Differences in Affective Dynamics Among Eating-Disorder Diagnostic Groups

Abstract

Emotion-regulation theories suggest that affect intensity is crucial in the development and maintenance of eating disorders. However, other aspects of emotional experience, such as lability, differentiation, and inertia, are not as well understood. This study is the first to use ecological momentary assessment (EMA) to examine differences in several daily negative affect (NA) indicators among adults diagnosed with anorexia nervosa (AN), bulimia nervosa (BN), or binge-eating disorder (BED). We used EMA data from three large studies to run a series of linear mixed models; the results showed that participants in the AN and BN groups experienced significantly greater NA intensity and better emotion differentiation than participants in the BED group. Alternatively, the BN group demonstrated significantly greater NA lability than the AN group and greater NA inertia than the BED group. These results suggest that several daily affective experiences differ among eating-disorder diagnostic groups and have implications toward distinct conceptualizations and treatments.

It has long been established that emotion regulation is central to the understanding of psychopathology (Gross & Jazaieri, 2014; Gross & Muñoz, 1995; Sheppes, Suri, & Gross, 2015). Indeed, affective experience has been the focus of numerous empirical investigations for the past several decades. With respect to eating-disorder psychopathology, several theories posit that affect-related mechanisms may play an important role in the etiology and maintenance of anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED). After recent “microlongitudinal” research, including the use of ecological momentary assessment (EMA) data (Stone & Shiffman, 1994), we now understand patterns of temporal trajectories of affect surrounding various disordered eating behaviors (e.g., Berg et al., 2013; Engel et al., 2013; Haynos et al., 2017; Smyth et al., 2007). However, additional nuances of emotional experience, such as the dynamics of affective lability, emotion differentiation, and emotional inertia, have not yet been comprehensively examined among the eating-disorder diagnostic groups.

Theories of Affect in the Eating Disorders

Historically, models of eating-disorder psychopathology have included affect as a construct that helps explain how disordered eating behaviors and cognitions develop and are maintained. With respect to binge-eating behaviors, restraint theory (Herman & Polivy, 1980), the affect regulation model of eating pathology (Hawkins & Clement, 1984), escape theory (Heatherton & Baumeister, 1991), and the resource depletion model of disordered eating (Loth et al., 2016) all hypothesize that negative affect (NA) functions as a precipitant of binge eating. In turn, this causal link between NA and binge eating has been incorporated into several theories on the development and maintenance of BN and BED (e.g., Fairburn, 2008; Safer, Telch, & Chen, 2009; Wonderlich et al., 2015). NA has also been linked to dietary restriction and incorporated into emotion regulation theories regarding the development and maintenance of AN (Haynos & Fruzzetti, 2011; Wildes, Ringham, & Marcus, 2010).

In accordance with these theories, the importance of negative affect intensity (i.e., mean level of negative affect) has received a great deal of empirical support within the eating-disorder literature. Indeed, studies regarding the influential relationships between momentary negative affect intensity and the occurrence of disordered eating behaviors have been particularly beneficial toward the overall understanding of eating-disorder psychopathology (Berg et al., 2013; Engel et al., 2013; Hilbert & Tuschen-Caffier, 2007; Lavender et al., 2016; Smyth et al., 2007; Wonderlich et al., 2015). However, one critique regarding emotion modeling in the eating-disorder literature is that the approach has been largely unidimensional in its emphasis on intensity. Researchers of other types of psychopathology have examined different dynamic affective processes and how they interact with symptoms in an individual’s naturalistic environment (for a review, see Trull, Lane, Koval, & Ebner-Priemer, 2015). Consequently, employing a similar approach to emotion modeling would build on the existing affect intensity literature to further understanding regarding the role of affect among eating-disorder diagnostic groups.

Affective Dynamics

A burgeoning area of research involves the characterization and examination of additional dynamic affective experiences beyond that of affect intensity. As outlined by Trull and colleagues (2015), affective lability, emotion differentiation, and emotional inertia are also meaningful constructs that account for differences in emotional experience over time and that may ultimately have mechanistic implications toward underlying psychopathology. Affective lability (otherwise referred to as affective instability) represents the extent to which levels of affect change or fluctuate over successive time intervals for a given individual (Ebner-Priemer et al., 2007). This combination of within-persons variability (i.e., how much the measured level of affect differs from its person mean) and temporal dependency (i.e., how consistent these measurements are across successive time points) has been proposed as crucial for the reliable and valid characterization of affective lability in emotion research (Trull et al., 2015). High levels of affective lability represent broader difficulties with emotion regulation, which may ultimately signal greater severity of overall psychopathology, such as that seen in borderline personality disorder (American Psychiatric Association, 2013).

Emotion differentiation (or emotion granularity) refers to the ability of an individual to distinguish discrete emotional states (e.g., anger vs. fear vs. sadness) in the moment (Barrett, 2004; Barrett, Gross, Christensen, & Benvenuto, 2001). It has been hypothesized that, by more precisely identifying momentary emotions, individuals gain access to specific information that assists them in selecting the most applicable emotion regulation strategy to employ (Barrett et al., 2001; Schwarz & Clore, 2003). Consequently, poor emotion differentiation suggests that individuals experience emotions in broad generalizations, such as “good” or “bad,” which results in the greater likelihood of engaging in inappropriate or ineffective emotion-regulation strategies (Barrett et al., 2001). Over time, this pattern of maladaptive emotion regulation may result in the development or maintenance of psychopathology.

