Rethinking Emotional Eating: Retrospective and Momentary Indices of Emotional Eating Represent Distinct Constructs

Abstract

Emotional eating (EE) has been proposed as a key weight loss barrier. However, most investigations of EE rely on retrospective self-reports, which may have poor construct validity. This study evaluated concordance between a common self-report EE measure and a novel method for assessing momentary EE using ecological momentary assessment (EMA). We further assessed the utility of both measures for predicting both BMI and weight outcomes. Participants with overweight or obesity enrolled in a weight loss trial (N = 163) completed a self-report measure of EE and underwent an EMA protocol that assessed momentary emotions and eating behaviors. Momentary EE was derived from EMA data using generalized linear mixed-effects models. Linear regression models examined associations between both EE measures and concurrent BMI as well as weight losses over 30 months. Retrospectively self-reported EE and momentary EE were negatively correlated with one another (r = −0.27). Higher momentary EE and higher retrospectively reported EE both predicted higher concurrent BMI, and higher retrospectively reported EE predicted poorer weight loss outcomes at all time points (p < 0.05). By contrast, higher momentary EE predicted improved weight outcomes at 1-year and 2-year follow-up (p < 0.05). Our findings extend prior research suggesting that retrospective self-report EE measures capture a different construct than intended and suggest that momentary EE could predict improved weight loss outcomes.

1. Introduction

Obesity is a major public health issue facing most industrialized countries, and rates of obesity continue to rise (Centers for Disease Control and Prevention, 2020). For decades, emotional eating (EE); i.e., eating or overeating in response to negative emotions (Greeno & Wing, 1994; Macht, 2008) has been proposed as a key contributor to excess weight (Bruch, 1964; Greeno & Wing, 1994). An estimated 57–90% of adults with overweight or obesity report high levels of EE (Ganley, 1989; Gibson, 2012; Péneau et al., 2013), and greater EE is associated with preference of high energy-density foods (Nguyen-Michel et al., 2007) and adverse health outcomes, including greater body mass index (BMI) (Anglé et al., 2009; Péneau et al., 2013) and weight gain over 1–2 years (Dohle et al., 2014; Koenders & van Strien, 2011; Sung et al., 2009). In addition, EE appears to pose a barrier to weight loss (Frayn & Knäuper, 2018), making it difficult for people to refrain from overeating following negative emotions. For example, greater baseline EE has been found to predict less weight loss after a year-long treatment period (Teixeira et al., 2010), and poorer likelihood of achieving 7% weight loss following six months of treatment (though not at post-treatment or follow-up) (Delahanty et al., 2013). Importantly, reductions in EE during treatment appear to predict improved weight loss outcomes (Braden et al., 2016; Butryn et al., 2009), indicating that EE may be a modifiable intervention target.

While the existing literature appears to support a key role for EE in overweight and obesity, it largely relies upon retrospective self-report measures to assess overall EE tendencies (“retrospective EE”), the validity of which have been called into question (Bongers & Jansen, 2016; Braden et al., 2020). Common retrospective EE measures (e.g., the Dutch Eating Behavior Questionnaire (Van Strien et al., 1986), the Emotional Eating Scale (Arnow et al., 1995) and the revised version of the Three Factor Eating Questionnaire—R18 (Karlsson et al., 2000) consist of face-valid items asking participants to self-report their propensity to eat or desire food after experiencing various emotions. From a theoretical standpoint, such measures may be biased, such as by one’s current emotional state, or may measure a different construct than intended. For example, individuals often misattribute the cause of their behavior (Gantman et al., 2017) and this misattribution may extend to the realm of overeating. In a study by Adriaanse et al. (2016), for instance, after completing a bogus taste test, participants were randomly assigned to a condition in which they were either informed that they had eaten an appropriate amount or that they had overeaten. Afterward, participants retrospectively self-reported their general EE tendency and how they felt prior to the taste test. Compared to participants low in retrospective EE, self-proclaimed emotional eaters attributed their supposed overeating more strongly to prior negative affect, despite actual levels of negative affect and quantity of food consumed not differing between conditions. These results suggest that those high in retrospective EE believe that their overeating occurs in response to negative emotions, regardless of whether this is actually the case.

Further calling the validity of retrospective EE measures into question, researchers have found equivocal support for concordance between retrospectively-assessed EE and EE assessed using different modalities. In experimental tests of EE, for example, retrospective EE is sometimes (e.g., (Van Strien et al., 2012), though often not (Bongers & Jansen, 2016; Braden et al., 2020), related to various laboratory-assessed EE measures, in which food intake is assessed in a bogus taste test following negative mood inductions (e.g., mood-inducing film clips and/or music, the recollection of negative memories, or false feedback following a stressful task). Similarly, in prior research there has been mixed support for a relationship between retrospective EE and EE assessed in daily life utilizing daily diary paradigms (in which participants reported on the behaviors and experiences at the end of the day), including amongst individuals with overweight/obesity (Brogan & Hevey, 2013; Conner et al., 1999; Newman et al., 2007; O’Connor et al., 2008). If retrospective EE accurately captures real-world EE, then one would expect a greater prospective relationship between negative emotions and subsequent overeating in daily life amongst individuals self-reporting higher retrospective EE. While some researchers have found support for such a relationship (O’Connor et al., 2008), others have not (Adriaanse et al., 2011; Brogan & Hevey, 2013).

