Affect, interoception, and disordered and intuitive eating behaviors: Examining momentary mediational associations among women with eating disorder pathology

Abstract

Objective:

Affect and interoception (the ability to sense/connect with bodily sensations) are transdiagnostic self-regulatory factors that promote maladaptive eating behaviors at the between-person level (via cross-sectional research). However, no research has examined the mechanistic role of interoception dimensions in associations between affect, and disordered and intuitive eating behaviors (DEBs, IEBs) at the within-person level via ecological momentary assessment (EMA).

Method:

The present study addressed this research gap by examining whether: (1) momentary interoception dimensions (“not distracting,” “self-regulation,” “body listening and trust”) differentially mediated momentary associations between negative and positive affect (predictors), and DEBs (dietary restriction, overeating, loss of control eating) and IEBs (“eating for physical versus emotional reasons,” “reliance on hunger and satiety cues,” “unconditional permission to eat,” “body-food choice congruence”) outcomes; (2) these associations varied across affect, interoception, and eating behavior types. Women (N = 150) with eating disorder pathology (M_age = 20.95, SD = 4.14) completed 4 EMA surveys on mobile devices daily for 10 days.

Results:

Multilevel structural equation models showed that greater momentary negative affect was associated with more distracting from uncomfortable/painful bodily sensations at subsequent assessments (i.e., less “not distracting” interoception) and, in turn, less “eating for physical versus emotional reasons” IEB later that day. Greater momentary positive affect was associated with greater subsequent “body listening and trust” interoception and, subsequently, greater “reliance on hunger and satiety cues” IEB. No momentary indirect effects linking affect, interoception, and DEBs were significant.

Conclusions:

These results may be leveraged to optimize eating disorders research and novel real-time interventions.

1.0. Introduction

Eating disorders (EDs) are prevalent health concerns that are associated with high mortality risk and psychophysiological morbidity (Romano et al., 2022; Udo & Grilo, 2019; van Hoeken & Hoek, 2020). It is well-established that individuals with greater ED pathology engage in disordered eating behaviors (DEBs; e.g., loss of control eating) more frequently than individuals with less ED pathology (Romano et al., 2020, 2022). Likewise, individuals with greater ED pathology less frequently engage in intuitive eating behaviors (IEBs; i.e., health-promoting eating behaviors characterized by eating based on hunger and satiety cues versus other factors, such as cognitive dietary/food rules and affect) than individuals with less ED pathology (Linardon et al., 2021). Improving the understanding of factors that are associated with both maladaptive DEBs and adaptive IEBs may provide important nuanced insight that could be leveraged for testing as targets in future EDs prevention and treatment efforts. This could, in turn, help decrease individuals’ risks for adverse health long-term.

1.1. Affect and Disordered Eating Behaviors

Prior research that has examined state-based factors that are associated with ED pathology in daily life has consistently identified negative and positive affect as central processes that can contribute to DEB maintenance (Berg et al., 2017; Schaefer et al., 2020; Wonderlich et al., 2022). Such research has been informed by affect-focused EDs theories, which posit that worsening affect in the hours prior to and improved affect following DEBs can promote conditioned responses that maintain these behaviors (Hawkins & Clement, 1984; Mason et al., 2021). In line with these theoretical tenets, research using ecological momentary assessment (EMA) and multilevel modeling (see Berg et al., 2017) has generally shown that negative affect increases and, in some cases, positive affect decreases in the hours preceding individuals’ disinhibited DEBs (e.g., loss of control eating, binge eating), dietary restriction, and compensatory behaviors (Berg et al., 2017; Haedt-Matt & Keel, 2011; Schaefer et al., 2020; Wonderlich et al., 2022). In turn, negative affect generally decreases and, in some cases, positive affect increases immediately following these DEBs (Berg et al., 2017; Haedt-Matt & Keel, 2011; Schaefer et al., 2020; Wonderlich et al., 2022).

1.2. Interoception

Although findings from the extant EMA literature consistently indicate that momentary affective states and their trajectories could be critical treatment targets, empirically supported interventions for EDs that target affect dysregulation generally result in suboptimal outcomes, with over half of individuals remaining symptomatic post-treatment and approximately two-thirds relapsing longer-term (Linardon, 2018; Linardon & Wade, 2018; Sala et al., 2023). It is plausible that additional factors beyond dynamic moment-to-moment or day-to-day changes in affect may promote individuals’ engagement in DEBs in their natural environments (e.g., while at home, work/school) and, if addressed in future interventions, could improve treatment outcomes. Transdiagnostic self-regulatory processes may be especially important to consider, particularly interoception.

Although operationalized in different ways across disciplines, interoception is generally conceptualized as the ability to sense and integrate information from one’s bodily sensations, such as hunger, satiety, heartbeat, and pain (Garfinkel et al., 2015; Khalsa et al., 2018, 2022; Mehling et al., 2018). The measurement of interoception includes distinct objective (assessed via in-lab heartbeat perception tasks and electrocardiograms) and subjective (self-reported via surveys) dimensions (Garfinkel et al., 2015). Based on Garfinkel and colleagues’ (2015) conceptualization, objective dimensions of interoception include “interoceptive accuracy” (i.e., individuals’ abilities to accurately detect their internal bodily sensations) and “interoceptive awareness” (i.e., individuals’ metacognitive awareness of their interoceptive accuracy; Garfinkel et al., 2015). Subjective dimensions of interoception are broadly referred to as “interoceptive sensibility” (i.e., individuals’ self-perceptions of their ability to be internally self-focused and interoceptively aware in varied ways) and, based on Mehling and colleagues’ (2018) conceptualization, includes eight different dimensions: Noticing; Not Distracting; Not Worrying; Attention Regulation; Emotional Awareness; Self-Regulation; Body Listening; Body Trust (Garfinkel et al., 2015; Mehling et al., 2018). Difficulties with objective and subjective interoception dimensions are believed to uphold vital roles in the development and maintenance of EDs by promoting deficits in individuals’ abilities to accurately sense and respond to their hunger and satiety cues, and via experiential avoidance of eating- and body-related interoceptive sensations (Brown et al., 2017; Khalsa et al., 2018, 2022).

Cross-sectional research with ED samples has indicated that greater deficits in interoception dimensions are robustly linked to worse (i.e., higher negative and lower positive) affect at the between-person level (Brown et al., 2017; Khalsa et al., 2022). Also, a growing evidence-base has focused on examining (primarily bivariate) associations between different interoception dimensions, DEBs, and IEBs at the between-person (i.e., cross-sectional) level. This cross-sectional research has generally found that individuals with greater ED pathology exhibit greater deficits in objectively and subjectively measured interoception dimensions than individuals with less ED pathology (Herbert et al., 2021; Jenkinson et al., 2018; Olatunji et al., 2018; Pollatos & Georgiou, 2016). Likewise, initial cross-sectional research has generally identified positive between-person associations between objectively measured interoception dimensions and IEBs among non-clinical and EDs samples (DeVille et al., 2021; Herbert et al., 2013; Richard et al., 2019), and positive between-person associations between subjective interoception dimensions and IEBs among non-clinical samples (Jeune et al., 2024; Oswald et al., 2017; Robinson et al., 2021). However, notable variations in the latter pattern have been identified based on the subjective interoception dimension and IEB under consideration (see below; Herbert et al., 2013; Herbert, 2021; Jeune et al., 2024; Oswald et al., 2017). Collectively, these between-person findings suggest that people with greater interoception exhibit lower DEB severity and greater IEBs.