Emotional inertia represents how predictable levels of affect are from one moment to the next. Although conceptually similar to affective lability, emotional inertia is considered a separate metric because it focuses solely on the temporal-dependency aspect of affect independent of the magnitude of within-persons variability (Kuppens, Allen, & Sheeber, 2010; Trull et al., 2015). Thus, higher levels of emotional inertia suggest inflexibility of emotion experience over time (Hollenstein, 2015; Suls, Green, & Hillis, 1998). Because it is expected that emotional experience will change in response to different contexts and environmental needs, such inflexibility becomes particularly problematic and potentially indicative of psychopathology and psychological maladjustment (Kashdan & Rottenberg, 2010; Kuppens et al., 2010).

Albeit limited, there is some existing research linking affective dynamics, as measured from daily diary data, with maladaptive behaviors among individuals diagnosed with an eating disorder. Several studies that have examined affective lability among women with BN found that affect lability was greater on days with bulimic behaviors (i.e., binge eating and purging) than days without (Berner et al., 2017; Selby et al., 2012) and was associated with more frequent binge-eating episodes (Anestis et al., 2010) and other impulsive behaviors (Anestis et al., 2009). Previous research has also shown that heightened affect lability predicted dietary restriction among a sample of women with AN (Lavender et al., 2013). With respect to emotion differentiation, one study among women with AN found that poor emotion differentiation was associated with an array of maladaptive behaviors, including more frequent vomiting, laxative use, dietary restriction, exercise, and self-weighing (Selby et al., 2014). Taken together, these studies have created the foundation for examining affective dynamics in the eating disorders. However, research that comprehensively examines these different affect metrics among several distinct diagnostic groups is needed to truly highlight the role of affective dynamics in eating disorders.

Utility of Ecological Momentary Assessment to Assess Emotion Constructs

EMA methods, including the repeated assessment of relevant constructs throughout the day within an individual’s natural environment, are used to calculate and capture the various affective dynamics. EMA is accomplished through the use of handheld computers and smartphone devices, which makes it possible for an individual to record psychological data as they occur in the moment. One primary benefit of EMA is that it reduces the risk that an individual will forget or remember incorrectly important details that happened in the past (Gorin & Stone, 2001). Thus, EMA allows reliable assessment of affect without biases associated with retrospective recall (Ebner-Priemer & Trull, 2009). However, a more pertinent benefit of EMA as it relates to this study is that it allows for the examination of microtemporal relationships in affect over time. Thus, EMA overcomes limitations of traditional, nonmomentary, retrospective self-report measures that require individuals to aggregate their emotional experience (Ebner-Priemer & Trull, 2009). Indeed, as Trull and colleagues (2015) noted in their review,

emotional experience is dynamic, not static, and methods that provide multiple microassessments of emotional experience across time and situations can shed light on the dynamic nature of various affective processes that are theorized to be central to the development and course of psychopathology. (p. 355)

Present Study

Although emotional experience is becoming increasingly more relevant in the eating-disorder literature, no study to date has comprehensively examined eating-disorder diagnostic differences across several indices of emotion dynamics. Given that affect seems important in understanding the development and maintenance of eating-disorder psychopathology, research that illuminates how the complexity of emotional experience unfolds on a daily basis in the natural environment and explores whether any noteworthy differences in affective dynamics emerge between distinct eating-disorder diagnostic groups is needed. Thus, the goal of the current study was to use EMA data in an exploratory research design to examine whether group differences emerged in the daily experience of negative affective intensity, affective lability, emotion differentiation, and emotional inertia among individuals diagnosed with AN, BN, or BED. Although research has examined eating-disorder group differences on conceptually similar trait-based variables (e.g., Goldner, Srikameswaran, Schroeder, Livesley, & Birmingham, 1999; Nowakowski, McFarlane, & Cassin, 2013; Westen & Harnden-Fischer, 2001) and momentary measures of affective intensity (e.g., De Young et al., 2013; Goldschmidt et al., 2013; Hilbert & Tuschen-Caffier, 2007; Le Grange et al., 2013), specific a priori hypotheses were not identified because this research is the first to comprehensively explore eating-disorder-group differences in these specific momentary, state-based affective constructs of NA lability, differentiation, and inertia.

Method

Participants

The current research included a total of 363 participants. This larger sample comprised three smaller samples from separate research protocols: an EMA protocol for women meeting diagnostic criteria for anorexia nervosa (EMA AN; Engel et al., 2013), an EMA protocol for women meeting diagnostic criteria for bulimia nervosa (EMA BN; Smyth et al., 2007), and a treatment protocol including EMA data for men and women meeting diagnostic criteria for binge eating disorder (EMA BED; Peterson et al., 2019). Demographic information for age, body mass index (BMI), and gender for all three samples can be found in Table 1, and additional participant information is described below.

Table 1.

Sample Characteristics

Statistic	Study
	EMA AN (n = 118)	EMA BN (n = 133)	EMA BED (n = 112)
Mean age (years)	25.32 (8.36)a	25.34 (7.61)a	39.97 (13.37)b
Race/ethnicity
White	96.6%	97.0%	92.7%
Black	1.7%	0.0%	0.0%
American Indian	0.0%	1.5%	0.0%
Asian	0.0%	0.8%	0.9%
Hispanic	0.0%	0.0%	1.8%
“Other”	1.7%	0.8%	4.5%
Mean BMI (kg/m2)	17.15 (1.03)a	23.92 (5.21)b	35.13 (8.66)c
Women	100%a (n = 118)	100%a (n = 133)	82.30%b (n = 93)
Mean number of signals per day	6.00 (0.64)a	6.08 (0.67)a	4.22 (1.18)b
Mean EMA adherence rate	87.28%a	91.88%b	76.49%c

Note: Values in parentheses are standard deviations unless otherwise noted. Values for race/ethnicity may not add to 100% because of rounding. Cells sharing a common subscript are not statistically significantly different from one another at p < .01. EMA = ecological momentary assessment; AN = anorexia nervosa; BN = bulimia nervosa; BED = binge eating disorder; BMI = body mass index.

