers, & Defares, 1986). The subscale assesses two aspects to emotional eating. One assesses eating behaviour in response to the diffusion of emotion, and the other assesses eating behaviour in response to clearly defined emotions (Strien, 2002). The measure has been found to have strong factorial validity and convergent reliability across a variety of populations, including populations attending weight management programmes (Wardle, 1987). The DEBQ-em has demonstrated satisfactory psychometric properties (Barrada, Strien, & Cebolla, 2016), and high factorial validity (van Strien, Herman, Anschutz, Engels, & de Weerth, 2012), and internal consistency with a Cronbach alpha coefficient value of 0.94 ( Van Strien et al., 1986). Cronbach's alpha for the present study was 0.98 indicating good internal consistency.

Observe subscale of the KIMS scale
The Observe subscale of the KIMS is a 12-item scale which uses a five-point Likert scale to operationalise mindful awareness skills (Baer et al., 2004). Mindful awareness skills have been described in the literature as skills in internally and externally observing stimuli (Dimidjian & Linehan, 2003;Kabat-Zinn, 1990). These skills were of interest to the research given the inverse association between alexithymia, the sub-clinical inability to identify and describe emotions about the self, and mindful awareness (Baer et al., 2004;Baer et al., 2006). The observe scale has high content validity and adequate to good test-retest reliability with a Cronbach's alpha coefficient value of 0.91 reported (Baer et al., 2004). Cronbach''s alpha for the present study was 0.93 indicating strong internal reliability.

Attitudes towards emotional expression scale
The Attitudes towards Emotional Expression (AEE) scale is a 20-item scale which uses a five-point Likert scale to operational attitudes towards emotional expression (Laghai & Joseph, 2000). The AEE assesses four dimensions to attitudes to emotional expression: the view that emotions should be controlled (control). The view that the expression of emotion is a sign of weakness (weakness); the tendency to hide the expression of emotions to others (non-expression); and the view that emotional expression will lead to rejection from others or damage others (social) (Joseph et al., 1994). The AEE scale has shown good internal consistency and convergent validity with a Cronbach alpha coefficient value of 0.93 reported (Joseph et al., 1994;Laghai & Joseph, 2000;Meyer et al., 2010). Cronbach's alpha for the present study was 0.97, indicating good internal consistency.

Ethical considerations
The research adhered to the British Psychological Society (BPS) code of ethical standards for human research (BPS, 2009).

Procedure
An online version of the questionnaire was developed using Survey Monkey (www. surveymonkey.com). The questionnaire was piloted at one weight management clinic, and adjusted to improve questionnaire readability. Recruitment took place in clinics in the Republic of Ireland. Participants were provided with written information detailing the nature of the study. Willing participants were asked to give written consent and complete the online questionnaire using the iPads provided. Administrators took measurements of participant BMI. Participants were later provided with debriefing information and thanked for their participation. Data were collected between 17 February 2016 and 16 June 2016. Data were stored securely online with Survey Monkey and later stored on an external hard drive. The study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Guidelines (Appendix) (Von Elm et al., 2007).

Data analysis
The data were analysed using the Statistical Package for the Social Science (SPSS), version 22 and PROCESS macro version 2.16 (Hayes, 2015). Little's missing completely at random test was executed to address missing data. Descriptive statistics were executed for all variables. Bivariate correlations were executing using Pearson's Product Moment correlation coefficient, to assess the relationship between all variables. A hierarchical regression was conducted to examine an affect phobia model of emotional eating. A mediation analysis tested the hypothesis that attitudes towards emotional expression mediate the relationship between emotional eating and body weight.