Prospective prediction of eating is important to assess because EE is an inherently temporal construct—elevations in affect are proposed to lead to subsequent (over)eating. One ideal method for accomplishing such prospective prediction is ecological momentary assessment (EMA), a methodology which assesses psychological, cognitive, and behavioral phenomena as they occur in the moment during one’s typical daily routine (Stone & Shiffman, 1994). Like daily diary assessments, EMA captures EE naturalistically. EMA has gained popularity within behavioral medicine and the study of eating behavior (Engel et al., 2016; Haedt-Matt & Keel, 2011) for its enhanced ecological validity and ability to decrease recall bias. Further, and of relevance to the assessment of emotional eating, EMA can capture temporal associations between constructs. That is, by repeatedly sampling experiences and behaviors throughout a day, over several days, researchers can examine whether elevations in momentary self-reports (e.g., affect) precede, or increase the likelihood of, subsequent events (e.g., overeating).

However, despite the strong rationale for measuring EE (a temporal construct) with EMA (a temporal measurement method), limited research has examined whether emotions prospectively predict self-reported eating behaviors in daily life (Boh et al., 2016; Forman et al., 2017; Goldschmidt et al., 2014). One study by Goldschmidt and colleagues (2014) used EMA to examine the relationships between internal experiences (including emotional states) and eating behaviors amongst adults with overweight/obesity. Negative affect was not consistently elevated prior to or following eating or overeating, casting some doubt on the hypothesis that emotions at a given timepoint are a key driver of subsequent overeating amongst individuals with overweight/obesity.

To our knowledge, only one study (from which three papers have been published) has examined the prospective relationship between momentary emotions and overeating in a weight loss sample (Forman et al., 2017; Goldstein et al., 2018; Manasse et al., 2018). In this study, Forman and colleagues utilized EMA to examine whether various internal experiences predicted the likelihood of a subsequent dietary lapse from a weight control diet amongst adults with overweight/obesity. A dietary lapse was defined as eating or drinking likely to cause weight gain, and/or put weight loss/maintenance at risk. Consistent with the findings of Goldschmidt and colleagues (2014), within-person fluctuations in emotional states (i.e., elevations in emotions compared to a participant’s average level of emotion) did not affect dietary lapse likelihood, further calling into question the construct validity of retrospective EE.

Yet, no research, to our knowledge, has evaluated the concordance between retrospective EE measures and EE assessed in a momentary way, i.e., using EMA paradigms. EMA, which is capable of multiple, real-time assessments of emotions and eating, is an ideal method of capturing “momentary EE” (i.e., the degree to which negative emotions prospectively predict likelihood of subsequent dietary lapse/overeating episode). Given that momentary EE measurement approaches are more consistent with the conceptualization of EE as a temporal construct (i.e., momentary fluctuations in emotional states preceding actual events of overeating), an investigation of the concordance between retrospective and momentary EE could improve our understanding of whether retrospective self-report EE measures actually capture EE. As such, in the present study we examined the concordance between retrospective EE and a novel EMA-derived approach to capture momentary EE amongst weight loss-seeking adults. We examined both overall levels of EE (i.e., EE following negative emotions, generally), and EE for three specific emotions (loneliness, anxiety, and sadness) that participants reported on in both the retrospective EE measure and the EMA protocol administered in the current study. We hypothesized that retrospective EE would be positively associated with momentary EE for both overall levels of EE and for EE corresponding to specific emotion items, though we hypothesized that this association would be modest, given relatively weak relationships observed between retrospective EE and laboratory EE in prior research (Bongers & Jansen, 2016).

Further, we examined the relationship between retrospective EE and momentary EE and weight and weight loss outcomes. First, we examined associations with current BMI, hypothesizing that greater retrospective and momentary EE would be associated with greater BMI. The aims of these analyses were to evaluate the construct validity of retrospective EE measures (i.e., by determining whether retrospective EE was associated with BMI in a similar manner to momentary EE). Next, we examined the prospective relationships between both EE measurement approaches and percent weight loss (PWL) at post-treatment and at 1-year and 2-year follow-up. The aims of these analyses were to further evaluate the construct validity of retrospective EE measures, and also to assess the predictive clinical utility of each approach (i.e., the ability to identify which individuals may fare better or worse during treatment). Consistent with past research, we hypothesized that greater momentary and retrospective EE would be associated with less PWL at post-treatment and follow-up, for both overall levels of EE and for EE corresponding to individual emotion items (loneliness, anxiety, sadness). All hypotheses were specified prior to data analysis.

2. Methods

2.1. Participants

One hundred eighty-nine adults with overweight or obesity (BMI 27–50 kg/m²) were recruited from the Philadelphia area to participate in a randomized trial (registered on ClinicalTrials.gov, identifier CT00746265) comparing the efficacy of standard behavioral treatment (SBT) and acceptance-based behavioral treatment (ABT) for weight loss, the primary outcomes of which have been published previously (Forman et al., 2016; Forman et al., 2019). All participants provided written informed consent during the enrollment process. For the present secondary analyses, only mid-treatment data (6-months into treatment) were used for both EE measures. This decision was made because at baseline, retrospective EE was assessed before the start of treatment, whereas EMA data was collected after the start of treatment, a potential confound. The sample at mid-treatment (N = 163) had a mean age of 52.45 (SD = 9.82). 83.8% of participants were women and 70.4% were White, 24.7% were Black/African American, 3.5% were Non-White Hispanic and 1.4% were Asian. The mean BMI for the sample at mid-treatment was 32.58 (SD = 6.17), ranging from 22.41 to 48.11.