EMA AN

The sample for this study included 118 women who met full (n = 59) or subthreshold (n = 59) criteria for AN as outlined by the fourth edition of the Diagnostic and Statistical Manual for Mental Disorders (DSM–IV; American Psychiatric Association, 1994). Women with subthreshold AN met all of the DSM–IV criteria for AN except for (a) BMI between 17.5 and 18.5, (b) absence of amenorrhea, or (c) absence of cognitive features (Engel et al., 2013). The sample was largely White (96.6%), and a majority had at least some college education (90.7%; see further details in Engel et al., 2013). A total of 601 participants were initially screened by phone for eligibility; 166 of those individuals (27.6%) participated in further screening at one of the research study sites, and 121 participants (20.1%) were ultimately eligible for participation and subsequently enrolled in the study. Data from 3 participants (2.5%) were dropped because their adherence rates on the EMA protocol were less than 50%, which resulted in a final sample of 118.

EMA BN

The sample included 133 women who met full DSM–IV criteria for BN. The sample was largely White (96.9%), and 60.0% of participants had at least some college education (for further details, see Smyth et al., 2007). Initially, 154 participants were screened for eligibility criteria via phone and in-person assessment, and 143 of those participants (92.9%) were eligible for participation and began the EMA protocol. Seven participants did not complete the full EMA protocol (4.9%), and data from an additional 3 participants (2.1%) were dropped because of poor adherence rates, which resulted in the final sample of 133.

EMA BED

The sample included 112 men and women who met criteria for BED as outlined by the fifth edition of the Diagnostic and Statistical Manual for Mental Disorders (DSM–5; American Psychiatric Association, 2013). The sample was largely White (91.4%), and a majority had received a college degree (68.8%; for further details, see Peterson et al., 2019). Initially, 713 individuals were screened for eligibility criteria via phone and in-person assessment, and 112 of those participants (15.7%) were eligible for participation and began the EMA protocol.

Procedure

The three studies received approval from the institutional review boards across all sites, and all participants provided informed consent before their participation.

EMA AN and EMA BN

Participants were recruited from various clinical and community settings. Participants in the EMA BN group were recruited from research sites in Fargo, North Dakota, and Minneapolis, Minnesota (Smyth et al., 2007), and participants in the EMA AN group were recruited from the same research sites plus an additional site located in Chicago, Illinois (Engel et al., 2013). After completing the initial phone screening, participants were scheduled for either two (EMA AN) or three (EMA BN) separate on-site visits to complete a physical examination (to ensure medical stability) and a structured interview to confirm eligibility criteria.

Participants were trained in the use of handheld computers so they could complete the EMA protocol. All participants were allowed practice days (which were not included in analyses) to ensure their comfort level and familiarity with the EMA protocol. Participants then completed the EMA protocol for 2 weeks. During that time, participants were asked to return to the research site approximately two to three times (EMA AN) or three to four times (EMA BN) so that we could download and obtain the data from their handheld computers. Participants were provided feedback on their adherence rates during these visits. In addition to being paid $100 per week for participation in the EMA protocol, participants were eligible to receive a $50 bonus if their adherence rates in responding to signals was greater than 80% (EMA AN) or 85% (EMA BN).

The EMA protocols for the EMA AN and BN groups included signal, interval, and event-contingent recordings. However, just the six semirandom signals per day and one end-of-day signal were included in the analyses because we did not want affect ratings associated with the specific eating-disorder behaviors assessed in the event-contingent signals to bias our measurement of general daily affective dynamics among these groups. This resulted in a total of seven possible signals per day for both the AN and BN groups. Times for semirandom signals were selected within 30 min before or after the following six “anchor” time points: 8:30 a.m., 11:10 a.m., 1:50 p.m., 4:30 p.m., 7:10 p.m., and 9:50 p.m. These randomly selected times were then programmed into the handheld computer at the onset of the study. During all signals, participants responded to a series of questions, including prompts to rate their current mood.

EMA BED

Individuals were recruited from clinics in Fargo and Minneapolis for participation in a clinical trial for an adaptation of integrative cognitive-affective therapy (Wonderlich, Peterson, & Smith, 2015) for the treatment of BED (ICAT-BED; Peterson et al., 2019). After initial screening, participants scheduled an in-person assessment to confirm diagnosis.

Before beginning the psychotherapy treatment trial, all participants completed 7 days of an EMA protocol. During the initial study assessment visit, participants were trained on the use of the EMA program, “Real Time Assessment in the Natural Environment” (ReTAINE; Neuropsychiatric Research Institute, 2019), on their smartphones. Participants were allowed 1 day of practice with the EMA protocol, and those data were not used in final analyses. Following the 1-day trial period, participants completed the EMA protocol over the next 7 days. Research coordinators contacted all participants by phone 2 to 4 days into the EMA protocol to provide feedback regarding their adherence.

