Working out measurement overlap in the assessment of maladaptive exercise

Abstract

Objective.

Although exercise is generally considered healthy, many individuals engage in maladaptive exercise (e.g., compulsive in nature). Several definitions of maladaptive exercise exist, leading to multiple, varied assessment tools; assuming homogeneity across these assessments contributes to low consensus in etiological models.

Method.

We used a Jaccard Index to quantify content overlap among 15 commonly-used self-report instruments measuring maladaptive exercise, with 31 features identified across 224 items.

Results.

The most common features were exercise to control weight/shape and to avoid negative affect (both included in 9/15 instruments), or compensate for calories consumed (8/15 instruments). Overlap among instruments was low (0.206) and no features were common across all instruments.

Conclusions.

Findings generally support theoretical models of exercise in eating pathology. However, instruments most commonly used to assess maladaptive exercise measure heterogenous content. Careful consideration should be taken when comparing findings derived from differing instruments, when synthesizing literature on maladaptive exercise, and when selecting instruments to measure specific maladaptive exercise features.

Introduction

Exercise is generally associated with positive physical and mental health outcomes; however, exercise can become maladaptive in nature. Exercise that is characterized as addictive (Cook et al., 2014; Miller & Hormes, 2022) or compulsive (Dittmer, Jacobi, et al., 2018; Meyer et al., 2011) is widely reported across community samples (Birche et al., 2017; Scharmer et al., 2020; Taranis & Meyer, 2011), and is a transdiagnostic behavior in the context of psychopathology (Miller & Hormes, 2022), including among those diagnosed with eating disorders (EDs) (Dittmer, Jacobi, et al., 2018; Meyer et al., 2016). Particularly when coupled with eating pathology, maladaptive exercise is associated with elevated risk for suicide (Smith et al., 2013), poorer clinical outcome (Dalle Grave et al., 2008; Erskine et al., 2016), and lower quality-of-life (Cook et al., 2014).

Given the negative sequelae associated with maladaptive exercise across samples, significant research effort has been devoted to understanding and measuring this behavior. Many possible definitions have been proposed and debated (e.g., actual behavior, motivation for weight/shape control, cognitive/affective response or expectancies, compensatory nature, tolerance and withdrawal, increasing positive affect, achievement/social incentives) (Dittmer, Jacobi, et al., 2018; Mond & Gorrell, 2021; Scharmer et al., 2020), and as a result, researchers and clinicians might now select one of many available scales to assess maladaptive exercise. Given the wide number of measures currently used, disparate item content could lead to substantially different conclusions between studies. In addition, it may also reflect changes in researchers’ and clinicians’ understanding of maladaptive exercises across time. For example, within the field of EDs, exercise behavior was traditionally studied for its ‘excessive’ frequency and duration (Davis & Fox, 1993), whereas more recent theoretical conceptualizations have highlighted the importance of also assessing motivation for exercise (Adkins & Keel, 2005; Mond & Gorrell, 2021) and short-term reward consequences (Kolar et al., 2023; Kolar & Gorrell, 2021). Thus, studies conducted using only definitions of frequency and duration to categorize exercise as maladaptive may: (1) not capture those who exercise at lower frequencies and durations who nonetheless would report a maladaptive psychological relationship with exercise, or (2) given differences in exercise engagement across social groups (e.g., athlete populations), may over-pathologize exercise of those who engage in regular exercise that is not compulsive or life-interfering (i.e., sport engagement). Thus, various assessments of maladaptive exercise may either emphasize or miss multiple maladaptive exercise features, resulting in disparate conclusions about the nature of maladaptive exercise.

Taken together, there are several challenges within the current system of measuring maladaptive exercise that point to questionable assessment practices (Flake & Fried, 2019; Harris et al., 2020). Perhaps most notably, it is often assumed that scales of maladaptive exercise across several different theoretical models such as the affect-regulation (Kolar & Gorrell, 2021), or compulsive exercise (Meyer et al., 2011) are all measuring the same construct and therefore, can be used interchangeably. Given the proliferation of both definitions and assessments of maladaptive exercise, it is not surprising that not all measures assess all features of the behavior (e.g., some focus more on motivation and do not include frequency or duration) which may lead to ‘knowledge gaps’ where features are assumed to be absent or present, but are actually not assessed. Assessment heterogeneity can also limit replication and generalizability of research when variables of interest are measured using scales that are assessing different, but related, underlying constructs (Flake & Fried, 2019). If scales are assumed to be homogeneous when they are not, their use creates a fragmented literature from which it is difficult to draw consensus around etiological models. On the other hand, if measures with significant overlap are assumed to capture independent constructs and their respective literatures do not overlap, opportunities for theoretical integration of underlying processes may be missed. Ultimately, lack of coherence or difficulties in integration across assessments could hinder the development of unified theoretical models from which to build interventions.