2.2. Procedures

2.2.1. Intervention and Assessment Visits

Participants were randomized to receive one of two behavioral weight loss treatments (BT) consisting of 25 group sessions across 12 months. Both treatment conditions (SBT and ABT) included behavioral components similar to the Diabetes Prevention Program (DPP) protocol, i.e., daily self-monitoring of calorie intake, prescriptions for balanced-deficit diet, and stimulus control (Diabetes Prevention Program Research Group, 2002). In addition, the ABT condition integrated material from acceptance and commitment therapy (ACT) (Hayes et al., 2011) and dialectical behavioral therapy (DBT) (Salsman & Linehan, 2006), including (1) clarification of how freely chosen, personal life values can motivate persistence in weight control efforts, (2) willingness to accept discomfort or reduction of pleasure and (3) awareness of how internal and external cues affect eating.

Participants completed a baseline, mid-treatment (6-month), and post-treatment (1-year) assessment as well as 1- and 2-year follow-up assessments in the laboratory. In this paper, we present a secondary data analysis using data collected at mid-treatment, post-treatment, 1-year and 2-year follow-ups. During the mid-treatment assessment (the focus of our EE concordance analyses) participants were weighed and completed psychological measures. Participants then completed an EMA protocol during the two weeks following mid-treatment. Drexel University’s Institutional Review Board approved all study procedures.

2.2.2. EMA Protocol

Participants were provided Android players (Samsung Galaxy Player 4.0) with custom EMA applications at mid-treatment. Consistent with prior EMA studies (Forman et al., 2017; Goldschmidt et al., 2014; Mason et al., 2018), the week-long EMA protocol contained event-contingent surveys (i.e., participants were instructed to report any instance in which they experienced a dietary lapse, which could include eating a larger meal or snack than intended, eating when not intended, or eating a type of food that the participant intended to avoid) and signal-contingent surveys at semi-random prompts intervals across waking hours (within 30 minutes of 9:30am, 12:00pm, 2:30pm, 5:00pm, 7:15pm, 9:30pm). EMA surveys asked participants to report whether a dietary lapse had taken place since the previous survey, as well as the degree to which they were experiencing various internal states (e.g., negative affect, hunger) and external cues (e.g., the presence of tempting foods). If participants did not respond immediately to a prompt, they received a reminder to complete their survey once every 5 minutes until either survey completion or until the prompt expired (after 45 minutes). Participants lost $1 from $42 total possible compensation for each missed survey. The EMA protocol was designed to examine momentary predictors of dietary lapses (Forman et al., 2017). For the present analyses, we recognized that we could exploit items within the EMA protocol (namely, negative emotions predicting dietary lapses), to assess emotional eating. Furthermore, dietary lapses (eating/drinking that could put a calorie goal/weight control at risk) should, in the context of a weight loss sample, specifically capture the patterns of eating (indulgent-, unplanned-, and/or over-eating) thought to characterize emotional eating (Adriaanse et al., 2011; Bennett et al., 2013; Kemp et al., 2013; Kuijer & Boyce, 2012).

2.3. Measures

A summary of EE measurement approaches examined in the current study, and their definitions, is presented in Table 1.

Table 1.

Summary of emotional eating measurement approaches examined in the current study.

Emotion	Retrospective Measure	Momentary Measure
Sadness	Retrospective Sadness EE: Score on the “blue” item of the emotional eating (EE) subscale of the Three- Factor Eating Questionnaire	Momentary Sadness EE: EMA-derived coefficients representing an individual’s tendency to lapse from a weight control diet following momentary elevations in sadness, relative to the individual’s average level of sadness, across timepoints.
Anxiety	Retrospective Anxiety EE: Score on the “anxious” item of the emotional eating (EE) subscale of the Three- Factor Eating Questionnaire	Momentary Stress EE: EMA-derived coefficients representing an individual’s tendency to lapse from a weight control diet following momentary elevations in stress, relative to the individual’s average level of stress, across timepoints.
Loneliness	Retrospective Loneliness EE: Score on the “lonely” item of the emotional eating (EE) subscale of the Three-Factor Eating Questionnaire	Momentary Loneliness EE: EMA-derived coefficients representing an individual’s tendency to lapse from a weight control diet following momentary elevations in loneliness, relative to the individual’s average level of loneliness, across timepoints.
Composite of all emotions	Retrospective Overall EE: Average of score on the “blue,” “anxious,” and “lonely” items (the emotional eating (EE) subscale) of the Three- Factor Eating Questionnaire	Momentary Overall EE: EMA-derived coefficients representing an individual’s tendency to lapse from a weight control diet following momentary elevations in negative emotions, averaging across three emotion states (sadness, stress, loneliness), relative to the individual’s average level of negative emotion, across timepoints.