The EMA protocol included five semirandom signals per day and one end-of-day signal (for a total of six possible signals per day). Times for semirandom signals were selected within 30 min before or after the following five anchor time points: 9:00 a.m., 12:00 p.m., 3:00 p.m., 6:00 p.m., and 9:00 p.m. These randomly selected times were then programmed into the ReTAINE system at the onset of the study. During all signals, participants responded to a series of questions, including prompts to rate their current mood.

Measures

Diagnostic interviews

Diagnoses for all participants were confirmed using a structured clinical interview conducted by a master’s- or doctoral-level clinician. Within each study, a second independent assessor rated diagnoses from interview recordings for a random sample of 25% of participants, thus allowing for estimates of interrater reliability. For the EMA AN sample, diagnoses were confirmed from the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID I/P; First, Spitzer, Gibbon, & Williams, 1995). Interrater reliability (κ) for a current AN diagnosis was .93. For the EMA BN sample, diagnoses were also confirmed using the SCID I/P. Interrater reliability (κ) for diagnoses among the EMA BN sample was 1.00. For EMA BED, diagnoses were confirmed using the Eating Disorder Examination (EDE; Fairburn & Cooper, 1993) semistructured interview. Interrater reliability (intraclass correlation coefficient), based on a subsample of 20% of study participants selected at random (n = 22), ranged from .842 (shape concerns) to .980 (restraint) on EDE scales and was .940 for the EDE global score.

Daily self-report

Across all studies, participants rated their current emotions in daily diary entries on a 5-point Likert scale (1 = very slightly or not at all, 5 = extremely). For each study, NA items in the EMA protocol were selected in order of highest factor loadings from the Positive and Negative Affect Schedule–Expanded Form (PANAS-X; Watson & Clark, 1994) on the basis of their theoretical or clinical relevance to the targeted population. A total of eight NA items were included in EMA AN (Engel et al., 2013), 11 NA items in EMA BN (Smyth et al., 2007), and 20 NA items in EMA BED (Peterson et al., 2019). Reliability estimates (coefficient α) for the selected NA items in each study ranged from good (.80 in EMA BED) to excellent (.92 for EMA AN and BN; Engel et al., 2013; Schaefer et al., 2019; Smyth et al., 2007). Seven NA items were used in all three studies and were subsequently used for the affective dynamics calculations in the current study: afraid, angry at self, ashamed, disgusted, dissatisfied with self, nervous, and sad.

Affective dynamics

All affective dynamics were calculated at the daily level to allow for novel comparisons of daily affective experience patterns across groups. Although previous group-difference approaches have calculated a single affective dynamic index for each individual (i.e., a “trait” approach; see Trull et al., 2015 for a review), Tomko and colleagues (2015) found unique results when calculating emotion differentiation at the occasion (i.e., signal) and daily levels. Given that the EMA BED group received just 1 week of EMA with six possible signals per day and the EMA AN and BN groups received 2 weeks of EMA with seven possible signals per day, we chose to calculate indices of NA intensity, lability, differentiation, and inertia at the daily level to control for any possible bias related to the systematic difference in duration of the EMA protocol. Furthermore, we sought to preserve the dynamic nature of these constructs and increase power to detect significant effects in analyses by calculating affective dynamics at the daily level.

Intensity

Daily NA intensity was calculated by averaging ratings across the seven NA items during each signal and then taking the average of these scores across the entire day of signaling. Higher scores represent greater levels of daily NA intensity.

Lability

Consistent with the approach taken in previous eating-disorder research (e.g., Berner et al., 2017), daily NA temporal lability was calculated using the mean square successive difference (MSSD). MSSD reflects both increases and decreases in overall ratings of NA variability and temporal dependency as they occur between successive time intervals (e.g., estimating change from Time 1 to Time 2, etc.). We followed procedures outlined by Jahng, Wood, and Trull (2008) and modified the equation for MSSD by using adjusted successive difference (ASD), calculated as

$$\begin{array}{l}\mathrm{ASD}=\frac{x_{(i+1)jk}-x_{ijk}}{{[(t_{(i+1)jk}-t_{ijk})/Mdn(t_{(i+1)jk}-t_{ijk})]}^{\mathrm{\lambda }}};\\ \mathrm{MSSD}=\frac{1}{N-1}{\displaystyle {\sum }_{i=1}^{N-1}{(\mathrm{ASD})}^2,}\end{array}$$

where t_{(i + 1)jk} − t_ijk is the time interval between observation x_{(i + 1)jk} and x_ijk and Mdn(t_{(i + 1)jk} − t_ijk) is the median of time intervals for i, j, and k. x_ijk represents the observation at occasion i within time unit j (day) for individual k. MSSD was then aggregated within day to create a daily average measure of NA lability for each participant. Higher MSSD scores represent greater daily NA lability.

Differentiation

Daily NA differentiation was calculated from intraclass correlation coefficients (ICCs) with absolute agreement across all negative emotion ratings within a given day, which resulted in an average coefficient value for each day of signaling for each participant. The use of ICCs in calculating estimates of emotion differentiation is consistent with previous research (e.g., Kashdan, Ferssizidis, Collins, & Muraven, 2010; Selby et al., 2014; Shrout & Fleiss, 1979). Higher ICC values (representing greater between-signal and lower within-signal variability) indicate that emotion ratings were more closely correlated to one another, representing poor ability to differentiate among emotions. However, for ease of interpretation, NA differentiation scores were reversed by subtracting the ICC value from 1, resulting in higher scores representing better daily NA differentiation.