For these reasons, there is a need to empirically evaluate content overlap in how we are measuring maladaptive exercise. The psychometric properties of some measures have been studied in relation to sport engagement (Di Lodovico et al., 2019), for their predicative ability for eating pathology (Adkins & Keel, 2005; Scharmer et al., 2020), and in relation to one another in their initial validation across samples (Goodwin et al., 2011). However, no study has evaluated and quantified the content overlap of self-report instruments that purportedly measure exercise in the context of EDs. Modeled after similar studies in depression (Fried, 2017), mania (Chrobak et al., 2018), obsessive compulsive disorder (Visontay et al., 2019), and anxiety (Wall & Lee, 2021), we aimed to provide an item content analysis of self-report measures of maladaptive exercise in EDs. Based on prior findings across other diagnostic groups (Fried, 2017; Wall & Lee, 2021), we hypothesized that item overlap would be relatively modest.

Methods

Identification of instruments

We evaluated 11 full scales assessing maladaptive exercise in populations with eating pathology as well as commonly sourced exercise items from four scales assessing general eating pathology (see Table 1); sources were selected from those used in ED samples, and included in recent reviews and chapters (Lampe & Gorrell, 2023; Rizk et al., 2020). Instruments were included based on their frequency in the literature, appearance in studies comparing multiple scales, and citation count based on a consensus among the seven authors. We did not include measures assessing exercise for muscularity, as these measures were not as commonly used in ED samples.

Table 1.

Instruments included and number and percent of specific features captured per instrument

Scale Type	Scale Name	Author (year)	Total citations	Cronbach’s alpha	N Scale Items Included	Features Captured N (%)
Full scales	EMI-2	Markland (1997)	705	0.69–0.95	51	14 (45.16)
	EDS	Hausenblas (2002)	658	--	20	8 (29.03)
	REI	Cash (1994)	388	0.91	20	14 (45.16)
	EPSI	Forbush (2013)	333	0.95	5	4 (12.90)
	CES	Davis (1993)	329	0.77	8	6 (19.35)
	EAI	Griffiths (2005)	286	0.84	6	7 (22.58)
	OEQ	Steffen (1999)	113	0.84	20	14 (45.16)
	CET	Goodwin (2011)	80	0.88	24	12 (38.71)
	EBQ	Luomidis (1998)	80	067–0.87	21	8 (29.03)
	EOQ	Yates (1999)	62	0.92	27	14 (45.16)
	EED	Danielsen (2015)	32	0.90	11	9 (29.03)
Sourced items	EDE-Q	Fairburn (2008)	1322	0.70–0.93	1	3 (9.68)
	EDDS	Stice (2000)	1044	0.91	1	2 (6.45)
	EDI-SC	Wear (1987)	259	0.95	3	3 (9.68)
	YBC-EDS	Bellace (2012)	32	0.76	6	5 (16.13)

Note. “Total citations” refers to the number of google scholar citations for the validation study of each measure as of publication of this article; EDDS: Eating Disorder Diagnostic Scale (Stice et al., 2000); EDE-Q: Eating Disorder Examination Questionnaire (Fairburn & Beglin, 2008); EDI-SC: Eating Disorders Inventory Symptom Checklist (Wear & Pratz, 1987); EPSI: Eating Pathology Symptoms Inventory (Forbush et al., 2013); YBC-EDS: Yale-Brown-Cornell Eating Disorder Scale (Bellace et al., 2012); CES: Commitment to Exercise Scale (Davis et al., 1993); EAI: Exercise Addiction Inventory (Griffiths et al., 2005); EBQ: Exercise Beliefs Questionnaire (Loumidis & Wells, 1998); EDS: Exercise Dependence Scale (Hausenblas & Downs, 2002b); EED: Exercise and Eating Disorder questionnaire (Danielsen et al., 2015); CET: Compulsive Exercise Test (Goodwin et al., 2011); EMI-2: Exercise Motivations Inventory, 2^nd edition (Markland & Ingledew, 1997); EOQ: Exercise Orientation Questionnaire (Yates et al., 1999); OEQ: Obligatory Exercise Questionnaire (Steffen & Brehm, 1999); REI: Reasons for Exercise Inventory (Cash et al., 1994).