1.3. Areas Warranting Further Assessment

There are two vital research gaps in the DEBs and IEBs evidence-base that warrant further examination. First, as discussed above, EMA-assessed within-person associations between worse momentary affect (i.e., higher negative and lower positive affect) and greater momentary engagement in DEBs are well-established (Berg et al., 2017; Haedt-Matt & Keel, 2011; Schaefer et al., 2020; Wonderlich et al., 2022), as are bivariate between-person (cross-sectional) associations among worse affect and interoception dimensions relative to worse DEBs and IEBs (Herbert et al., 2021; Jenkinson et al., 2018; Khalsa et al., 2022). Although transdiagnostic theories posit that deficits in interoception can promote the onset and maintenance of EDs (Herbert, 2021; Khalsa et al., 2022), and two initial EMA studies have examined associations between momentary interoception and DEBs (Romano et al., 2024; Velkoff & Smith, 2023a,b), to our knowledge no prior research has used EMA to examine whether distinct types of momentary interoception differentially mediate momentary associations between negative and positive affect relative to DEBs or IEBs in the natural environment. Directly evaluating whether these theoretically supported mechanistic associations (Herbert, 2021; Khalsa et al., 2022) emerge at the momentary level could provide information about how individuals’ experiences of their bodily states influence their eating behaviors in real-time, and help identify novel treatment targets.

Second, subjectively measured interoception (via self-report surveys; see above) includes several dimensions that assess individuals’ self-perceptions of their abilities to accurately perceive, connect with, and use information from their bodily sensations in different ways (Garfinkel et al., 2015; Mehling et al., 2018). For example, some of these dimensions capture the extent to which individuals allow themselves to (1) experience (versus distract or experientially distance from) uncomfortable bodily sensations and (2) experience their bodies and bodily sensations as safe and trustworthy (Garfinkel et al., 2015; Mehling et al., 2018). Despite the multi-dimensionality of this construct (Garfinkel et al., 2015; Khalsa et al., 2018, 2022; Mehling et al., 2018), in one of the only two existing studies that used EMA to examine associations between interoception and eating behaviors in the natural environment (via bivariate associations only, not mediation), interoception was assessed using one survey item that broadly asked participants how connected they felt to their bodies (Velkoff & Smith, 2023a,b). Established between-person research has indicated that subjective interoception dimensions capturing individuals’ abilities to (1) trust their bodies and (2) listen to their bodies for insight and information exhibit especially strong inverse associations with ED pathology (Brown et al., 2017, 2020; Martini et al., 2021). Also, previous between-person research—but no prior within-person research using EMA—has identified positive associations between different subjective interoception dimensions and IEBs, and found considerable variations in these associations based on the interoception dimensions and IEBs under consideration (Jeune et al., 2024; Oswald et al., 2017; Robinson et al., 2021). Interoception dimensions (e.g., body trust, body listening) and IEBs (e.g., how much individuals rely on hunger and satiety cues to guide their eating, and eat for physical versus emotional reasons) capturing more direct connections with bodily states appear to be particularly strongly linked at the between-person level (Herbert et al., 2013; Herbert, 2021; Jeune et al., 2024; Oswald et al., 2017), compared to associations between constructs capturing more distal connections (e.g., “not worrying” [while experiencing adverse physiological sensations] interoception, how much one gives oneself “unconditional permission to eat” intuitive eating). Examining the comparative salience of different interoception dimensions as mechanisms that may differentially explain individuals’ engagement in different DEBs and IEBs in the natural environment can extend findings from between-person research. Such research can provide insight into processes that may promote individuals’ eating behaviors on a day-to-day basis and answer novel within-person research questions, such as “at times when individuals report worse affect than usual, do they subsequently report worse interoception and, in turn, greater DEBs and less IEBs later that day?”

1.4. Study Purpose

The present study aimed to examine whether: (1) different types of momentary interoception mediated momentary associations between negative and positive affect relative to women with ED pathology’s engagement in distinct DEBs (dietary restriction, loss of control eating, overeating); (2) different types of momentary interoception mediated momentary associations between negative and positive affect relative to women with ED pathology’s engagement in distinct IEBs (“eating for physical versus emotional reasons,” “reliance on hunger and satiety cues,” “unconditional permission to eat,” “body-food choice congruence”) in the natural environment; (3) these associations varied due to the affect valence, interoception dimension, and eating behavior under consideration. It was hypothesized that at times when women reported greater negative and less positive affect than usual, they would subsequently report worse interoception in the next several hours and, in turn, greater DEBs and less IEBs use later that day (Herbert, 2021; Herbert et al., 2013; Khalsa et al., 2022). Mediational pathways including IEBs as outcomes were expected to exhibit stronger associations than those with DEBs because IEBs are predicated upon internal experiencing—a core component of interoception—whereas engagement in DEBs often involves experiential avoidance (i.e., attempts to distance oneself from one’s bodily sensations; Herbert, 2021; Khalsa et al., 2022; Tylka & Kroon Van Diest, 2013). Given the limited prior research in this area, it was unclear whether these mediational pathways would differ in magnitude relative to specific IEBs and DEBs outcomes. Consequently, whether these pathways differed in relation to specific eating behavior outcomes served as an exploratory aim.

2.0. Method

2.1. Participants and Procedures

Institutional review board approval (approval number: 1529089–2, 1/2/2020) was obtained and all procedures were performed in accordance with the ethical standards in the 1964 Declaration of Helsinki and its amendments. All participants provided electronic informed consent prior to study participation. Women with elevated ED pathology were recruited through an undergraduate psychology research pool and completed screening items within a broader baseline survey administered via Qualtrics to determine whether they met eligibility criteria. Eligibility criteria included: (1) identified as women; (2) ages 18–35; (3) elevated ED pathology (EDE-Q Global scores ≥2.3; Fairburn & Beglin, 1994; Mond et al., 2004). This EDE-Q cutoff aptly identified ED cases among community-based women (Mond et al., 2004), and exceeds mean EDE-Q Global scores for young women (Luce et al., 2008; Mond et al., 2006). Also, individuals who identified as women were selected because women aged 18–35 years exhibit elevated ED-risk and ED-related impairment (Udo & Grilo, 2018). Eligible participants were emailed instructions about how to download and use the LifeData app on their smartphones, from which the EMA surveys were remotely deployed. For the following 10 days, participants received prompts from the LifeData app to complete four brief (3–5-minutes), semi-randomly prompted surveys per day that were sent approximately every 4 hours throughout each study day.

An a priori power analysis indicated that ≥139 participants were needed to detect small-to-medium effects and power ≥ 0.80. This translates to an N_effective of 164–253 participants (West et al., 2011). Given that the present sample included 150 women, a sample of this size was sufficient to test all study aims.