Inertia

As suggested in previous research (Kuppens et al., 2010; Trull et al., 2015), daily NA inertia was calculated from a within-persons autoregressive estimate within a generalized estimating equation (GEE) model with linear response function.¹ The equation used to calculate daily NA inertia was

$$Y_{jk}{(NA)}_t={\mathrm{\beta }}_{0jk}+{\mathrm{\beta }}_{1jk}{(NA)}_{\left(t-1\right)jk}+r_{jk},$$

where NA at time t − 1 was used to predict NA at time t within time-unit j (day) for individual k. After we aggregated signals by day, this approach yielded a coefficient for each day of signaling, which was subsequently used as the score for NA inertia. Higher scores indicate less daily NA inertia (i.e., greater flexibility of NA over time).

Data-analytic strategy

A series of linear mixed models were conducted to test differences among eating-disorder groups in daily affective dynamics. For instances in which the outcome variable was skewed (i.e., NA intensity and lability), a generalized linear model with a gamma distribution was used to test group differences. Analyses were based on all available data, and missing data were not imputed. Dependent variables for these analyses were daily measures of affect dynamics (i.e., NA intensity, NA lability, NA differentiation, NA inertia). Models included a fixed intercept, main effects for covariates (age, gender, race, signal frequency, other affective dynamics variables), and a fixed effect for eating-disorder group. Marginal means adjusted for covariates were calculated for each group, and a series of pairwise comparisons with Bonferroni-adjusted confidence intervals were employed when testing group differences.

Results

Sample characteristics and preliminary analyses

Averages by sample for age, BMI, and number of signals per day and percentages for gender are found in Table 1. A series of one-way analysis of variance (ANOVA) analyses revealed several significant differences among groups. The BED sample was significantly older than the AN and BN samples, F(2, 359) = 85.76, p < .001. As expected, the BED sample had significantly fewer daily EMA signals than the AN and BN samples, F(2, 364) = 177.67, p < .001. In addition, as anticipated, all groups were significantly different from one another with respect to BMI; the BED sample demonstrated the highest average BMI, followed by the BN sample and then the AN sample, F(2, 364) = 290.71, p < .001.

Given the significant differences by age and number of signals per day, these variables were entered as covariates in subsequent analyses. Gender was also entered as a covariate in analyses to control for the effects of the small sample of men included in EMA BED. Race/ethnicity was entered as an additional covariate in analyses to control for group differences in the overall percentage of participants identifying as White. BMI was not entered as a covariate in analyses because it represents a central theoretical component in each of the DSM–5-based eating-disorder diagnoses; thus, controlling for BMI would unjustly bias our test of diagnostic group comparisons.

In the interest of determining whether daily affective dynamics were significantly related to one another, a series of linear mixed models was conducted within the combined full sample. One affective dynamic was selected as the dependent variable, and the remaining three affective dynamics were entered as main effects into the model. Results produced an unstandardized estimate for each main effect, which represents the unique relationship between the predictor and dependent variable. All three remaining affective dynamics were entered as covariates in subsequent analyses to control for potential effects of the other affective dynamics on the group differences in the affective dynamic outcome of interest (see Table 2).

Table 2.

Relationships Among Daily Affective Dynamics in the Full Sample

Outcome	Estimate (b)	SE	p
Predictor: NA intensity
NA lability	.005	.004	.183
NA differentiation	−.006	.002	.003
NA inertia	.000	.005	.956
Predictor: NA lability
NA intensity	.088	.074	.233
NA differentiation	−.287	.009	< .001
NA inertia	−.301	.021	< .001
Predictor: NA differentiation
NA intensity	−.353	.123	.004
NA lability	−.802	.024	< .001
NA inertia	−.474	.036	< .001
Predictor: NA inertia
NA intensity	.029	.053	.577
NA lability	−.156	.011	< .001
NA differentiation	−.088	.007	< .001

Note: Boldface type indicates statistically significant values (p < .05). NA = negative affect.

Given that there were no missing data at the variable level for completed signals, overall adherence to completing the EMA protocol (otherwise representing the overall instances of missing data in each group) was calculated for each study by dividing the total number of completed random signals (i.e., random signals used in analyses) by the total number of signals delivered throughout the course of the study. Average adherence rates greater than or equal to 80% included 79.4% of the EMA AN sample, 92.8% of the EMA BN sample, and 54.4% of the EMA BED sample. A one-way ANOVA revealed significant group differences in average adherence rates for the EMA protocol (see Table 1).

To investigate whether missing-data rates affected each of our models of interest, we ran a series of pattern mixture models (Hedeker & Gibbons, 1997). To complete these analyses, we dichotomized participants on the basis of their adherence scores (median split) and added this new missing-data variable as a main effect and in an interaction with group into the model. If a significant interaction between the missing-data and group variables were obtained, estimates were averaged across the two adherence groups to create an overall estimate for the group effect on the outcome variable of interest (Hogan & Laird, 1997).

No significant interactions between the missing-data and group variables were obtained for the NA intensity and lability outcomes, which suggests that missing data did not significantly affect our conclusions about the relationship among groups and these two types of affective dynamics. However, we did obtain a significant interaction among the missing-data and group variables for the daily NA differentiation and NA inertia outcomes. Average estimates, standard errors, and significance values were calculated following the procedures described by Hogan and Laird (1997) and compared with the original model (i.e., the model not containing effects for missing data). The estimates and significance values for both NA differentiation and inertia were not different between the averaged model and the original model, which ultimately suggests that missing data did not significantly affect the results for group in the original model. Thus, the original models including pairwise comparisons were retained for both the NA differentiation and inertia analyses.