Preliminary data analyses
The data were screened and preliminary analyses were performed to ensure no violation of the assumptions of normality, linearity and homoscedasticity. A reliability analysis was completed for each of the study's scales. A missing value analysis was completed. Little's missing completely at random test found no significance, indicating that data were missing at random χ 2 = 29.48, df = 41, p =.91.
Descriptive statistics are presented in Table 1. The mean score for the emotional eating subscale DEBQ-em was 43.84 (SD = 15.18). Inspection of the distribution of scores indicated a skewness towards the higher end of the scale, suggesting higher than average scores of emotional eating. High scores suggested that the population were highly likely to emotionally eat. The mean score for the AEE scale was 58.89 (SD = 21.51). Visual analysis of the scale scores was quite flat, yet did not meet the assumptions of platykurtosis, indicating a number of extreme cases. Overall, the distribution indicated neutral to negative attitudes to emotional expression. The mean score for the Observe subscale was 33.02 (SD = 9.83). This score indicated that the sample had normally distributed mindful awareness skills. The relationship between all variables was assessed in a correlation matrix using Pearson's product-moment correlation coefficient (see Table 2). There was a strong negative correlation between AEE and Observe (r = −0.56, p < .01), and between the DEBQ-em and Observe (r = −0.52, p < .01). A strong positive correlation was also found between the DEBQ-em and AEE scales (r = 0.69, p < .01); however, multicollinearity was not observed. Gender, age and BMI scores were not strongly correlated with other variables.
A hierarchical regression analysis was employed to evaluate how well age, gender, BMI, AEE and mindful awareness skills predicted emotional eating. Age and gender were controlled for in Step 1, explaining 8% of the variance in emotional eating. BMI was controlled for in Step 2, explaining 7% of the variance in emotional eating. AEE was entered in Step 3, explaining 39% of the variance. Observe was entered in Step 4, explaining 1% of the variance in emotional eating. The overall model was significant (F (1, 88) = 14.25, p = .000, R2 = 0.56, Adj R2 = 0.53). Only AEE was a statistically significant predictor of emotional eating (β = 0.59, p = .000). The regression analyses were then extended to evaluate how well each of the AEE subscales predicted emotional eating (see Table 3.). The model was significant (F (1, 88) = 14.25, p = .000, R2 = 0.56, Adj R2 = 0.53). The analysis revealed that only the subscale, weakness, the view that the expression of emotion is a sign of weakness, was a significant predictor of emotional eating (β = 0.31, p = .048).
A further hierarchical regression analysis was employed to evaluate how well age, gender, the AEE subscales, the DEBQ-em subscales and Observe skills predicted BMI  (see Table 4.). Age and gender were controlled for in Step 1, explaining 2% of the variance in BMI. The four AEE subscales were controlled for in Step 2, explaining 11% of the variance in BMI. The two DEBQ-em subscales were entered in Step 3, explaining 5% of the variance. Observe was entered in Step 4, explaining 2% of the variance. The overall model was significant (F (1, 87) = 2.68, p = .009, R2 = 0.22, Adj R2 = 0.14).
Only the AEE subscale, control (β = 0.39, p = .026), and the DEBQ-em subscale diffuse (β = 0.37, p = .045) were statistically significant predictors of BMI. A mediation analysis was used to test the hypothesis that attitudes towards emotional expression mediate the relationship between emotional eating and BMI (see Figure 1). The assumptions of mediation analyses were violated and no significant indirect effect was found (see supplementary online material 1). A post hoc analysis revealed a significant mediation effect of AEE on emotional eating through Observe skills (see Figure 2)

Discussion
The present study aimed to explore the direct and indirect effects of a number of previously identified variables on emotional eating and BMI (Finnegan et al., 2014) in a  clinically overweight and obese population attending a weight management programme, using objective clinical measures of body weight. Emotional eating and BMI were the principal outcome variables. As hypothesised, emotional eating directly, positively and significantly predicted BMI. The DEBQ-em subscale diffuse, where eating is used as a defence to diffuse negative emotions had a significant effect on BMI. As hypothesised, attitudes towards emotional expression were significantly, directly and positively associated with BMI, where highly negative attitudes predicted higher BMI. The subscale control had the most significant effect of all AEE subscales on BMI. As hypothesised, attitudes towards emotional expression significantly, directly and positively predicted emotional eating. The subscale weakness was the most significant predictor of emotional eating. Mindful awareness skills were not predictive of BMI or emotional eating. However, exploration of these skills on attitudes towards emotional expression had significant indirect effects on emotional eating. Exploration of attitude towards emotional expression (AEE) on emotional eating did not reveal a significant indirect effect on BMI, as had been hypothesised. Age and gender did not significantly predict emotional eating or BMI.