2.2. Measures

2.2.1. Affect

Five negative affect (guilty, distressed, nervous, scared, upset) and four positive affect (excited, strong, enthusiastic, alert) items from the Positive and Negative Affect Schedule (PANAS; Watson et al., 1988) assessed participants’ affect (1=Not at all, 5=Extremely). Higher scores reflect greater negative and positive affect, respectively. These items were selected due to their frequent use in prior EMA studies assessing DEBs (Berg et al., 2017; Haedt-Matt & Keel, 2011). Prior research has shown that the PANAS exhibits good internal consistency, acceptable-to-good test-retest reliability, and good convergent validity with other affect measures among college, clinical, and community-based samples (Watson et al., 1988). Multilevel confirmatory factor analyses (MCFA) confirmed that a two-factor negative and positive affect structure exhibited good fit for the present data.

2.2.2. Interoception

Following minor adaptations in item wording to capture momentary- vs. trait-level experiences, 8-items from the Multidimensional Assessment of Interoceptive Awareness-2 (MAIA-2: Mehling et al., 2018) were used to assess four interoception dimensions (2-items per subscale; 0=Not at all true, 5=Completely true): (1) “not distracting” (not distracting oneself from uncomfortable bodily sensations); (2) “self-regulation” (regulating distress by turning attention to one’s bodily sensation, e.g., via deep breathing); (3) “body listening” (actively listening to bodily sensations for information); (4) “body trust” (experiencing one’s body as safe and trustworthy). Higher scores reflect better interoception. An example item from the “body trust” subscale that was used in the present study includes “Since my last survey, I trusted my body sensations.” Prior research has shown that these dimensions exhibit strong associations with ED symptoms in EDs samples (Brown et al., 2017, 2020). The 8-items were selected, as they have exhibited high factor loadings in EDs, community-based, and college samples (Brown et al., 2017; Mehling et al., 2018). MCFA indicated that a three-factor structure combining the Body Listening and the Body Trust subscales into one “body listening and trust” factor exhibited good fit in the present sample.

2.2.3. Disordered Eating Behaviors

Six-items assessed how strongly participants agreed that they had engaged in DEBs since their last EMA surveys (0=Strongly Disagree, 6=Strongly Agree): (1) overeating (consuming an usually large amount of food, given the circumstances; 1-item); (2) loss of control eating (feeling that you cannot stop eating after starting; 1-item); (3) dietary restriction (skipped meals, refused food/drinks, replaced meals with low/no calorie substances, and limited food amount to control weight/shape; 4-items). Higher scores reflect greater DEBs. The items were adapted for the momentary-level from well-established trait-level EDs measures that have exhibited good psychometric properties among adults with ED pathology (Fairburn & Beglin, 1994; Forbush et al., 2013; Garner et al., 1982), prior EMA research (Haedt-Matt & Keel, 2011; Schaefer et al., 2020; Wonderlich et al., 2022), and are consistent with DSM-5-TR DEB conceptualizations (American Psychiatric Association, 2022). MCFAs showed that the best-fitting factor structure was a unidimensional dietary restriction factor with the four dietary restriction items and that the single overeating and loss of control eating items warranted separate examination.

2.2.4. Intuitive Eating Behaviors

Nine-items from the Intuitive Eating Scale-2 (IES-2; Tylka & Kroon Van Diest, 2013) were adapted for the momentary level to assess participants’ engagement in the following IEBs (1=Strongly Disagree, 5=Strongly Agree): (1) “eating for physical rather than emotional reasons” (e.g., eating to quell hunger versus negative affect; 2 items); (2) “unconditional permission to eat” (giving oneself unconditional permission to eat versus relying upon food rules; 2 items); (3) “reliance on hunger and satiety cues” (using hunger and satiety cues to guide one’s eating behaviors; 3 items); (4) “body-food choice congruence” (eating foods that align with one’s nutritional and energy-related needs; 2 items). Higher scores reflect greater IEBs. An example item from the “reliance on hunger and satiety cues” subscale that was used in the present study includes “Since my last survey, I listened to my body to tell me when to eat.” The 9-items were selected given evidence of high factor loadings on their respective subscales in college samples and their theoretical relevance for assessing within-person IEBs (Tylka & Kroon Van Diest, 2013). The IES-2 has exhibited good construct validity, internal consistency, test-retest reliability, and convergent validity with measures of ED symptoms among college students (Tylka & Kroon Van Diest, 2013). MCFAs indicated that this four-factor structure exhibited good fit for the data.

2.3. Statistical Analyses

All analyses were run using R 4.2.0 using the lavaan, lme4, and lmerTest packages. Prior to hypothesis testing, intercorrelations were examined among the assessed constructs to determine if there was a high degree of empirical overlap. As shown in Table S2 of the Supplement, the resultant intercorrelations indicated that, although there is some (expected) overlap between certain assessed constructs, there is also a high degree of variance that was not accounted for via these correlations. Also, prior to hypothesis testing, missing data analyses were run following recommended procedures (Enders, 2022). Results showed that the data were generally missing at random relative to participants’ compliance. Compliance was controlled in all models to adjust for this form of missingness, per recommendations (Enders, 2022). All models used maximum likelihood estimation. All residual-based multivariate assumption tests (e.g., linearity, homoscedasticity, multivariate normality, etc.) were met. Also, all constructs of interest were approximately normally distributed at the univariate level, as evidenced by normally distributed histograms. Notably, any minor deviations from univariate normality were not expected to bias the results, given the large sample size at level 1 and use of a model estimator (maximum likelihood estimation) that is robust to small to moderate deviations from normality (Curran et al., 1996).

Second, MCFAs were run to identify the best-fitting factor structures. This included using deviance tests to assess alternative factor structures that combined EMA items and factors for parsimony. As described in the Supplement, results indicated that the structures specified in the Measures section exhibited the best-fit for the data and were used in the analyses.

Third, four two-level (L) multilevel structural equation models (MSEMs) with a 1–1-1 structure were run using maximum likelihood estimation and latent variable decomposition (Lüdtke et al., 2008). Momentary observations at L1 were nested within participants at L2. MSEMs controlled for compliance (given the missingness analyses) and included momentary negative and positive affect as predictors, interoception dimensions as mediators, and (separately) eating behavior outcomes (dietary restriction, overeating, loss of control eating, and the four IEBs).

To appropriately assess the mediational effects (Kline, 2023), indicators for predictors’ latent factors were lagged twice (time [t]-2), indicators for mediators’ latent factors were lagged once (t-1), and outcomes were not lagged. The last survey on the previous day was not lagged to the first survey the next day, nor were non-consecutive surveys when participants missed surveys. Direct effects from predictors to outcomes were included at L1 and L2. Due to the data structure’s complexity and convergence difficulties that emerged when random slopes were added, fixed slopes and random intercepts were modeled at both levels. Further, although additional autoregressive effects were tested (outcomes at t-1 and t-2, mediators at t-2), convergence difficulties prevented their inclusion.

Inadequate fit that emerged during the model building process (described in the Supplement) was addressed by examining sources of misfit and taking recommended actions (Kline, 2023). One modification was made at a time to gauge each changes impact, ensure that all modifications were justified theoretically, and were only made if the Akaike Information Criteria and sample-size adjusted Bayesian Information Criteria decreased (Kline, 2023). Adequate-to-good fit for the MCFAs and MSEMs was defined as Comparative Fit Index and Tucker Lewis Index >0.90, Root Mean Square Error of Approximation <0.08, and Standardized Root Mean Square Residuals <0.08 (Kline, 2023).