2.3.1. Weight

Weight was measured at each assessment using a standardized Seca scale accurate to 0.1 kg. Height was measured with a stadiometer to establish BMI (kg/m²).

2.3.1. Retrospective EE

Retrospective EE was evaluated using the EE subscale of the 18-item version of the Three Factor Eating Questionnaire (TFEQ), a well-validated measure of eating behaviors (Anglé et al., 2009). The reliability of this subscale was adequate in our sample (α = 0.85) (Karlsson et al., 2000). Utilizing a Likert scale ranging from 1 (“Definitely False”) to 4 (“Definitely True”), participants reported the degree to which their general tendency to eat or overeat in response to three emotions: anxiety (“When I feel anxious, I find myself eating.”), sadness (“When I feel blue, I often overeat.”), and loneliness (“When I feel lonely, I console myself by eating.)” We averaged scores on these three items to create a composite “Retrospective Overall EE” score.

2.3.2. Negative Emotions

Given our paper’s focus on EE concordance, we used the negative affect EMA items (sad/down, stressed, lonely) adapted from the Positive and Negative Affect Schedule (PANAS) (Watson et al., 1988) that aligned with TFEQ negative affect items (blue, anxious, lonely). At each EMA survey, participants reported the degree to which they were currently sad/down, stressed, lonely, irritated/angry, and bored on a 5-point Likert scale ranging from 1 (Not at all) to 5 (Extremely). We computed mean momentary negative emotion at each EMA timepoint and created a composite momentary negative emotion variable (“overall negative emotion”) by averaging all negative affect items. For timepoints where responses on one or more emotion item were missing but at least one emotion item was completed (1%), overall negative emotion was computed from the completed items.

2.3.3. Dietary Lapses

At each EMA survey, participants reported whether they lapsed from their diet by responding to the following question: “Since the last time you completed this survey, did you have a dietary lapse? A dietary lapse is eating or drinking likely to cause weight gain, and/or put weight loss/maintenance at risk.” Participants then had the option to indicate the type of dietary lapse that they experienced (eating a larger portion than intended, eating at an unintended time, eating an unintended food), results for which are reported elsewhere (Forman et al., 2017). Supporting the construct validity of this dietary lapse measure, our prior work using this same measure has found that lapses occur most frequently in evenings and on weekends, and when at home (Forman et al., 2017), consistent with reports of greater caloric intake on weekends (An, 2016; Racette et al., 2008) and when self-regulatory resources are depleted (i.e., evenings, which are most often spent at home) (Hofmann et al., 2012). Supporting the predictive validity of this measure, dietary lapse frequency has been found to be negatively associated with weight loss (Forman et al., 2017).

2.4. Analyses

The data analytic plan was specified prior to conducting any analyses. For all analyses including EMA data, participants who did not report lapsing at any point during the mid-treatment assessment period were excluded. To generate a momentary index of EE for each participant, we utilized a two-level generalized linear mixed-effect model (GLMM) using a binomial distribution with a logit link function, including composite momentary negative emotion (i.e., participants’ mean level of emotion across timepoints) as a fixed within-subjects predictor of the probability of lapse at the next timepoint (effect-coded as 0 = no lapse reported, 1 = lapse reported), with a subject-level random intercept and random slope. Consistent with conceptualization of emotional eating as reflecting overeating, or unhealthy eating (Devonport et al., 2019), momentary EE was defined as a negative emotion increasing the likelihood of subsequent lapse (defined as eating or drinking likely to cause weight gain, and/or put weight loss/maintenance at risk), relative to a participant’s baseline. As such, composite momentary negative emotion was mean-centered within subjects and EMA predictors were lagged by one timepoint (within-days only) and standardized to have variance of one and mean of zero, consistent with statistical recommendations for examining within-subjects effect using repeated-measures designs (Curran & Bauer, 2011; Singer et al., 2003) and also consistent with prior EMA studies on dietary lapses and eating behavior (Crochiere et al., 2019; Forman et al., 2017; Williams-Kerver et al., 2020). Additional exploratory GLMM models included individual momentary negative affect items (sad/down, stressed, lonely) as predictors of lapse at each EMA timepoint.

All GLMM models used an unstructured covariance matrix and restricted maximum likelihood estimation (REML) and allowed for covariances between random slopes and intercepts. Models were fit using the nlminb (optimization using PORT routines) methods for parameter optimization (R packages optimx (Nash et al., 2020), lme4 (Bates et al., 2014) and lmerTest (Kuznetsova et al., 2018)). Individual-participant coefficients, representing the mean change to each individual’s log-transformed odds of lapsing associated with a one-unit increase in momentary negative emotion at the previous timepoint, were then extracted from each model by adding model-estimated subject-level variance components to group-level fixed effects, which allowed us to obtain indices of momentary sadness EE, momentary stress EE, momentary loneliness, and overall momentary EE (see Table 1). To aid in interpretation, coefficients were converted into odds ratios through exponentiation (see Table 3). For more model specification details, see Supporting Information (Tables S1 and S2).

Table 3.

Descriptive statistics for momentary EE coefficients and odds ratios derived from generalized linear mixed-effect models (GLMM) using EMA data.