Differences by eating-disorder diagnosis

NA intensity

Results of a generalized linear model with a gamma distribution that examined the impact of covariates on daily NA intensity revealed several significant fixed effects: age, Wald χ² = 4.24, p = .039; gender, Wald χ² = 21.91, p < .001; signal frequency, Wald χ² = 6.43, p = .01; NA lability, Wald χ² = 37.84, p < .001; and NA inertia, Wald χ² = 12.16, p < .001. The remaining covariates did not emerge as significant (ps = .186–.211). After we controlled for covariates, group emerged as a significant fixed effect, Wald χ² = 25.97, p < .001. Bonferroni-corrected post hoc analyses demonstrated that the AN and BN groups had significantly higher daily NA intensity than the BED group (Table 3). However, there was not a significant difference between the AN and BN groups.

Table 3.

Estimated Marginal Means and Pairwise Comparisons from Linear Mixed Models Examining Group Differences in Affective Dynamics

Variable	Estimated marginal mean						Pairwise comparison
	AN		BN		BED		AN vs. BN			AN vs. BED			BN vs. BED
	M	CI	M	CI	M	CI	M	p	Direction of effect	M	p	Direction of effect	M	p	Direction of effect
NA intensity	2.18 (0.13)	[1.92, 2.46]	2.12 (0.12)	[1.90, 2.38]	1.62 (0.07)	[1.49, 1.77]	0.05 (0.10)	1.0	AN = BN	0.55 (0.13)	< .001	AN > BED	0.50 (0.12)	< .001	BN > BED
NA lability	0.27 (0.03)	[0.22, 0.34]	0.35 (0.04)	[0.28, 0.44]	0.27 (0.03)	[0.23, 0.33]	−0.08 (0.02)	.003	AN < BN	0.00 (0.03)	1.0	AN = BED	0.08 (0.03)	.058	BN = BED
NA differentiation	0.58 (0.05)	[0.48, 0.68]	0.62 (0.05)	[0.52, 0.72]	0.48 (0.05)	[0.38, 0.58]	−0.04 (0.01)	.002	AN < BN	0.10 (0.02)	< .001	AN > BED	0.14 (0.02)	< .001	BN > BED
NA inertia	−0.08 (0.12)	[−0.31, 0.16]	−0.07 (0.12)	[−0.30, 0.17]	−0.19 (0.12)	[−0.42, 0.04]	−0.01 (0.03)	1.0	AN = BN	0.12 (0.05)	.053	AN = BED	0.13 (0.05)	.032	BN > BED

Note: Bold values indicate statistically significant findings (p < .05). AN = anorexia nervosa; BN = bulimia nervosa; BED = binge eating disorder; CI = 95% confidence interval; NA = negative affect.

NA lability

Results of a generalized linear model with a gamma distribution that examined the impact of covariates on daily NA lability revealed several significant fixed effects: signal frequency, Wald χ² = 10.15, p < .001; NA intensity, Wald χ² = 100.25, p = .006; NA differentiation, Wald χ² = 7.42, p = .006; and NA inertia, Wald χ² = 7.54, p = .006. The remaining covariates did not emerge as significant (ps = .096–.466). After we controlled for covariates, group emerged as a significant fixed effect, Wald χ² = 15.36, p < .001. Bonferroni-corrected post hoc analyses demonstrated that the BN group had significantly higher daily NA lability than the AN group (Table 3). However, no significant differences emerged between the AN and BED or between the BN and BED groups.

NA differentiation

Results of a linear mixed model that examined the impact of covariates on daily NA differentiation revealed several significant fixed effects: age, F(1, 3753) = 38.20, p < .001; race, F(1, 3753) = 7.95, p = .005; signal frequency, F(1, 3753) = 42.83, p < .001; NA intensity, F(1, 3753) = 24.10, p < .001; NA lability, F(1, 3753) = 1426.22, p < .001; and NA inertia, F(2, 3753) = 162.19, p < .001. The remaining covariate (gender) did not emerge as significant (p = .475). After we controlled for covariates, group emerged as a significant fixed effect, F(2, 3753) = 21.13, p < .001. Bonferroni-corrected post hoc analyses demonstrated that the BN group had significantly higher daily NA differentiation than the AN group. In addition, both the AN and BN groups demonstrated significantly higher daily NA differentiation than the BED group (Table 3).

NA inertia

Results of a linear mixed model that examined the impact of covariates on daily NA inertia showed that daily NA lability and NA differentiation emerged as significant fixed effects: F(1, 3753) = 159.72, p < .001 and F(1, 3753) = 162.19, p < .001, respectively. The remaining covariates did not emerge as significant effects (ps = .070–.916). After we controlled for covariates, group emerged as a significant fixed effect, F(2, 3753) = 3.31, p = .036. Bonferroni-corrected post hoc analyses demonstrated that the BN group had significantly higher scores on inertia compared with the BED group (Table 3). However, there were no significant differences between the AN and BN groups or between the AN and BED groups.

Discussion

In this exploratory investigation, we examined whether significant group differences in daily NA dynamics emerged among individuals diagnosed with one of three different eating disorders (i.e., AN, BN, or BED). Results demonstrated several interesting patterns of findings. The AN and BN groups had significantly higher daily NA intensity than the BED group. However, individuals with BED had poorer NA emotion-differentiation abilities than individuals with AN or BN. The BN group was significantly higher than the AN group on daily NA lability and differentiation, which suggests that individuals with BN experience more labile NA but are better NA differentiators. Finally, the BN group had higher scores for daily NA inertia than the BED group, which suggests that individuals with BN experience more persistent NA compared with those with BED.