3.0. Results

3.1. Participant Characteristics

Table 1 presents descriptive statistics for the 150 women. On average, participants had a BMI of 28.13 (SD = 6.67) and were 20.95 years old (SD = 4.14). Most identified as White (n = 94, 62.67%), approximately half as “only heterosexual” (n = 72, 48.00%), and approximately one-fifth as Hispanic/Latinx (n = 26, 17.33%).

Table 1.

Participant Characteristics and Descriptive Statistics

	M (SD)	n (%)	Range	ICC
Age	20.95 (4.14)		18.00–35.00
Racial Identity
American Indian or Alaska Native		1 (0.67%)
Asian or Asian American		11 (7.33%)
Black or African American		20 (13.33%)
Native Hawaiian or Other Pacific Islander		1 (0.67%)
Other Race		7 (4.67%)
Multi-Racial		16 (10.67%)
White		94 (62.67%)
Latinx		26 (17.33%)
Sexual Identity
Only Heterosexual		72 (48.00%)
Mostly Heterosexual		28 (18.67%)
Bisexual		36 (24.00%)
Mostly Homosexual, Lesbian, or Gay		3 (2.00%)
Only Homosexual, Lesbian, or Gay		6 (4.00%)
Other Sexual Identity		3 (2.00%)
Prefer Not to Answer		2 (1.33%)
Eating Disorder Examination Questionnairea
Global Score	3.73 (0.87)		2.45–5.82
Restraint Subscale	3.09 (1.56)		0.00–6.00
Eating Concern Subscale	2.56 (1.31)		0.00–5.80
Shape Concern Subscale	4.44 (1.03)		1.88–6.00
Weight Concern Subscale	4.17 (1.01)		1.80–6.00
Body Mass Index	28.13 (6.67)		17.57– 52.47
Affect
Positive Affect	8.81 (3.65)		1.00–20.00	0.42
Negative Affect	8.97 (4.51)		1.00–25.00	0.48
Interoception
Not Distracting	3.26 (1.53)		0.00–5.00	0.55
Self-Regulation	1.67 (1.39)		0.00–5.00	0.63
Body Listening and Trust	2.13 (1.29)		0.00–5.00	0.65
Disordered Eating Behaviors
Dietary Restriction	1.32 (1.51)		0.00–6.00	0.51
Loss of Control Eating	0.91 (1.57)		0.00–6.00	0.32
Overeating	0.67 (1.34)		0.00–6.00	0.25
Intuitive Eating Behaviors
EPER	4.40 (0.95)		1.00–5.00	0.37
UPE	2.85 (1.24)		1.00–5.00	0.38
RHSC	3.02 (1.14)		1.00–5.00	0.57
BFCC	2.71 (1.18)		1.00–5.00	0.34

Note. EPER = eating for physical rather than emotional reasons intuitive eating; UPE = unconditional permission to eat intuitive eating; RHSC = reliance on hunger and satiety cues intuitive eating; BFCC = body-food choice congruence intuitive eating; M reflects grand means.

^aValues on the Eating Disorder Examination Questionnaire reflect participants’ responses to the trait-level version of this questionnaire that was administered to participants at baseline, prior to the ecological momentary assessment portion of the present study.

Participants responded to 4,261 out of 6,000 possible EMA surveys, reflecting good compliance (compliance rate = 71.02%; missing data = 28.98%). Intraclass correlation coefficients indicated that 25%–51% of the variability in DEBs and 34%–57% in the IEBs stemmed from trait-level between-person differences between participants, whereas the remaining 49%–75% and 43%–66% (respectively) was due to moment-to-moment variability (plus error). See Table 1. Also, as shown in Table S2 of the Supplement, at the within-person level intercorrelations ranged from 0.03 to 0.44 and, at the between-person level, intercorrelations ranged from −0.01 to 0.57.

3.2. Affect, Interoception, and Eating Behaviors

A description of the MSEM model building process is presented in the Supplement. Tables 2–5 present the results of the final MSEMs that examined associations among negative and positive affect (predictors), interoception dimensions (mediators), and—in separate models—dietary restriction (Table 2), loss of control eating (Table 3), overeating (Table 4), and IEBs (Table 5; outcomes).

Table 2.

Multilevel Structural Equation Model Examining Associations among Negative and Positive Affect, Interoception, and Dietary Restriction

	Within-Person			Between-Person
	b (SE)	p	β	b (SE)	p	β
X → M
Negative Affect →
Not Distracting	−0.276 (0.098)	.005	−0.116	−1.154 (0.290)	<.001	−0.454
Self-Regulation	−0.071 (0.084)	.394	−0.037	0.025 (0.240)	.917	0.010
Body Listening and Trust	−0.045 (0.079)	.563	−0.023	−0.217 (0.229)	.344	−0.094
Positive Affect →
Not Distracting	0.007 (0.047)	.885	0.006	0.093 (0.184)	.615	0.050
Self-Regulation	0.108 (0.043)	.011	0.109	0.787 (0.172)	<.001	0.442
Body Listening and Trust	0.068 (0.039)	.084	0.067	0.802 (0.166)	<.001	0.473
M → Y
→ Dietary Restriction
Not Distracting	−0.087 (0.047)	.066	−0.077	−0.178 (0.090)	.047	−0.239
Self-Regulation	−0.049 (0.108)	.650	−0.035	0.216 (0.131)	.100	0.276
Body Listening and Trust	0.071 (0.094)	.449	0.052	−0.246 (0.145)	.089	−0.300
X → Y
→ Dietary Restriction
Negative Affect	0.174 (0.113)	.121	0.065	0.271 (0.248)	.275	0.143
Positive Affect	−0.023 (0.056)	.689	−0.016	0.142 (0.174)	.414	0.102
Indirect Effects
Negative Affect → Not Distracting → Dietary Restriction	0.024 (0.015)	.116	0.009
Positive Affect → Not Distracting → Dietary Restriction	−0.001 (0.004)	.885	−0.000
Negative Affect → Self-Regulation → Dietary Restriction	0.004 (0.009)	.687	0.001
Positive Affect → Self-Regulation → Dietary Restriction	−0.005 (0.012)	.655	−0.004
Negative Affect → Body Listening and Trust → Dietary Restriction	−0.003 (0.007)	.648	−0.001
Positive Affect → Body Listening and Trust → Dietary Restriction	0.005 (0.007)	.488	0.003

Note. χ²(296) = 638.968, p < .001; CFI = 0.953; TLI = 0.942; RMSEA = 0.029 (90% CI = 0.026, 0.032); SRMR_within = 0.041; SRMR_between = 0.078; Momentary observations at level 1 were nested within participants at level 2; indictors for the predictors’ (negative affect, positive affect) latent factors were lagged twice, indicators for the mediators’ (interoception) latent factors were lagged once, and indicators for the dependent variable (dietary restriction) were not lagged; latent variable decomposition was used to decrease bias in parameter estimates at level 2 (Lüdtke et al., 2008); this model was estimated using maximum likelihood estimation.

Table 5.