Variable	Mean Coefficient (SD)	Mean Odds Ratio (SD)	% Sample with OR > 2	OR Range
Momentary Sadness EE	0.29 (1.03)	2.05 (1.85)	37.6%	0.03–10.62
Momentary Stress EE	0.40 (1.70)	4.22 (5.67)	48.9%	0.01–33.53
Momentary Loneliness EE	1.23 (0.73)	4.30 (2.77)	77.4%	0.45–13.97
Momentary Overall EE	0.78 (2.37)	13.20 (25.93)	56.4%	0.01–172.98

Note: N = 133 for all momentary EE measures.

We then examined concordance between retrospective and momentary EE using Spearman correlations, as well as the association between each measure and current BMI. To examine prospective associations between EE and PWL, both retrospective and momentary EE were included as predictors of PWL from baseline to post-treatment, 1-year and 2-year follow-up, using ordinary least-squares regression. Additional exploratory models examined prospective associations between individual emotions (i.e., stress/anxiety, loneliness, sadness) included in each measure and PWL, considering prior research suggesting that different individual emotions (e.g., stress versus sadness) may differentially predict subsequent eating behavior (Reichenberger et al., 2018). Consistent with an intent-to-treat approach, all missing weight data (none at mid-treatment; N = 14 at post-treatment, N = 20 at 1-year follow-up; N = 28 at 2-year follow-up) were imputed using maximum likelihood estimation.

3. Results

At mid-treatment, 163 participants had retrospective EE (TFEQ) data available. After excluding those who reported no lapses during the assessment period, 133 participants had momentary EE data available (representing 6,018 total surveys), and 124 participants had both retrospective and momentary EE data available (thus, data for 124 participants were used in analyses). Average EMA survey compliance (i.e., proportion of surveys completed) among included participants was 90.6% (SD = 15.3% and ranged from 42.0% to 100.0%. Participants reported a mean of 4.09 dietary lapses (SD = 3.76) over the assessment period, ranging from 0 to 17. Mean momentary sadness was 1.27 (SD = 0.62); 1.73 (SD = 0.88) for stress; 1.23 (SD = 0.57) for loneliness, and 1.41 (SD = 0.54) for overall negative emotion.

Descriptive statistics for retrospective EE are presented in Table 2, and descriptive statistics for momentary EE coefficients and corresponding odds ratios are presented in Table 3. Fifty-six percent of participants showed odds ratios greater than 2 on overall momentary EE, indicating that more than half of participants experienced at least a two-fold increase to their odds of reporting a dietary lapse following a one-point increase in overall momentary negative emotion. Odds ratios of two or greater may indicate clinically significant effects (Ferguson, 2016), suggesting that most participants in our sample experienced a clinically meaningful effect of increased momentary negative emotion on their odds of lapsing from their diet.

Table 2.

Descriptive statistics for Retrospective EE.

Retrospective (TEFQ) EE	Mean (SD)	% Scoring ≥ 3 (“Mostly True”)
Anxious	2.57 (0.90)	50.7%
Blue	2.43 (0.92)	46.0%
Lonely	2.13 (0.89)	32.0%
Retrospective Overall EE	2.38 (0.79)	30.0%

Note: TFEQ = Three-Factor Eating Questionnaire, 18-item version (Anglé et al., 2009). N = 163. Transforming the EE subscale score to a 0–100 scale, as is convention (Romon et al., 2004), the mean Retrospective EE subscale was 46.24 (SD = 26.15).

Spearman correlations between our two methods of assessing EE indicated that, contrary to our hypothesis, momentary overall EE was moderately negatively correlated with retrospective overall EE (r = −0.27, p < 0.001). Similarly, momentary sadness, stress and loneliness EE were weakly to moderately negatively correlated with their retrospective counterparts (see Table 4). Consistent with hypotheses, Spearman correlations between retrospective EE and BMI (r = 0.20, p = 0.03), and momentary EE and BMI (r = 0.29, p = 0.001), were both positive.

Table 4.

Spearman correlations between momentary EE and retrospective EE

	1	2	3	4	5	6	7	8
1. Momentary Sadness EE	1.00	0.96**	0.81**	0.89**	−0.18*	−0.19*	−0.14	−0.20*
2. Momentary Stress EE	–	1.00	0.98**	0.96**	−0.30**	−0.29**	−0.33**	−0.35**
3. Momentary Loneliness EE	–	–	1.00	0.94**	−0.31**	−0.31**	−0.32**	−0.35**
4. Momentary Overall EE	–	–	–	1.00	−0.24**	−0.23*	−0.24**	−0.27**
5. Retrospective Sadness EE	–	–	–	–	1.00	0.71**	0.63**	0.90**
6. Retrospective Anxiety EE	–	–	–	–	–	1.00	0.58**	0.87**
7. Retrospective Loneliness EE	–	–	–	–	–	–	1.00	0.83**
8. Retrospective Overall EE	–	–	–	–	–	–	–	1.00

^**p ≤ .01,

^*p ≤ .05.

Note. Retrospective EE is equivalent to the emotional eating subscale score on the Three-Factor Eating Questionnaire, 18-item version (TFEQ-R18) (Anglé et al., 2009); Momentary EE was derived from EMA data. N = 124.