Findings in context
The finding that emotional eating had a significant positive effect on BMI supports previous findings that human emotion is central to eating and disordered eating behaviour (Canetti et al., 2002;Polivy & Herman, 2002). This supports the theory that emotional eating is an atypical response to autonomic nervous system arousal (Topham et al., 2011;Torres & Nowson, 2007;Van Strien et al., 2013), and adds credence to the growing body of evidence that stress may contribute to obesity (Holmes, Ekkekakis, & Eisenmann, 2010;Kupeli et al., 2017). Similar to previous research, high levels of emotional eating were associated with higher BMI (Goldbacher et al., 2016;Nolan et al., 2010). This supports the findings that emotional eating may be a causal factor in the development of obesity (Robbins & Fray, 1980;Van Strien et al., 2009), and therefore increase the risk of development of a number of chronic health conditions (WHO, 2015). Self-reported emotional eating was higher than average. However, participants were attending a cognitive behavioural therapy (CBT)-based weight management programme, where an individual with disordered eating may be more likely to attend. This may explain some of the variance in this finding.
The finding that AEE had a significant effect on emotional eating and BMI helps validate a review of affect regulation theory in obesity, which suggests that negative emotions become triggers for eating (Leehr et al., 2015). This finding is in keeping with the findings of previous research (Espeset et al., 2012;Finnegan et al., 2014;Ford et al., 2011;Haslam et al., 2012;Meyer et al., 2010). The belief that emotion is a sign of weakness had the most significant effect of all AEE subscales on emotional eating. This was supported by the previous findings of similar studies with non-clinical samples of women (Haslam et al., 2012;Meyer et al., 2010). This finding further supports an affect phobia model of emotional eating, where emotions are invalidated in the childhood (Corstorphine, 2006). Interestingly, the belief that emotions should be controlled had the most significant effect on BMI. Previously, control has been associated with eating pathology and weight concern (Meyer et al., 2010).
Surprisingly, however, AEE did not mediate the relationship between emotional eating and BMI. This suggests that AEE does not explain the relationship between emotional eating and BMI. Other potential factors may mediate the relationship. For example, previous research highlights the role of negative emotional states (Van Strien et al., 2013), while others have highlighted the role of self-compassion . Ubiquitously it is acknowledged that obesity is heterogeneous in its aetiology (Marcus & Wildes, 2009).
Unexpectedly, mindful awareness skills were not predictive of emotional eating. These skills have been linked with awareness in the experience of the here and now (Dimidjian & Linehan, 2003;Kabat-Zinn, 1990). The present study's finding is contrary to previous research, where greater mindful awareness skills have been associated with reduced automatic reactivity, and enhanced regulation in behaviour (Keng et al., 2011). However, these skills were found to mediate the relationship between attitudes towards emotional expression and emotional eating, indicating that mindfulness may have some role to play in emotional eating (Alberts, Thewissen, & Raes, 2012;Mason et al., 2016). This finding is similar to a recent study, which found that mindful observation modulates the link between motivational states and traits on eating behaviour in response to appetitive stimuli (Papies, Pronk, Keesman, & Barsalou, 2015). The present finding supports previous findings that mindful awareness skills may be inversely associated with alexithymia (Baer et al., 2004;Baer et al., 2006). Furthermore, the present study supports previous research, where alexithymia has previously been linked to emotional eating (Pinaquy, Chabrol, Simon, Louvet, & Barbe, 2003;Pink, Williams, & Lee, 2016;Van Strien & Ouwens, 2007), particularly in severe obesity (Noli et al., 2010). The finding that mindful awareness skills were not predictive of BMI score was unexpected. However, a small cohort of studies have also reported this (Finnegan et al., 2014;Kearney et al., 2012).
Age and gender did not appear as significant predictors of emotional eating or BMI, contrary to previous research. Prior research indicated that young adults and females have a greater tendency to emotionally eat (Cornelis et al., 2014;Finnegan et al., 2014;Gibson, 2012). In a similar vein, older adults and men tend to have higher BMI scores (Okorodudu et al., 2010). However, it is noted by the authors that the present findings may be unique to a homogeneous sample of overweight and obese adults attending a CBT-based weight management programme.

Theoretical contribution: an affect phobia model of emotional eating
The current study's findings support an affect phobia model of emotional eating (Finnegan et al., 2014;McCullough, 2003). The model suggests that childhood environments lead to the development of negative attitudes to emotional expression, where negative emotions may be perceived negatively, and subsequently avoided or controlled. This experience may result in internal conflict in an individual, where defences such as emotional eating are adopted to decrease emotional distress. Such disruption to the development of emotions may lead to the sub-clinical inability to identify and describe emotions about the self, namely alexithymia (Pink et al., 2016). Eating in response to the internal experience of negative emotions may lead to increased body weight in the context of chronic stress.