Multilevel Structural Equation Model Examining Associations among Negative and Positive Affect, Interoception, and Intuitive Eating Behaviors

	Within-Person			Between-Person
	b (SE)	p	β	b (SE)	p	β
X → M
Negative Affect →
Not Distracting	−0.325 (0.100)	.001	−0.127	−1.174 (0.326)	<.001	−0.406
Self-Regulation	−0.068 (0.079)	.390	−0.035	0.106 (0.258)	.681	0.038
Body Listening and Trust	−0.082 (0.068)	.226	−0.045	−0.214 (0.244)	.381	−0.082
Positive Affect →
Not Distracting	0.048 (0.051)	.349	0.036	0.059 (0.186)	0.750	0.031
Self-Regulation	0.101 (0.042)	.016	0.100	0.829 (0.175)	<.001	0.454
Body Listening and Trust	0.087 (0.036)	.015	0.090	0.813 (0.167)	<.001	0.471
M → Y
→ EPER
Not Distracting	0.072 (0.030)	.015	0.100	0.140 (0.059)	.017	0.278
Self-Regulation	0.006 (0.071)	.936	0.006	−0.151 (0.088)	.085	−0.286
Body Listening and Trust	0.081 (0.067)	.225	0.081	0.124 (0.096)	.196	0.223
→ UPE
Not Distracting	−0.073 (0.046)	.109	−0.070	−0.058 (0.075)	.435	−0.090
Self-Regulation	−0.194 (0.115)	.091	−0.139	−0.418 (0.112)	<.001	−0.615
Body Listening and Trust	0.172 (0.106)	.104	0.117	0.725 (0.126)	<.001	1.009
→ RHSC
Not Distracting	0.018 (0.025)	.477	0.027	0.147 (0.066)	.026	0.203
Self-Regulation	−0.061 (0.063)	.330	−0.069	−0.349 (0.102)	.001	−0.463
Body Listening and Trust	0.207 (0.059)	<.001	0.223	0.802 (0.116)	<.001	1.005
→ BFCC
Not Distracting	−0.029 (0.040)	.465	−0.027	0.010 (0.064)	.871	0.018
Self-Regulation	−0.091 (0.099)	.357	−0.063	−0.120 (0.098)	.221	−0.204
Body Listening and Trust	0.193 (0.092)	.035	0.127	0.304 (0.107)	.005	0.488
X → Y
→ EPER
Negative Affect	−0.173 (0.070)	.014	−0.094	−0.363 (0.183)	.047	−0.249
Positive Affect	−0.015 (0.037)	.687	−0.015	−0.067 (0.118)	.568	−0.070
→ UPE
Negative Affect	0.076 (0.108)	.486	0.028	0.080 (0.222)	.717	0.043
Positive Affect	0.079 (0.057)	.165	0.056	−0.105 (0.150)	.484	−0.085
→ RHSC
Negative Affect	−0.002 (0.060)	.968	−0.001	0.278 (0.203)	.171	0.133
Positive Affect	0.046 (0.032)	.146	0.051	−0.013 (0.135)	.925	−0.009
→ BFCC
Negative Affect	−0.110 (0.095)	.246	−0.039	−0.201 (0.196)	.306	−0.123
Positive Affect	0.018 (0.050)	.712	0.013	0.208 (0.131)	.113	0.193
Indirect Effects
Negative Affect → Not Distracting → EPER	−0.023 (0.012)	.044	−0.013
Positive Affect → Not Distracting → EPER	0.003 (0.004)	.375	0.004
Negative Affect → Self-Regulation → EPER	−0.000 (0.005)	.936	−0.000
Positive Affect → Self-Regulation → EPER	0.001 (0.007)	.936	0.001
Negative Affect → Body Listening and Trust → EPER	−0.007 (0.008)	.388	−0.004
Positive Affect → Body Listening and Trust → EPER	0.007 (0.007)	.278	0.007
Negative Affect → Not Distracting → UPE	0.024 (0.016)	.143	0.009
Positive Affect → Not Distracting → UPE	−0.004 (0.004)	.413	−0.002
Negative Affect → Self-Regulation → UPE	0.013 (0.017)	.442	0.005
Positive Affect → Self-Regulation → UPE	−0.020 (0.014)	.163	−0.014
Negative Affect → Body Listening and Trust → UPE	−0.014 (0.015)	.332	−0.005
Positive Affect → Body Listening and Trust → UPE	0.015 (0.011)	.174	0.011
Negative Affect → Not Distracting → RHSC	−0.006 (0.008)	.490	−0.003
Positive Affect → Not Distracting → RHSC	0.001 (0.002)	.575	0.001
Negative Affect → Self-Regulation → RHSC	0.004 (0.006)	.519	0.002
Positive Affect → Self-Regulation → RHSC	−0.006 (0.007)	.366	−0.007
Negative Affect → Body Listening and Trust → RHSC	−0.017 (0.015)	.252	−0.010
Positive Affect → Body Listening and Trust → RHSC	0.018 (0.009)	.044	0.020
Negative Affect → Not Distracting → BFCC	0.009 (0.013)	.481	0.003
Positive Affect → Not Distracting → BFCC	−0.001 (0.002)	.571	−0.001
Negative Affect → Self-Regulation → BFCC	0.006 (0.010)	.531	0.002
Positive Affect → Self-Regulation → BFCC	−0.009 (0.011)	.392	−0.006
Negative Affect → Body Listening and Trust → BFCC	−0.016 (0.015)	.293	−0.006
Positive Affect → Body Listening and Trust → BFCC	0.017 (0.011)	.111	0.011

Note. χ²(554) = 1329.173, p < .001; CFI = 0.943; TLI = 0.930; RMSEA = 0.031 (90% CI = 0.029, 0.034); SRMR_within = 0.039; SRMR_between = 0.080; EPER = eating for physical rather than emotional reasons Intuitive eating; UPE = unconditional permission to eat Intuitive eating; RHSC = reliance on hunger and satiety cues intuitive eating; BFCC = body-food choice congruence intuitive eating; momentary observations at level 1 were nested within participants at level 2; indictors for the predictors’ (negative affect, positive affect) latent factors were lagged twice, indicators for the mediators’ (interoception) latent factors were lagged once, and indicators for the dependent variable (intuitive eating behaviors) were not lagged; latent variable decomposition was used to decrease bias in parameter estimates at level 2 (Lüdtke et al., 2008); this model was estimated using maximum likelihood estimation.

Table 3.

Multilevel Structural Equation Model Examining Associations among Negative and Positive Affect, Interoception, and Loss of Control Eating

	Within-Person			Between-Person
	b (SE)	p	β	b (SE)	p	β
X → M
Negative Affect →
Not Distracting	−0.308 (0.101)	.002	−0.127	−1.156 (0.287)	<.001	−0.453
Self-Regulation	−0.059 (0.082)	.467	−0.031	0.026 (0.239)	.915	0.011
Body Listening and Trust	−0.034 (0.077)	.663	−0.017	−0.233 (0.227)	.304	−0.101
Positive Affect →
Not Distracting	0.005 (0.046)	.913	0.004	0.121 (0.185)	.512	0.064
Self-Regulation	0.112 (0.041)	.006	0.116	0.801 (0.175)	<.001	0.441
Body Listening and Trust	0.083 (0.039)	.032	0.080	0.806 (0.168)	<.001	0.470
M → Y
→ Loss of Control Eating
Not Distracting	−0.050 (0.054)	.358	−0.033	−0.109 (0.095)	.253	−0.139
Self-Regulation	−0.175 (0.126)	.167	−0.089	0.029 (0.133)	.829	0.035
Body Listening and Trust	0.109 (0.106)	.304	0.059	−0.071 (0.151)	.637	−0.082
X → Y
→ Loss of Control Eating
Negative Affect	0.192 (0.126)	.129	0.051	0.844 (0.274)	.002	0.424
Positive Affect	0.092 (0.062)	.138	0.049	−0.062 (0.186)	.740	−0.042
Indirect Effects
Negative Affect → Not Distracting → Loss of Control Eating	0.015 (0.018)	.389	0.004
Positive Affect → Not Distracting → Loss of Control Eating	−0.000 (0.002)	.914	−0.000
Negative Affect → Self-Regulation → Loss of Control Eating	0.010 (0.016)	.517	0.003
Positive Affect → Self-Regulation → Loss of Control Eating	−0.020 (0.016)	.220	−0.010
Negative Affect → Body Listening and Trust → Loss of Control Eating	−0.004 (0.009)	.690	−0.001
Positive Affect → Body Listening and Trust → Loss of Control Eating	0.009 (0.010)	.352	0.005