Content analysis

All scales and sourced items together comprise 224 items. A content analysis was used to determine constructs assessed among items. As it is impossible to objectively determine which features were assessed across all items due to high heterogeneity in the literature and because there is no way to determine clearly whether two similarly worded items are meant to measure the same features, a group consensus approach was used. First, EL and SG reviewed all items separately to identify features assessed; any disagreement in feature categorization was resolved through discussion (EL/SG). EL and SG then prepared a codebook that defined each of the identified features (see Table 2). All seven authors reviewed the codebook and set, by group consensus, final category definitions. This process identified 31 distinct features of maladaptive exercise assessed across all 224 items.

Table 2.

Maladaptive exercise features and their definitions.

Feature	Definition	N Instruments Featured (out of 15)
Avoid negative affect	Engagement in exercise to avoid anticipated negative emotions (e.g., guilt; anxiety; worry) if one did not engage in the exercise.	9
Weight/shape control	Engagement in exercise with the aim of altering one’s weight/shape or protecting against anticipated weight gain/shape changes if one did not exercise.	9
Compensate for calories	Engagement in exercise to compensate for, or “work-off” calories consumed.	8
Rigid/Routine	Engagement in exercise in accordance with a strict schedule/routine.	7
Decrease negative affect	Engagement in exercise with the aim of reducing negative emotions/distress.	6
Appearance	Engagement in exercise to alter or maintain the way one looks to others. Not specific to weight/shape.	6
Increase positive affect	Engagement in exercise with the aim of increasing positive emotional experiences (e.g., via endorphins or self-efficacy/confidence)	5
Obligation	Engagement in exercise because one feels one must or should (e.g., to adhere to a regimen).	5
Frequency	The frequency with which one engages in exercise over a given amount of time.	5
Physical consequences	Engagement in exercise despite experiencing or risking negative physical ramifications (e.g., injury; illness; overtraining).	5
Psychosocial consequences	Engagement in exercise despite experiencing or risking negative psychosocial ramifications (e.g., negatively impacting relationships; lack of socialization)	5
Athletic achievement/goal attainment	Engagement in exercise motivated by a sense of accomplishment via working toward or meeting personal/athletic benchmarks.	4
Decrease/avoid physical discomfort	Engagement in exercise with the goal of reducing or avoiding physical discomfort (e.g., bloating; feelings of fatness) which would otherwise be experienced.	4
Health improvement/ maintenance	Engagement in exercise with the aim of improving or maintaining one’s physical/mental health.	4
Building physical ability/skill/function	Exercise engagement with the aim of maintaining or advancing one’s physical abilities, skills, or overall functional capabilities.	4
Social engagement	Engagement in exercise as a way to spend time/connect with other people.	4
Perseveration/Rumination	Having thoughts/preoccupation/daydreams about exercise when one is not actively engaged in exercising.	3
Restlessness	Feeling “fidgety” or like one needs to be moving at all times.	3
Cognitive enhancement	Engagement in exercise with the aim of improving one’s cognitive ability or focus.	3
Enjoyment	The enjoyment one gets from exercise engagement.	3
Social desirability	Engagement in exercise with the aim of being considered valuable or interesting to others.	3
Pre-compensation	Engagement in exercise to compensate for calories/food one anticipates consuming at a later time.	2
Compulsion	In response to insistent urges, a subjective experience of exercise as activity that is “driven,” “out of control,” or compelled.	2
Inability to reduce/cut-out	Lack of ability to reduce or eliminate exercise engagement despite attempts to do so.	2
Tolerance	Needing increased exercise engagement (e.g., longer duration, higher intensity) to achieve the same effect (e.g., affective state; weight control) over time.	2
Build muscle/strength	Engagement in exercise with the aim of increasing one’s muscle mass or strength.	2
Duration	The duration for which one engages in exercise.	2
Exercise intensity	The strenuousness of exercise (e.g., pushing oneself; breathlessness).	2
Importance	Viewing exercise as a top priority or as a reflection of engagement with personal values.	2
Activity type	The type of activity one engages in for exercise (e.g., cardiovascular, stretching, strength training, or specific sports).	1
Lack of Enjoyment	The extent to which one actively does not enjoy exercise engagement.	1