Results of linear regression analyses (Table 5) revealed that, consistent with hypotheses, retrospective EE negatively predicted weight loss at post-treatment (b = −4.06, SE = 0.82, t = −5.96, p < 0.001), 1-year (b = −3.86, SE = 0.94, t = −4.12, p < 0.001) and 2-year follow-up (b = 2.33, SE = 1.07, t = −2.17, p = 0.03). Similar patterns of results were observed with retrospective sadness, anxiety, and loneliness EE (see Table S3), wherein retrospective EE for each specific emotion predicted weight loss at post-treatment and 1-year follow-up for all individual emotion items, and also at 2-year-follow for retrospective anxiety and loneliness, but not sadness EE (p’s < 0.05). Results at 2-year follow-up were in the same direction and reached significance for retrospective anxiety EE and retrospective loneliness EE, but not retrospective sadness EE. Figure 1 presents model-estimated weight losses at each assessment as a function of EE measurement method (retrospective and momentary).

Table 5.

Results of linear regression models predicting percent weight loss from baseline at post-treatment, 1-year follow-up, and 2-year follow-up using Momentary Overall EE and Retrospective Overall EE.

		Post-Treatment (12-Month) PWL				1-Year Follow-Up PWL				2-Year Follow-Up PWL
Model	Variables	b	SE	t	p	b	SE	t	p	b	SE	t	p
1	Intercept	11.61	0.72	−16.19	< 0.001	6.33	0.79	8.03	< 0.001	3.74	0.86	4.33	< 0.001
	Momentary Overall EE	0.47	0.29	−1.64	0.10	0.76	0.32	−2.38	0.02	0.75	0.35	−2.15	0.03
2	Intercept	22.21	2.11	10.52	< 0.001	16.67	2.42	6.90	< 0.001	10.46	2.75	3.80	< 0.001
	Retrospective Overall EE	−4.06	0.82	−4.96	< 0.001	−3.86	0.94	−4.12	< 0.001	−2.33	1.07	−2.18	0.03

Note. N = 124. PWL: Percent Weight Loss. Momentary EE is derived from Ecological Momentary Assessment (EMA) data. Retrospective EE is equivalent to the emotional eating subscale score on the Three-Factor Eating Questionnaire, 18-item version (TFEQ-R18) (Anglé et al., 2009).

Figure 1.

Linear regression model-estimated percent weight change at baseline, mid-treatment, post-treatment, 1-year and 2-year follow-up for retrospective EE assessed by the TFEQ (A) and momentary EE assessed using EMA (B). Error bars represent model-estimated standard error of the mean. Red lines indicate 1 standard error above the mean estimate and blue lines indicate 1 standard error below the mean estimate. Percent weight change rather than percent weight loss (PWL) is depicted by convention.

Contrary to hypotheses, the opposite pattern was observed for retrospective EE (Table 5)— momentary overall EE positively predicted weight loss at 1-year (b = 0.76, SE = 0.32, t = 2.3, p = 0.02) and 2-year follow-up (b = 0.75, SE = 0.35, t = 2.15, p = 0.03), which was contrary to hypotheses. Results at post-treatment were in the same direction but did not reach statistical significance (b = 0.47, SE = 0.29, t = 1.64, p = 0.10). Similar patterns of results were observed with momentary sadness, stress, and loneliness EE (see Table S3), such that momentary EE for each specific emotion significantly predicted weight loss at 1- and 2-year follow-up points at a significant (p < 0.05) or trend (p = .07) level. Results at post-treatment were in the same direction and reached significance for momentary stress EE and momentary loneliness EE (p < 0.05), but not momentary sadness EE.

To account for the possibility that treatment condition (effect-coded as ABT = 1, SBT = −1), EMA compliance, mid-treatment BMI or mid-treatment depression levels (BDI-II) might influence relationships between EE and weight loss, analyses were repeated by including these variables as predictors in each model. Since findings were unchanged when controlling for all four of these variables, results are presented from models including only EE as a predictor. As an exploratory follow-up analysis, we examined whether EE measures interacted with treatment condition in predicting weight loss by including interaction terms between EE and condition in the linear regression analyses described above. Results did not provide support for an interaction between condition and retrospective EE in predicting weight loss at post-treatment (b = 0.04, SE = 0.83, t = −0.05, p = 0.96), 1-year follow-up (b = −0.90, SE = 0.96, t = −0.95, p = 0.35) or 2-year follow-up (b = −1.71, SE = 1.09, t = −1.58, p = 0.12). Similarly, results did not support an interaction between condition and momentary EE in predicting weight losses at post-treatment (b = 0.03, SE = 0.29, t = 0.10, p = 0.92), 1-year follow-up (b = 0.23, SE = 0.32, t = 0.71, p = 0.48) or 2-year follow-up (b = 0.19, SE = 0.35, t = 0.55, p = 0.58).

4. Discussion

EE is proposed as a key driver of overeating and weight loss barrier (Frayn & Knäuper, 2018). However, the existing literature largely relies upon retrospective self-report measures of EE, leaving open the possibility that while retrospective EE has predictive utility, it measures a different construct than expected such that the true relationship between EE and weight outcomes may yet be undetermined. Our work extends literature on the relationship between EE and overweight/obesity by exploring a novel method of capturing EE (momentary EE), and examining the concordance between this novel method of measuring momentary EE and a traditional retrospective EE measure. Our novel EE measurement method used an approach more consistent with the temporal conceptualization of EE (in which emotions are thought to precede overeating), and allowed us to examine, for the first time, individual differences in momentary EE and their association with weight outcomes. Our momentary EE measurement approach thus contributes to the multimodal assessment of EE and allowed us to further elucidate what retrospective EE may, or may not, capture, which would aid in clarifying understanding of the true relationship between EE and weight outcomes.