Note. χ²(199) = 519.204, p < .001; CFI = 0.950; TLI = 0.936; RMSEA = 0.033 (90% CI = 0.030, 0.037); SRMR_within = 0.043; SRMR_between = 0.079; Momentary observations at level 1 were nested within participants at level 2; indictors for the predictors’ (negative affect, positive affect) latent factors were lagged twice, indicators for the mediators’ (interoception) latent factors were lagged once, and the dependent variable (loss of control eating) was not lagged; latent variable decomposition was used to decrease bias in parameter estimates at level 2 (Lüdtke et al., 2008); this model was estimated using maximum likelihood estimation.

Table 4.

Multilevel Structural Equation Model Examining Associations among Negative and Positive Affect, Interoception, and Overeating

	Within-Person			Between-Person
	b (SE)	p	β	b (SE)	p	β
X → M
Negative Affect →
Not Distracting	−0.274 (0.095)	.004	−0.115	−1.213 (0.332)	<.001	−0.417
Self-Regulation	−0.040 (0.077)	.606	−0.021	0.096 (0.260)	.713	0.034
Body Listening and Trust	−0.039 (0.074)	.598	−0.019	−0.236 (0.249)	.342	−0.089
Positive Affect →
Not Distracting	0.003 (0.045)	.944	0.003	0.089 (0.184)	.629	0.047
Self-Regulation	0.114 (0.040)	.005	0.118	0.803 (0.174)	<.001	0.442
Body Listening and Trust	0.081 (0.038)	.032	0.079	0.805 (0.168)	<.001	0.466
M → Y
→ Overeating
Not Distracting	−0.012 (0.048)	.808	−0.009	−0.078 (0.074)	.289	−0.130
Self-Regulation	−0.105 (0.111)	.344	−0.061	0.142 (0.110)	.197	0.225
Body Listening and Trust	0.082 (0.093)	.379	0.050	−0.193 (0.121)	.112	−0.290
X → Y
→ Overeating
Negative Affect	0.007 (0.105)	.948	0.002	0.385 (0.230)	.094	0.220
Positive Affect	0.022 (0.054)	.679	0.013	−0.052 (0.149)	.726	−0.046
Indirect Effects
Negative Affect → Not Distracting → Overeating	0.003 (0.013)	.811	0.001
Positive Affect → Not Distracting → Overeating	−0.000 (0.001)	.948	−0.000
Negative Affect → Self-Regulation → Overeating	0.004 (0.009)	.650	0.001
Positive Affect → Self-Regulation → Overeating	−0.012 (0.013)	.369	−0.007
Negative Affect → Body Listening and Trust → Overeating	−0.003 (0.007)	.652	−0.001
Positive Affect → Body Listening and Trust → Overeating	0.007 (0.008)	.416	0.004

Note. χ²(230) = 638.335, p < .001; CFI = 0.945; TLI = 0.930; RMSEA = 0.035 (90% CI = 0.032, 0.038); SRMR_within = 0.045; SRMR_between = 0.080; Momentary observations at level 1 were nested within participants at level 2; indictors for the predictors’ (negative affect, positive affect) latent factors were lagged twice, indicators for the mediators’ (interoception) latent factors were lagged once, and the dependent variable (overeating) was not lagged; latent variable decomposition was used to decrease bias in parameter estimates at level 2 (Lüdtke et al., 2008); this model was estimated using maximum likelihood estimation.

3.2.1. Affect (X)→ Interoception (M)

Across models, there were significant inverse associations between the negative affect and “not distracting” interoception factors at both the within-person and between-person levels, but not between negative affect and the “self-regulation” or “body listening and trust” interoception factors at either level. That is, at times when participants reported greater negative affect than usual, they subsequently reported greater attempts to distract themselves from bodily sensations that felt uncomfortable or painful at their next assessments. Further, women who generally reported greater negative affect than other women also generally exhibited greater attempts to distract themselves from adversely-perceived bodily sensations throughout the study.

Across models, there were significant positive associations between positive affect and the “self-regulation” and “body listening and trust” interoception factors at the within-person and between-person levels, except the within-person association for “body listening and trust” was not significant when dietary restriction was the outcome. In other words, at times when women reported greater positive affect than usual, they subsequently reported greater attempts to turn their attention to their bodily sensations to help them decrease distress (e.g., via deep breathing; greater “self-regulation”) and greater attempts to listen to and trust their bodily sensations (greater “body listening and trust”) at their subsequent assessments. Further, women who generally reported greater positive affect than other women generally reported greater “self-regulation” and “body listening and trust” throughout the study. Associations between positive affect and the “not distracting” interoception factors were not significant at either level in any model.

3.2.2. Interoception (M)→ Disordered Eating Behaviors (Y)

There was a significant inverse association between the “not distracting” interoception and dietary restriction factors at the between-person—but not within-person—level. That is, women who generally reported that they did not attempt to distract themselves from bodily sensations that felt uncomfortable or painful reported less dietary restriction than women who reported more distracting from bodily sensations throughout the study. No interoception factors were associated with loss of control eating or overeating at the within-person or between-person levels.

3.2.3. Interoception (M)→ IEBs (Y)

At the within-person level, greater “body listening and trust” interoception was associated with greater “reliance on hunger and satiety cues” and—separately—“body-food choice congruence” IEBs, but not “eating for physical versus emotional reasons” or “unconditional permission to eat” IEBs. That is, at times when women reported greater attempts to listen to and trust their bodily sensations than usual, they subsequently reported greater reliance on their hunger and satiety cues to guide their eating behaviors and that they chose nutrient-dense foods that could support their bodies’ needs at their next assessments. Also, there was a significant positive within-person association between “not distracting” interoception and the “eating for physical versus emotional reasons” IEB. That is, at times when women reported fewer attempts to distract themselves from bodily sensations that felt uncomfortable or painful, they subsequently reported greater eating for physical versus emotional reasons at their next assessments. No interoception dimensions were associated with “unconditional permission to eat” IEB at the within-person level.