In a second step, six authors (KS, DK, KC, MC, DC, and SG) independently rated which features were assessed in each of the 224 items; EL collated results. Items were considered to assess a specific feature if ≥ 4/6 raters identified that feature for that item; items could be considered within more than one category. Using this rule, authors reached consensus for 201 items (89.7%). All authors reviewed the remaining 23 items and determined feature categorization through group discussion and consensus.

Statistical Analysis

For full scales included in the analysis (excluding sourced items), we estimated content overlap using the Jaccard Index, which is a commonly used similarity coefficient for binary data (Evans, 1996). Jaccard Index scores range from 0 (no overlap among scales) to 1 (complete overlap) and is calculated by dividing the number of features two measures share (s) by the number of features unique to each of the two scales (u1 and u2) plus the number of shared features (i.e., s / (u1 + u2 + s)). We used guidelines for strength of correlation coefficients from Evans (1996), following prior literature (Fried, 2017; Wall & Lee, 2022): very weak 0.00–0.19, weak 0.20–0.39, moderate 0.40–0.59, strong 0.60–.79, and very strong 0.80–1.0. We also calculated the rate of unique features per scale (i.e., features that appear in no other scale). Analyses were conducted using R version 4.1.3 (R Core Team, 2022); data and code are available in Open Science Foundation (OSF) at osf.io/q65v4.

Results

Features appear in a mean of 3.97 instruments (including full scales and sourced items; mode = 2, median = 4). Of the 31 features, two (6.45%; lack of enjoyment; activity type) appear only in one single instrument, whereas no features were assessed across all instruments. The three most common specific features were: (i) exercise for weight/shape control and exercise to avoid negative affect (both featured in nine scales), (ii) exercise to compensate for calories consumed (featured in eight scales), and (iii) rigidity/routine around exercise (featured in seven scales). Table 2 lists how many instruments each of the features are included. Table 1 summarizes the number of specific features captured per instrument.

Scale overlap

Overlap among full questionnaires (excluding commonly sourced items from general ED instruments) was estimated via the Jaccard Index. The mean overlap among all scales is 0.229, which implies a ‘weak’ similarity of the scales (Evans, 1996); specific overlap among all individual scales, and mean overlap of each scale with all other 11 full scales, are presented in Table 3. Several notable findings emerged:

1. The Exercise Dependence Scale (EDS) and the Exercise Motivations Inventory (EMI-2) demonstrated no overlap in feature assessment. The EMI-2 captures a broad range of motivations for exercise (within six domains, including ill-health avoidance and weight management) but notably negates psychological dependence, a key feature of dependence-based theoretical models noted above, and the construct that the EDS was designed to characterize.

2. Very few full scales demonstrated moderate to strong levels of overlap. The highest overlap was between the EMI-2 and Reasons for Exercise Inventory (REI) (0.750); this is not unexpected, given that both scales are intended to assess a broad range of exercise motivations. The next highest overlap was between the Compulsive Exercise Test (CET) and the Exercise and Eating Disorder Questionnaire (EED) (0.500). The third-highest overlaps were between the Exercise Beliefs Questionnaire (EBQ) and both the EMI-2 and REI (0.467), characterizing beliefs about or motivations for exercise, respectively.

3. The scale with the highest degree of overlap with all 11 other full scales was the Exercise Orientation Questionnaire (EOQ) (0.282), followed by the REI (0.274), and the CET (0.267). All represent what would be considered weak similarity. The scale with the lowest overlap with all 11 other full scales was the EDS (0.139), followed by the Exercise Addiction Inventory (EAI) (0.164), and the Commitment to Exercise Scale (CES) (0.194). Each of these scales reflect quite different domains, namely exercise dependence, addiction, and commitment, respectively.

Table 3.

Overlap of item content of 10 maladaptive exercise scales: Jaccard Index.