Anorexia Nervosa Versus Bulimia Nervosa

The AN and BN groups differed significantly on scores for daily NA lability and differentiation but did not differ on scores for daily NA intensity and inertia. Previous research regarding trait-based personality dimensions has suggested that individuals with BN are more likely to be undercontrolled or emotionally dysregulated and that individuals with AN are likely to be overcontrolled or restricted (Claes et al., 2006; Westen & Harnden-Fischer, 2001). Given that NA lability is a prominent clinical feature of individuals who experience pervasive issues with emotion regulation (e.g., as seen with borderline personality disorder; American Psychiatric Association, 2013; Ebner-Priemer et al., 2007), the results from the current study (i.e., those demonstrating that the BN group scored higher on NA lability than the AN group) are consistent with the trait-based research in the eating-disorder literature. However, for NA differentiation, the BN group members displayed a greater ability to differentiate emotions, which suggests that they might have greater access to cognitive resources by which to inform their emotion regulation than individuals with AN. Previous research has also linked poor emotion differentiation with broader or more pervasive emotion-regulation difficulties (e.g., Hill & Updegraff, 2012; Suvak et al., 2011; Zaki, Coifman, Rafaeli, Berenson, & Downey, 2013). Thus, these empirical patterns ultimately reflect that momentary-emotion-regulation difficulties might manifest differently among individuals with AN as opposed to those with BN, such that individuals with BN may struggle with unpredictable emotions on a moment-to-moment basis, whereas individuals with AN may struggle to identify nuances in their momentary emotional experience.

Binge Eating Disorder Versus Anorexia Nervosa and Bulimia Nervosa

Regarding comparisons between the BED group and the AN and BN groups, significant differences emerged for daily NA intensity, differentiation, and inertia. Individuals with BED had significantly lower scores on daily NA intensity than did individuals with AN or BN, which suggests that individuals with AN or BN experience stronger negative emotions on a daily basis than do individuals with BED. Findings for NA inertia suggest that intensity of NA is more persistent for individuals with BN compared with individuals with BED. According to previous research on emotional inertia and psychological maladjustment (e.g., Kuppens et al., 2010), the combination of high levels of NA intensity and high levels of NA inertia are particularly suggestive of severe affective psychopathology, such as depression. Given that the BN group had higher scores for NA intensity and inertia, these findings suggest that individuals with BN may be at a higher risk for emotion regulation difficulties and may subsequently exhibit more severe psychopathology relative to individuals with BED.

In contrast, individuals with BED exhibited poorer daily emotion-differentiation abilities than individuals with AN or BN. In light of previous research suggesting that lower levels of emotion differentiation are linked to psychopathology and maladaptive behaviors (for a review, see Trull et al., 2015), our findings suggest that the unique affective psychopathology of individuals with BED may be linked to deficits in emotion differentiation rather than issues related to NA intensity, lability, or inertia.

Research and clinical implications

Taken together, these results regarding group differences in daily affective dynamics may expand our understanding of emotion-regulation differences among the eating-disorder diagnoses. For example, daily affective lability may be used to significantly distinguish BN from AN. On the other hand, daily NA intensity may be used to distinguish BED from both AN and BN, and daily NA inertia may be used to differentiate BED from BN. Ultimately, these findings may contribute information that is useful in distinguishing among eating-disorder diagnoses and subsequently inform future eating-disorder classification schemes.

The present findings also have several implications toward furthering the measurement and conceptualization of affect among the eating disorders. Past research in the eating disorders has examined conceptually similar affect constructs using traditional, nonmomentary, retrospective self-report measures. For example, group differences in alexithymia (i.e., pervasive issues with identifying, describing, and communicating information regarding emotions) have been studied in the eating disorders (for a review, see Nowakowski et al., 2013). Lability, when measured as a trait-level personality construct, has also been compared across eating-disorder diagnostic groups (Goldner et al., 1999; Westen & Harnden-Fischer, 2001). Furthermore, specific trait-based facets of difficulties with emotion regulation, such as emotional awareness (i.e., paying attention to emotions when they occur) and clarity (i.e., understanding the emotions one is experiencing), have been compared across eating-disorder diagnoses (Brockmeyer et al., 2014; Lavender et al., 2015; Svaldi, Griepenstroh, Tuschen-Caffier, & Ehring, 2012). However, these findings have limited ecological validity because they require participants to report on their own aggregated emotional experience. Indeed, two studies compared traditional self-report and EMA measures of affective lability among eating-disorder samples and demonstrated that EMA lability is a stronger predictor of several disordered eating behaviors than is self-report lability (Anestis et al., 2010; Lavender et al., 2013). Thus, the findings from these studies combined with the results of the current study suggest that momentary affective dynamics may provide more empirically rich data than measuring similar constructs using traditional, retrospective self-report.