There were multiple significant direct effects between the interoception dimensions and IEBs at the between-person level, with variations for different IEBs. For example, as shown in Table 5, all three interoception dimensions were associated with the “reliance on hunger and satiety cues” IEB at the between-person level. Specifically, women who generally reported fewer attempts to distract themselves from internal bodily sensations that felt uncomfortable or painful (more “not distracting” interoception), fewer attempts to turn their attention to their bodily sensations to help them decrease feelings of distress (less “self-regulation” interoception), and greater attempts to listen to and trust their bodily sensations (greater “body listening and trust” interoception) than other women also generally reported greater “reliance on their hunger and satiety cues” IEB to guide their eating behaviors throughout the study period. Less “self-regulation” interoception and greater “body listening and trust” interoception dimensions were also associated with greater “unconditional permission to eat” IEB at the between-person level, as were between-person associations between greater “not distracting” interoception and greater “eating for physical rather than emotional reasons” IEB, and between greater “body listening and trust” interoception and greater “body-food choice congruence” IEB.

3.2.4. Indirect Effects

Whereas none of the indirect effects linking negative and positive affect to the interoception factors and, in turn, DEBs were significant, two significant indirect effects emerged when IEBs were modeled as outcomes. Specifically, negative affect was associated with the “eating for physical versus emotional reasons” IEB via “not distracting” interoception, and positive affect was associated with the “reliance on hunger and satiety cues” IEB via “body listening and trust” interoception. That is, at times when women reported greater negative affect than usual, they subsequently reported greater attempts to distract themselves from bodily sensations that felt uncomfortable or painful at their next assessments (i.e., less “not distracting” interoception) and, in turn, less “eating for physical versus emotional reasons” at their subsequent assessments. Also, at times when women reported greater positive affect than usual, they subsequently reported greater attempts to listen to and trust their bodily sensations (more “body listening and trust” interoception) at their next assessments and, in turn, greater reliance on their hunger and satiety cues to guide their eating behaviors at their next assessments. No other indirect effects were significant.

4.0. Discussion

The present study used EMA to examine whether distinct interoception dimensions differentially mediated theoretically supported (Hawkins & Clement, 1984; Linardon et al., 2021; Mason et al., 2021) associations between negative and positive affect relative to women with ED pathology’s engagement in distinct DEBs and IEBs in the natural environment. Results identified circumscribed momentary mediational associations among negative and positive affect, interoception dimensions, and IEBs. Notable variations in this pattern of results emerged based on the type of affect, interoception, and IEB, as discussed below. Also, no momentary indirect effects linking affect, interoception, and DEBs were significant. Collectively, these findings provide insight that may be leveraged to enhance the EDs evidence-base and theories, and help increase the precision and efficacy of novel real-time EDs interventions.

4.1. Intuitive Eating Behaviors

In the present study, it is noteworthy that momentary interoception dimensions (“not distracting,” “body listening and trust”) differentially mediated momentary associations between negative and positive affect relative to IEBs involving high levels of somatic experiencing/connection (“eating for physical versus emotional reasons,” “reliance on hunger and satiety cues”), but none of the momentary mediational associations with DEBs as outcomes. To our knowledge, the present study was the first to examine these associations at the momentary level using EMA, which limits comparisons with prior research. However, these findings are consistent with and expand upon proxy evidence from between-person (cross-sectional) research with non-clinical samples that has shown that people who generally exhibit better subjective interoception dimensions than others also generally engage in more IEBs (bivariate associations; Jeune et al., 2024; Oswald et al., 2017; Robinson et al., 2021).

The present findings have several implications that warrant attention. First, it is noteworthy that at times when women reported greater negative affect than usual, they subsequently reported greater attempts to distract themselves from bodily sensations that felt uncomfortable or painful at their next assessments (i.e., less “not distracting” interoception) and, in turn, less “eating for physical versus emotional reasons” IEB later that day. This finding aligns with conceptual models of aberrant interoceptive processing, wherein individuals are believed to engage in more maladaptive and less adaptive behaviors as attempts to experientially distance themselves, or “escape,” from adverse bodily sensations (e.g., physiological manifestations of negative affect; Herbert, 2021; Khalsa et al., 2018, 2022). Further, the present finding extends this prior research by specifying the temporality of these dynamic patterns of association in the natural environment (e.g., while women are at home, work/school). Specifically, these data indicate that engaging in such experiential distancing at a given moment can have enduring impacts later that same day and could serve as an explanatory factor that underlies difficulties women may have with limiting emotion-focused eating behaviors and conversely basing their (intuitive) eating behaviors on physiological sensations (e.g., hunger, satiety). In future research, it may be helpful to further explore the temporality of these associations by assessing how long a sequential worsening of affect and experiential distancing on a given day continues to map onto less engagement in eating behaviors that are driven by physiological sensations versus emotional experiences (e.g., “does this extend to the next day, week, etc.?”).

Second, results showed that at times when women reported greater positive affect than usual, they subsequently reported greater attempts to listen to and trust their bodily sensations (more “body listening and trust” interoception) at their next assessments and, in turn, greater reliance on their hunger and satiety cues to guide their eating behaviors later that day. This finding complements the identified momentary indirect effect linking greater negative affect, less “not distracting” interoception, and less “eating for physical versus emotional reasons” IEB by demonstrating how theoretically supported transdiagnostic self-regulatory processes (e.g., affect, interoception) can promote a core health-promoting eating behavior (“reliance on hunger and satiety cues” IEB; Linardon et al., 2021; Tylka & Kroon Van Diest, 2013) in the natural environment from a strengths-based positive psychological perspective. This is important, as existing EDs research has primarily focused on examining correlates of eating behaviors from a deficit-based perspective, such as by examining factors that are associated with DEB severity, relapse, and chronicity (Sala et al., 2023; van Hoeken & Hoek, 2020). Yet, in line with positive psychological theory (Ryff, 2014), the absence of pathology (e.g., no current DEBs) does not inherently indicate that individuals are “well,” functioning adaptively, or engaging in behaviors like the “reliance on hunger and satiety cues” IEB that can promote beneficial health long-term (Linardon et al., 2021; Tylka & Kroon Van Diest, 2013). Given the limited EDs research using this framework, the present findings support conducting future research that also applies a strengths-based positive psychological perspective, such as by continuing to explore factors that are associated with health-promoting IEBs among EDs samples. Findings from studies with a positive psychological framework can broaden the understanding of how individuals with ED pathology may not only decrease their maladaptive symptoms (e.g., DEBs) but improve their health and wellness.

4.2. Disordered Eating Behaviors

It is unclear why sequential changes in affect and interoception that emerged over the course of day did not predict DEBs (dietary restriction, overeating, loss of control eating) later that day, given the lack of within-person research in this area for comparison. These non-significant findings contrast established cross-sectional between-person research that has identified robust associations between worse affect and interoception, relative to greater DEBs at the trait-level (Herbert et al., 2021; Jenkinson et al., 2018; Olatunji et al., 2018; Pollatos & Georgiou, 2016), and transdiagnostic EDs theories that posit that affect dysregulation and aberrant interoceptive processing can promote the onset and maintenance of EDs (Herbert, 2021; Hawkins & Clement, 1984; Khalsa et al., 2022; Mason et al., 2021). The present findings consequently highlight the importance of examining whether such trait-level associations observed in between-person research manifest at the within-person (momentary) level. Indeed, as shown by the present non-significant findings with DEB outcomes, trait-level processes (e.g., how biobehavioral processes manifest and covary for individuals in general and characterize their dispositional functioning) do not necessarily provide clarity about within-person (momentary) associations (e.g., how biobehavioral processes dynamically change across more discrete micro-temporal timescales, such as from moment-to-moment or day-to-day). Notably, these null findings may vary with a clinical sample with higher DEB severity and should be assessed in future research.