	CES	CET	EAI	EBQ	EDS	EED	EMI-2	EOQ	OEQ	REI	EPSI
CES	--	0.286	0.182	0.167	0.273	0.364	0.053	0.176	0.333	0.111	0.250
CET	0.286	--	0.189	0.250	0.111	0.500	0.300	0.300	0.368	0.368	0.143
EAI	0.182	0.189	--	0.154	0.364	0.067	0.105	0.313	0.167	0.105	0.000
EBQ	0.167	0.250	0.154	--	0.067	0.214	0.467	0.375	0.100	0.467	0.000
EDS	0.273	0.111	0.364	0.067	--	0.133	0.000	0.100	0.294	0.048	0.091
EED	0.364	0.500	0.067	0.214	0.133	--	0.211	0.353	0.353	0.278	0.182
EMI-2	0.053	0.300	0.105	0.467	0.000	0.211	--	0.474	0.217	0.750	0.000
EOQ	0.176	0.300	0.313	0.375	0.100	0.353	0.474	--	0.333	0.400	0.059
OEQ	0.333	0.368	0.167	0.100	0.294	0.353	0.217	0.333	--	0.217	0.286
REI	0.111	0.368	0.105	0.467	0.048	0.278	0.750	0.400	0.217	--	0.000
EPSI	0.250	0.143	0.000	0.000	0.091	0.182	0.000	0.059	0.286	0.000	--
Averages	0.199	0.256	0.149	0.205	0.135	0.241	0.234	0.262	0.243	0.249	0.092

Note. EDDS: Eating Disorder Diagnostic Scale; EDE-Q: Eating Disorder Examination Questionnaire; EDI-SC: Eating Disorders Inventory Symptom Checklist; EPSI: Eating Pathology Symptoms Inventory; YBC-EDS: Yale-Brown-Cornell Eating Disorder Scale; CES: Commitment to Exercise Scale; EAI: Exercise Addiction Inventory; EBQ: Exercise Beliefs Questionnaire; EDS: Exercise Dependence Scale; EED: Exercise and Eating Disorder questionnaire; CET: Compulsive Exercise Test; EMI: Exercise Motivations Inventory; EOQ: Exercise Orientation Questionnaire; OEQ: Obligatory Exercise Questionnaire; REI: Reasons for Exercise Inventory.

Discussion

Results generally suggest that various maladaptive exercise scales cannot be used interchangeably across research and practice settings, and integration and replicability of studies using the variety of measurements of maladaptive exercise that are available may be difficult, if not impossible. Given the prevalence rates of maladaptive exercise across athlete (Gorrell & Anderson, 2018), college (Scharmer et al., 2020), community (Birche et al., 2017), and clinical (Dittmer, Voderholzer, et al., 2018; Kolar et al., 2022) samples, careful consideration of how we measure this behavior is of great importance moving forward.

Key findings

Consistent with prior work in the assessment of other mental health symptoms such as depression (Fried, 2017) or anxiety (Wall & Lee, 2021), our first key finding is the overall lack of content overlap among maladaptive exercise scales given the large number of features assessed across scales. Moreover, there was no particular scale that appeared to overlap with a greater number of others such that no one scale stands out as an optimal representative of this body of assessment more broadly. Another important finding is that there was higher overlap at the item-level across all instruments for several specific features of exercise (i.e., exercise to avoid negative affect, for weight/shape control, to compensate for calories consumed, and reporting rigidity/routine around exercise; Figure 1) which are most closely aligned with ED symptoms and the most widely accepted theoretical models (Dittmer, Jacobi, et al., 2018; Kolar & Gorrell, 2021; Meyer et al., 2011; Mond & Gorrell, 2021). Despite this item-level evidence, several full instruments which center on shared theoretical models do not have high overlap, reflecting potentially important differences among each of the instruments despite their theoretical origins. For example, overlap between a scale that assesses exercise dependence (EDS) versus exercise addiction (EAI) was 0.364; similarly, the overlap between a scale that assesses exercise compulsion (CET) versus exercise obligation (OEQ) was 0.368. These findings may reflect shifting theoretical conceptualizations of exercise in the context of EDs from the late 1990’s through the early 2000’s, with more recent theoretical models overlapping only somewhat with pre-existing theoretical models of exercise.

Figure 1.

Co-occurrence of 31 maladaptive exercise features across 15 instruments.