Extending prior research in which retrospectively reported EE was found to be unrelated to EE assessed in a more objective way (e.g., quantity of food eaten following an experimental mood induction) (Boh et al., 2016; Bongers & Jansen, 2016; Brogan & Hevey, 2013), we found that retrospective EE was moderately and negatively associated with momentary EE. Our study thus lends support to a body of research suggesting that retrospective EE measures may not measure EE as intended (Bongers & Jansen, 2016). Indeed, while we expected to observe a modest, positive relationship between our two EE measures, we instead found a modest negative correlation, suggesting that these measurement approaches tapped into distinct, and potentially even opposing, constructs. It has been suggested that rather than reflecting “true” EE, high retrospective EE may instead reflect general lack of control when overeating, disinhibited eating, or concern over one’s eating, consistent with its associations with greater self-reported susceptibility to eat in response to external cues and greater laboratory-assessed food intake following mood inductions (Bongers & Jansen, 2016). Participants who score highly on retrospective EE measures may thus have a tendency to attribute overeating to negative affect, (Adriaanse et al., 2016) yet may actually eat in response to a variety of external and internal cues. Consistent with prior research (Frayn & Knäuper, 2018), we observed a significant, though weak positive relationship between retrospective EE and current BMI.

Our findings also replicate prior work on the relationship between weight loss and retrospective EE. The scores obtained for the TFEQ-R18 EE subscale in our sample were highly similar to a prior study evaluating TFEQ EE among individuals with obesity (Anglé et al., 2009). Additionally, consistent with past research that administered the TFEQ EE subscale to individuals with BMI ≥ 30, greater retrospective EE robustly predicted less weight loss (Dohle et al., 2014; Koenders & van Strien, 2011; Sung et al., 2009) at post-treatment and 1- and 2-year follow-ups. Similar findings emerged when examining prospective relationships between individual emotion items on the TFEQ EE subscale and weight loss. Of note, because analyses were found to be unchanged when controlling for BMI, it is unlikely that findings were driven by individuals with higher EE starting treatment at a higher BMI and thus losing a higher proportion of body weight. Our results thus support the predictive utility of retrospective EE, highlighting that brief self-report tools for EE consistently predict weight control. Indeed, given that the TFEQ EE subscale is easy and inexpensive to administer, especially relative to a complex EMA protocol, it arguably shows good pragmatic utility for identifying which individuals are most likely to struggle during weight loss treatment and may require additional support. However, despite the predictive validity of the TFEQ, its construct validity remains in question, as highlighted by findings related to our novel momentary EE measure, thus posing a challenge to understanding the true relationship between EE and weight loss.

Consistent with findings for retrospective EE, greater momentary EE was found to be positively associated with BMI, suggesting that these two measurement approaches might have tapped into distinct constructs (i.e., distinct facets of EE or other contributors to overeating) that were nonetheless both associated with higher body weight. However, in contrast to retrospective EE, greater momentary EE predicted improved weight outcomes in our sample at post-treatment and 1- and 2-year follow-ups, and this finding was generally upheld when examining different momentary emotions individually. While this finding further supports the distinctness of retrospective and momentary EE, in particular their distinct potential to predict long-term treatment outcomes, it also raises the question of what momentary EE actually measures. In service of future research, we speculate on possible interpretations. For one, the observed association between momentary EE and weight loss may reflect a “true” tendency for increased EE to predict improved weight loss outcomes following BT. Given that BTs, to at least a modest extent, target relationships between emotions and eating behaviors (Butryn et al., 2011), individuals for whom EE plays a major role in overeating may stand to benefit in particular from these treatments. Such an interpretation would require that our momentary EE measurement method enabled a more objective assessment of EE compared to retrospective measures, such as by capturing the temporal nature of EE. Further, relative to retrospective EE, momentary EE was presumably less susceptible to recall bias and “false positives” (as incorrectly reporting the presence of an emotion is presumed relatively unlikely when reporting current, as opposed to past, emotions). Therefore, to the extent that our momentary EE measure was able to capture EE more objectively, our results could suggest that greater EE may not predict worse weight outcomes and could in fact predict improved weight loss outcomes to the extent that it is targeted during BT.

However, it is also possible the observed positive association between momentary EE and weight loss resulted from participant traits that may have confounded our measurement of momentary EE. For example, participants’ abilities to accurately report dietary lapses or participants’ awareness of their emotional states may have confounded our measurement of momentary EE. Indeed, some evidence suggests that individuals with overweight/obesity have higher rates of alexithymia (i.e., an impoverished understanding of their emotions) and interoceptive awareness deficits (Casagrande et al., 2020; Willem et al., 2019) and thus might have greater difficulty detecting when they have lapsed from their diet or when they are experiencing negative affect (Willem et al., 2019). Additionally, individual differences in motivation to engage in treatment could have influenced the extent to which participants accurately reported their emotional states and dietary lapses. Thus, higher momentary EE may predict improved weight outcomes due to the confounding influence of other traits that tend to predict weight loss success.