4.3. Strengths and Limitations

The present study has multiple strengths. First, this study was the first to use EMA with a sample of women with ED pathology who exhibited elevated diversity related to race and sexual orientation to examine how moment-to-moment changes in different interoception dimensions mediate momentary associations between negative and positive affect, relative to women’s engagement in both DEBs and IEBs in the natural environment. This permitted the examination of when, why, and how transdiagnostic self-regulatory processes (affect, interoception) that are of central importance to understanding ED pathology (Haedt-Matt & Keel, 2011; Herbert, 2021; Khalsa et al., 2022) can differentially precipitate women with ED pathology’s engagement in both maladaptive and adaptive eating behaviors. The present findings have important implications for theory building and intervention development. Second, applying a positive psychology lens by examining factors that can promote health-promoting IEBs in the natural environment, rather than DEBs only, aligns with growing research showing that the absence of maladaptive symptoms (e.g., DEBs) does not inherently equate to the presence of health, wellness, and adaptive functioning, and that a lack of positive functioning is associated with ED relapse (Ryff, 2014; Tomba et al., 2019). These findings with IEBs outcomes consequently provide insight that may be leveraged to inform new real-time mHealth interventions that may not only decrease women’s DEBs but improve their health and wellness (Linardon et al., 2021; Tylka & Kroon Van Diest, 2013).

Despite these strengths, study limitations warrant attention. First, the present study included college women with elevated ED pathology. It is unknown whether these results generalize to youth, individuals who do not identify as women, individuals with other marginalized identities, or non-clinical or non-college samples. Future research would benefit from examining whether the assessed associations are replicated within these other important groups to gauge their generalizability. Second, the EMA items were either adapted from well-established measures designed for between-person research or selected from longer state-based measures designed to assess within-person processes to limit participant burden and compensate for the brevity of EMA (Shiffman et al., 2008). Although validity-related concerns due to measure adaptation are tempered by the confirmation of measures’ factor structures via MCFA, future research that includes more survey items to assess the DEBs and research that examines the tested statistical models in relation to DEBs that were not assessed in the present study (e.g., compensatory behaviors) will serve as a helpful extension of the present study. Third, participants received four brief semi-randomly prompted surveys each day of the 10-day EMA period and these prompts were sent approximately every 4 hours throughout each study day. This prompting schedule was chosen in an effort to best capture constructs of interest as close in time to when they generally occur (e.g., affect, interoception, eating behaviors) while also limiting participant burden (Shiffman et al., 2008). However, it is unclear whether the sequencing of participants’ affective experiences earlier in the day mapping onto participants’ interoception approximately four hours later and, in turn, DEBs and IEBs four hours after that may differ over longer or shorter timescales, given the novelty of the assessed research questions. Future research that directly assesses how these processes may manifest over varied timescales will serve as an important extension of the present study. Fourth, due to the data structure’s complexity and convergence difficulties that emerged when random slopes were added, fixed slopes and random intercepts were modeled at levels 1 and 2 of the assessed statistical models. Also, although additional autoregressive effects were tested (outcomes at t-1 and t-2, mediators at t-2), convergence difficulties prevented their inclusion. Random slopes and autoregressive effects attest to the extent to which the slopes of regression lines vary across different individuals and determine how strongly prior values (e.g., values at t-1 and t-2) influence a current value (e.g., values at t = 3 or 4), respectively. Future research is needed to assess whether the present findings differ in samples with data that also allow random slopes and autoregressive effects to be added. Fifth, based on the descriptive data shown in Table 1, it is possible that there was a floor effect for the DEBs, wherein a proportion of participants generally endorsed values on the DEBs items that indicate that they did not frequently engage in these behaviors, and that such ratings were less apparent for the assessed IEBs. Future research that includes samples with greater DEBs severity is needed to determine whether the present results differ for individuals with greater eating disorder severity.

4.4. Future Directions

The present findings have important implications for future research and practice. First, that the transdiagnostic self-regulatory factors (affect, interoception) were associated with women’s IEBs in the present study is in line with theory supporting these factors as central processes that can precipitate (mal)adaptive eating behaviors at the trait-level (Herbert, 2021; Khalsa et al., 2018, 2022). However, no existing conceptual models exist that: (1) account for the role of both affect and interoception as precipitating or explanatory factors that can promote the onset and maintenance of IEBs in the natural environment; or (2) explicate how such a framework might be integrated with existing ED models. Instead, IEBs are generally conceptualized via the Acceptance Model of Intuitive Eating (Avalos & Tylka, 2006). This model was developed based on primarily cross-sectional research and focuses on trait-level associations between (general) acceptance from others, body image constructs (e.g., body-related acceptance from others, body appreciation), and IEBs (Avalos & Tylka, 2006; Linardon et al., 2021). The present findings suggest that it may be helpful for future research to develop and further explore the validity of a complementary real-time IEB model that accounts for the role of affect and interoception as constructs that may explain individuals’ engagement in IEBs at the within-person level. Based on the present findings, this conceptualization may benefit from emphasizing the importance of enhancing individuals’ moment-to-moment attunement with their bodily sensations (affect, hunger, satiety) to promote engagement in health-promoting IEBs, and highlight how distancing oneself from these sensations can promote less IEBs. Given that the present study was the first to assess momentary associations among different affect, interoception, and IEB types via EMA, future research that aims to determine whether the present findings are replicated and examines whether such a framework may be a viable conceptual model that explains when, why, and how individuals engage in IEBs in the natural environment is needed to strengthen the theoretical underpinnings of research in this area.

Second, future research may benefit from: (1) developing real-time mHealth interoceptive exposure interventions that encourage individuals with ED symptoms to connect with their bodily sensations; (2) assessing how such a real-time mHealth intervention impacts both DEBs and IEBs in the natural environment. It is well-established that individuals with ED pathology experience difficulties with sensing and connecting to their bodily sensations (e.g., hunger, fullness, bloating; Herbert, 2021; Khalsa et al., 2022). These difficulties are believed to develop through a process of experiential avoidance that occurs when individuals with ED pathology repeatedly attempt to escape or distance themselves from these sensations by engaging in DEBs (Brown et al., 2017; Herbert, 2021; Khalsa et al., 2022). Consistent with EDs theories (Haedt-Matt & Keel, 2011; Hawkins & Clement, 1984; Mason et al., 2021), reductions in negative affect immediately following DEBs are believed to produce a conditioned response and negatively reinforced cycle that persists and contributes to ED maintenance. Using a novel mHealth interoceptive exposure intervention in future research can assess whether repeatedly exposing individuals with ED pathology to their adversely perceived bodily sensations extinguishes this conditioned responding, improves perceptual sensing (e.g., the ability to accurately sense hunger and satiety cues at a given moment), and maps onto improvements in DEBs and IEBs.

4.5. Conclusions

In the present EMA study, circumscribed momentary mediational associations among negative and positive affect, interoception dimensions, and IEBs were identified. However, notable variations in this pattern of results emerged across different affect, interoception, and IEB types. Also, no momentary indirect effects linking affect, interoception, and DEBs were significant. Collectively, these findings provide important insight that may be leveraged to enhance the EDs evidence-base and theories, and increase the precision and efficacy of novel real-time EDs interventions.

Data Availability:

The data that support the findings of this study are available from the corresponding author upon reasonable request.