Heterogeneity in exercise features

To what causes do we attribute this lack of overlap? One source of heterogeneity derives from the considerable number of theoretical models that are currently used to conceptualize maladaptive exercise behavior in the context of EDs (Mond & Gorrell, 2021). There is currently a lack of consensus in the field around any one theoretical model, or definition of maladaptive exercise more broadly; some work has been moving the field towards agreement (Noetel et al., 2017), efforts that should be continued. In the interim, the four conceptual motivations for maladaptive exercise (of many proposed in the literature) that we highlighted just above (exercise to avoid negative affect, for weight/shape control, to compensate for calories consumed, and reporting rigidity/routine around exercise) underscore heterogeneity in definitions, and each can be considered unique in their theoretical underpinnings. Therefore, based on a given theoretical stance, the selection of a given instrument will likely be optimally determined by the particular theoretical model with which the research team or clinician aligns. If a specific measure (e.g., the CET, which aligns with a compulsive theoretical model of exercise behavior (Meyer et al., 2011)) does not encapsulate all of the needed assessment questions, other full scales or sourced items may provide adjunctive coverage to capture additional maladaptive exercise features. Top features that stand out as relevant across theoretical models and multiple scales include affective regulation, weight/shape control, and exercising to compensate for caloric intake.

Another root of the heterogeneity can be attributed to group-consensus decisions to either combine or separate specific aspects of a feature category in the current study. For example, we chose to parse the broad domain of affect regulation. When we did this, we found that avoiding negative affect was present in nine of 15 instruments, decreasing negative affect was present in six of 15 instruments, and increasing positive affect was present in five of 15 instruments (Table 2). Clinically, there is a difference between a desire to avoid negative affect (i.e., exercising to prevent distress) versus decrease negative affect (i.e., exercising to reduce distress) and assessing for each may serve to better understand the function of exercise behavior for a given individual, most notably in a treatment setting. Further, exercising for mood improvement is only weakly correlated with measures of ED pathology (Taranis et al., 2011; Young et al., 2017), suggesting that increasing positive affect might be a less salient feature in driving maladaptive exercise behavior in clinical samples.

Regardless of the semantics we use to describe a theoretical model, or to title a measure, our primary objective in addressing heterogeneity is not to motivate the tearing down and then rebuilding of one measure that assesses all features. Rather, we aim to bring much needed clarity and precision around what measures are capturing which features. It is in the field’s best interest to acknowledge that maladaptive exercise is a heterogeneous construct; assuming we can use instruments interchangeably might come at a considerable cost. Even when constrained to 31 of its most homogeneous features, some of the instruments that we currently use to assess maladaptive exercise appear to combine, conflate, or even omit the features that they are purported to capture. For example, many full scales include items that refer broadly to affect regulation (e.g., EAI: “I used exercise as a way of changing my mood [e.g., to get a buzz, to escape, etc.]” or EOQ: “I feel better after I exercise”) but they lack specificity in regard to how this might occur (e.g., is a mood change achieved by decreasing/avoiding negative affect and/or by increasing positive affect?). We also note the propensity for using sourced single-items, or their combination, to assess maladaptive exercise. While a single item can hone specificity, serve as an important adjunctive tool along with a full scale, and easily capture the frequency of engaging in a certain type of behavior (e.g., exercise for weight/shape control), using a single item alone is naturally limited when variable cut-off scores are used across the literature, and important information can naturally be omitted (e.g., what other reasons for exercise might there be?).

Implications

The current study suggests that clinical conceptualizations of exercise behavior based on one measure (versus another) may be incomplete without further probing. Moreover, using only one measure, or sourced item, is likely to yield disparate results or may not be generalizable to all components and symptoms of maladaptive exercise. As such, clinicians and researchers should take care to align their assessment of exercise with the features they most want to capture (e.g., affect regulation). With that said, some measures may be more effective at assessing more commonly experienced features of maladaptive exercise than others. Within the context of EDs, exercise to control weight/shape or guilt arising when one is unable to exercise are most frequently associated with elevated eating pathology (Hausenblas & Downs, 2002a; Taranis & Meyer, 2011), indicating that items assessing these features may be best for use in epidemiologic studies or other settings where only a single item can be used. Findings here suggest that the OEQ, CET, and EED may be appropriate measures to consider for use if the goal is to capture these common features that are particularly likely to present in clinical samples with eating pathology (i.e., avoidance of negative affect, weight/shape control, compensating for calories, rigidity/routine around exercise). These four features broadly align with the theoretical model of compulsive exercise in EDs (Meyer et al., 2011). If it is important to assess for additional unique features that do not fall within this theoretical model (e.g., impairment), the CES includes items that assess for physical and psychosocial consequences of exercise engagement.