Strengths and limitations
To the knowledge of the authors, this study was novel in that previous studies have not explored the variables of interest in clinically overweight and obese populations attending weight management programmes using clinical measures of body weight. The use of a clinical measure of BMI in operationalising body weight was a major strength in the study's design. A major methodological weakness of previous studies was the use of self-report for height and weight (Gorber et al., 2007). Another strength of the study was the adherence to the STROBE guidelines (Von Elm et al., 2007), which improved the study's rigour.
Despite its strengths, this study was limited by the use of a self-report measure of emotional eating which may not have truly reflect eating behaviour (Evers, Stok, & de Ridder, 2010). However, this limitation has been contested ( van Strien, 2010). Another concern is that the study took place in the weight management clinic setting, which may have influenced participants' responses. In addition, the voluntary nature of recruitment must be acknowledged as it may reduce the generalisability of the study's findings due to a volunteer bias, where certain demographics are less likely to take part in voluntary studies (Heiman, 2002). Furthermore, participants were made aware of the nature of the study through its advertisement in clinics. Arguably the participants may have been selfselecting as interested in emotional eating as a factor related to their obesity. Moreover, participants were recruited from a CBT-based weight management programme, which may have increased the likelihood of a sampling bias. While the gender ratio was notably more balanced than in previous research, there exists a gender imbalance in participant recruitment. This is a limitation to the generalisability of the study's findings. This is a common phenomenon in the literature (Espeset et al., 2012;Finnegan et al., 2014;Ford et al., 2011;Haslam et al., 2012;Meyer et al., 2010). These potential biases are acknowledged as threats to the study's external validity.
Furthermore, while the variables of interest were grounded in the literature, it is highly likely that other underlying variables were present. The authors were aware that it remains unclear if emotional eating causally determines body weight, as it is highly considered that a number of uncontrolled for variables occur simultaneously. Finally, the study employed a cross-sectional design, whereby the expected mediation effect would have been difficult to interpret as causality cannot be assumed and the findings are ultimately limited.

Clinical implications of findings
These limitations are unsurprising given the exploratory nature of the research. However, the findings may have important implications for clinical practice. Despite its limitations, the present study highlights the importance of challenging negative attitudes to emotional expression and the development of adaptive emotion regulation strategies in populations with high levels of disordered eating in clinical settings. Given the growing global obesity trend (NCD Risk Factor Collaboration, 2016) and the strong association between emotional eating and BMI, the present study may have implications for an increased risk of the development of a number of chronic health conditions (WHO, 2015). Future weight management interventions should assess disordered eating as part of evidence-based practice.
Emotional eating is an important risk factor for overweight and obese adults in general, and in the context of weight management interventions. Researchers and clinicians should acknowledge the vulnerable nature of this population. Findings call for the implementation of CBT for emotional eating and improved access to psychological therapies (Murphy, Straebler, Cooper, & Fairburn, 2010).

Future research
Emotional eating is an important area for further public health research. Replication of the present study's findings is recommended. In addition, it is recommended that future studies employ longitudinal research designs and explore gender balanced populations to increase the generalisability of findings. Future research using larger sample sizes could further examine the phenomenon by comparing differences between overweight and obese adults. Similarly, future research could refine the present model by exploring novel variables which may be likely to contribute to emotional eating (e.g. mental health, attitudes towards emotional eating and attitudes towards obesity in the obese). Furthermore, qualitative research would add to health psychology's understanding of the phenomena and perhaps identify lurking variables.

Summary and conclusions
The present study highlights the prevalence of emotional eating in overweight and obese populations. The authors found emotional eating to be an important contributing factor to body weight; however, it is acknowledged that causality cannot be assumed. The research indicated that attitudes towards emotional expression play are central to emotional eating, thereby helping to validate recent developments in the affect phobia model of emotional eating. Validation of the model may provide future researchers and clinicians with a framework for the development and implementation of weight management interventions. Mindful awareness skills were limited in their contribution to BMI and emotional eating, however, these skills did help explain the relationship between AEE and emotional eating. This finding supports previous research investigating disordered eating and alexithymia and indicates that mindful awareness may have some contribution to interventions. It is important to emphasise that the findings reported require replication and validation.

Disclosure statement
No potential conflict of interest was reported by the authors. (b) Give reasons for non-participation at each stage n/a (c) Consider use of a flow diagram n/a Descriptive data 14* (a) Give characteristics of study participants (e.g. demographic, clinical, social) and information on exposures and potential confounders