The findings from the current study may also have clinical utility. In terms of treatment, virtually all contemporary affect-focused treatments for eating disorders—for example, cognitive behavioral therapy–enhanced (CBT-E; Fairburn, 2008), dialectical behavior therapy for binge eating and bulimia (DBT; Safer et al., 2009), and integrative cognitive-affective therapy (ICAT; Wonderlich et al., 2015)—tend to rely on intensity as the core clinical emotion metric. However, these data suggest that additional emotion metrics, including lability, differentiation, and inertia, could be considered as potential clinical targets and mechanisms for eating-disorder behaviors and related symptoms. Existing clinical treatments can also be adapted to specific eating-disorder diagnostic group. For example, interventions that target affect intensity may need to be adapted to target affect lability for individuals diagnosed with BN. Alternatively, treatments for BED may benefit from emphasizing emotion-awareness skills (e.g., mindfulness).

Limitations and future directions

Given the exploratory design of the current study, there are several important areas for future directions that should be noted. In addition, although this study demonstrates several strengths, including the use of EMA data to calculate daily affective dynamics and relatively large sample sizes for each of the three eating-disorder diagnostic groups, several limitations should be addressed.

Despite the emergence of significant differences among the eating-disorder diagnostic groups, it has yet to be determined whether individuals with eating-disorder psychopathology are different from other groups without eating disorders in daily affective dynamics. For example, it is possible that the differences among eating-disorder diagnostic groups found in this study are partially explained by the existence of comorbid psychopathology (e.g., mood, anxiety, and substance-use disorders). Furthermore, it is possible that the levels of each affective dynamic in these specific groups are no different from those observed among healthy individuals without psychopathology. Future research should seek to collect additional momentary affective data to explore whether eating-disorder diagnostic groups have deficits in daily affective dynamics relative to healthy participants or groups of individuals with other types of psychopathology.

The results of the current study did not take into account the relationship among the indices of affective dynamics and specific eating-disorder behaviors. Given the extent to which the eating-disorder literature has examined prospective relationships among NA intensity and a variety of disordered eating behaviors, future research should comprehensively examine and compare the links among NA lability, differentiation, and inertia with core behavioral eating-disorder symptoms (e.g., binge eating, purging, and dietary restriction). By examining these prospective relationships, future research may be able to determine whether specific indices of affective dynamics are differentially linked to discrete behaviors (e.g., NA lability may be more strongly associated with binge eating, whereas NA differentiation may have a greater link with compensatory behaviors).

One set of limitations to the current study includes issues related to the measurement of affect. There was no standardized procedure to select the NA items used in the calculation of affective dynamics. Although the same NA items were included in calculations for each of the three diagnostic samples, the number and combination of items likely differ from those used in previous affective-dynamics studies (see Trull et al., 2015), which limits generalizability of findings. In fact, similar concerns have been raised regarding the lack of standardization and consistency in the selection of affect items within the broader affective-dynamics literature (Brose, Schmiedek, Gerstorf, & Voelkle, 2020). Future research examining affective dynamics in the eating disorders should seek to establish a standardized set of affect items to be used in EMA data collection. In addition, positive affect items were not available for the calculation of affective dynamics. Given that past research, both in the eating-disorders field and the broader literature on affective dynamics, has examined positive affect (e.g., Haedt-Matt & Keel, 2011; Trull et al., 2015), future research on affective dynamics in the eating disorders should seek to include positive-affect items in EMA data collection.

Another set of limitations involves methodological differences among the EMA protocols. For example, the EMA BED study had the only group of participants that included men and that were seeking treatment, which may have affected our group-difference findings. Furthermore, the EMA BED data were collected approximately 13 years after the EMA BN study ended; thus, the span of time between the EMA protocols may have also affected our findings. Future research should seek greater consistency across samples (e.g., all treatment-seeking individuals, both women and men, greater diversity of participant ethnic and cultural backgrounds, and concurrent data collection). Finally, the systematic differences in the duration and number of EMA signals among the three groups led us to consider whether 1 week provided enough time to adequately capture affective dynamics within the EMA BED group. Subsequently, attempts were made to minimize these systematic differences by calculating the affective dynamics at the daily level, using a multilevel framework to test group differences, and controlling for the total number of signals within a single day. However, it should be noted that calculating affective dynamics at the daily level precludes direct comparisons with data from existing studies in which affective dynamics were calculated at the trait level. Further research is needed both to replicate the findings from the current study and to expand similar daily affective-dynamic comparisons to other diagnostic groups.

Conclusions

This study is the first to use EMA data to comprehensively calculate and compare multiple daily affective dynamics across eating-disorder diagnostic groups. Results demonstrated several significant group differences that suggest different patterns of affective experience among the eating-disorder diagnostic groups. These findings extend the previous work regarding affective intensity in the eating disorders and serve as the first step in examining the impact of multiple types of daily affective dynamics on maladaptive behaviors and related symptoms among individuals diagnosed with an eating disorder. Future research should seek to examine these affect dynamics in relation to additional disordered eating behaviors among the three diagnostic groups, which may ultimately provide further insight into the role of emotion regulation in eating-disorder psychopathology.

Transparency

Action Editor: Kelly L. Klump

Editor: Scott O. Lilienfeld

Author Contributions

G. A. Williams-Kerver developed the study concept. Initial testing and data collection were supported by S. A. Wonderlich, R. D. Crosby, S. G. Engel, C. B. Peterson, S. J. Crow, J. E. Mitchell, and D. Le Grange. G. A. Williams-Kerver and L. Cao performed data analysis and interpretation under the supervision of S. A. Wonderlich and R. D. Crosby. G. A. Williams-Kerver drafted the manuscript, and S. A. Wonderlich, R. D. Crosby, S. G. Engel, K. E. Smith, C. B. Peterson, and D. Le Grange provided critical revisions. All of the authors approved the final manuscript for submission.