Of note, maladaptive exercise is sometimes conceptualized as a primary behavior that stands apart from exercise in the context of eating pathology (considered a secondary behavior), within a behavioral addiction framework (Weinstein & Szabo, 2023). In situations where the maladaptive behavior is unrelated to calorie burning, a reported desire for weight/shape control, or other body image or eating concerns, the EDS or EAI (assessing dependence or addiction, respectively) might be appropriate. Taken together, it appears that across settings, it is critical to consider which symptoms are especially important to assess and choose measures accordingly given the research or clinical question, population being assessed, and the specific reason for using the assessment.

Future directions

Over and above the self-report measures included here, there have been recent calls for in-laboratory, acute measurement of response to exercise (Kolar & Gorrell, 2021; Schaumberg et al., 2021), both in helping to elucidate affect-regulatory mechanisms that maintain exercise (Kolar & Gorrell, 2021), and to better inform clinical research design that accounts for both the negative and positive reinforcement nature of exercise behavior (Coniglio et al., 2021). Specific to samples who report eating and weight concerns, a majority of work in assessment of exercise has focused either on restrictive EDs like anorexia nervosa (Dalle Grave et al., 2008) or on how exercise might be able to promote weight management (Bardone-Cone et al., 2016). This siloed approach prevents understanding of biobehavioral mechanisms that maintain exercise across transdiagnostic presentations of eating pathology, and the weight spectrum (Mond & Gorrell, 2021). All efforts should continue to be made to assess exercise in ways that allow for greater generalizability and harmonization across the field. For example, efforts should be made to understand how often instruments assessing disparate features of exercise have been assumed to assess similar constructs (e.g., within existing meta-analyses), as these results should be interpreted with caution. This also includes promoting consistency around the use of clinical thresholds and when relevant, assessment across the disease state and diagnoses. In addition, appropriate consideration is necessary for how intersectionality of sociodemographic and cultural differences is represented within the assessment tools we design and use.

Limitations

This study should be considered in terms of several limitations. As noted above and not unique to this study, using a less conservative approach in establishing categories as we have done likely overestimates heterogeneity among and between scales (Chrobak et al., 2018; Fried, 2017; Visontay et al., 2019). We chose this approach to enhance the clinical relevance of findings, but acknowledge that overlap would increase if we were more inclusive across features. Our choice of features themselves might bias findings; although we conducted the evaluation of which features to include or re-define in a systematic manner, it is possible that we have unintentionally omitted or mis-represented a given feature. The study is also limited by our focus on measures that are employed within samples with eating pathology (to the exclusion of measures assessing muscularity-oriented exercise behaviors which may occur outside the context of eating pathology). Our selection was based on the expertise of the authorship team and knowledge of the frequency of use of these measures in ED samples, and aimed at providing the greatest relevance for researchers and clinicians who are working within this space. However, we acknowledge that there are additional instruments assessing maladaptive exercise (e.g., specific to muscle-building) that may capture domains of concerning exercise behavior that we do not address here.

Conclusions

In conclusion, the current examination of instruments assessing maladaptive exercise behavior showed a general lack of content overlap. Heterogeneity is not so much considered problematic in this case, but instead as a cautionary underscore for the importance of not using scales interchangeably. The selection of a given instrument may need to consider both the most commonly measured features, but also other additional items that may be indicated based on the specific assessment question and/or theoretical conceptualization being tested. Furthermore, the heterogeneity in current instruments might inform future approaches to build coherent theoretical or computational models of maladaptive exercise based on potentially relevant mechanisms that can then be empirically tested.

Public Significance Statement.

Many, varied, tools exist for the assessment of maladaptive exercise (e.g., compulsive or compensatory) in the context of eating disorders. Assuming homogeneity across tools contributes to low consensus in the field. We used a Jaccard Index to quantify content overlap among 15 self-report instruments measuring maladaptive exercise. The most commonly used instruments measure heterogenous content. Careful consideration should be taken when synthesizing literature and selecting instruments to use in research.

Data Availability:

The data that support the findings of this study are openly available in Open Science Foundation (OSF) at osf.io/q65v4.