Although we believe our method for capturing EE in a momentary fashion presents a promising new measurement approach, several limitations bear noting. Just as retrospective EE measures require that individuals have an awareness of the relationship between emotions and eating behaviors, our momentary EE measurement method was limited by its dependence on participants’ abilities to recognize their emotional states and accurately report on dietary lapses, likely resulting in “false negatives” (failing to report EE when it occurs). Future research would benefit from attempts to disentangle momentary EE from potential confounds, such as factors related to an individual’s willingness to report on emotional states and lapses. For example, sensor technologies (Bedri et al., 2017) could be used to automatically detect eating episodes rather than relying on participants’ abilities to self-report dietary lapses.

Another potential limitation is that EE data were collected at mid-treatment, rather than baseline. Future studies should examine whether these longitudinal weight loss findings can be replicated when EE is assessed prior to treatment, especially because at mid-treatment in our study, participants had already been introduced to strategies for managing emotional influences on eating, which could have influenced the observed frequency or severity of EE in our sample. Indeed, by the mid-treatment assessment (which occurred 6 months into treatment), participants had already completed 19 of 25 sessions, including most, but not all, of the content covered in the ABT group related to managing the impact of emotional states on eating (e.g., willingness to experience discomfort, awareness of cues that trigger unhealthy eating). However, the levels of retrospective EE observed in the current sample were very similar to a prior study evaluating the TFEQ EE subscale among individuals with obesity (Anglé et al., 2009). Of note, while it could be argued that retrospective EE (a more trait-like measure) was impacted more by the intervention than momentary EE (a more state-like measure), we think this unlikely given prior evidence suggesting that retrospective EE measures can change substantially over time in response to BT (Boutelle et al., 2018; Juarascio et al., 2020; Levoy et al., 2017).

In addition, our EMA protocol asked participants to report dietary lapses (eating/drinking that could put a calorie goal/weight control at risk), whereas retrospective EE items inquire about eating or overeating. On the one hand, we view our operationalization of momentary EE as a strength, because dietary lapses should, for a weight loss sample, specifically capture the types of eating thought to characterize emotional eating (e.g., indulgent-, planned-, or over-eating) (Adriaanse et al., 2011; Bennett et al., 2013; Kemp et al., 2013; Kuijer & Boyce, 2012), as opposed to more general patterns of eating that may coincide with negative affect yet not capture emotional eating (e.g., eating a healthy planned meal after feeling sad). However, it is important to note that this methodological difference in the measurement of retrospective and momentary EE may have contributed to the lack of concordance between these measured observed in the present study. Another possible contributor to this lack of concordance was that our retrospective EE measure asked participants to estimate their level of emotional eating over the past several weeks, whereas our measure of momentary EE only evaluated emotions and eating behaviors at one specific moment in time. Future research is needed to examine other forms of momentary EE (e.g., likelihood of eating as opposed to lapsing following emotions), and their associations with retrospective EE and weight outcomes, including in non-weight loss-seeking samples. Further, while our momentary EE measure assessed self-reported stress, our retrospective EE measure assessed self-reported anxiety. While stress and anxiety are related to one another, they are distinct; stress is a response to, or perceived inability to cope with, demands (Baum, 1990), whereas anxiety is a more future-oriented emotion involving feelings of tension, nervousness, worry and apprehension (Spielberger, 2010). As such, future research should examine the concordance between matching items (e.g., momentary stress EE and retrospective stress EE). Future studies could also attempt to apply our same methodology to examining EE as it pertains to fluctuations in negative affect over varying timescales (e.g., hour by hour, over the course of an entire day, over several days). Lastly, our study did not directly assess the validity of momentary EE, for example by examining the concordance between momentary EE and EE examined in a controlled laboratory setting, which future research should also address.

In sum, our results support prior research raising the possibility that retrospective self-report measures of EE do not capture EE as intended, meaning that the relationship between “true” EE and weight loss outcomes may still be unknown. Our findings suggest that when EE is assessed in a more naturalistic way and with an individual difference measurement approach, higher EE is indeed associated with greater concurrent BMI but may not predict poorer weight outcomes and may even potentially predict improved weight outcomes following treatment, unlike retrospective EE measures. However, future research into this nascent area of research is needed and is especially important in ascertaining the true relationship between EE and weight loss outcomes, which could have important clinical implications. If our findings with momentary EE were to be replicated with other non-retrospective methods of assessing EE, this could suggest that “true” EE (i.e., increased food intake in response to negative emotions) may not impede weight loss independently of other, more general behavioral constructs captured by retrospective EE measures, such as disinhibited eating. Lastly, it is worth considering the perspective that both retrospective and momentary measurement approaches reflect valid yet distinct components of EE (e.g., emotional eating behavior vs. the perception of oneself as an emotional eater). In this way, both measurement approaches could make a unique and valuable contribution to scientific understanding of EE. Taken together, our results contribute to a growing body of work evaluating relationships between negative affect and eating behaviors and highlight the importance of multimodal assessment when attempting to delineate the role of EE in the etiology and treatment of overweight and obesity.