- Journal List
- HHS Author Manuscripts
- PMC11037076
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsem*nt of, or agreement with, the contents by NLM or the National Institutes of Health.
Learn more: PMC Disclaimer | PMC Copyright Notice
Int J Eat Disord. Author manuscript; available in PMC 2024 Apr 23.
Published in final edited form as:
Int J Eat Disord. 2018 Nov; 51(11): 1252–1260.
Published online 2018 Sep 28. doi:10.1002/eat.22957
PMCID: PMC11037076
NIHMSID: NIHMS1985836
PMID: 30265751
Lindsay P. Bodell, PhD,1,2 Katherine Jean Forney, MS,1,3 Jesus Chavarria, PhD,1 Pamela K. Keel, PhD,3 and Jennifer E. Wildes, PhD1
Author information Copyright and License information PMC Disclaimer
Abstract
Objective:
Research evidence supports the clinical significance of subjective feelings of loss of control over eating; however, limited attention has been given to how this construct is assessed. Two measures have been developed in recent years (i.e., Eating Loss of Control Scale [ELOC] and Loss of Control over Eating Scale [LOCES]), but further validation in clinical and non-clinical samples is needed.
Method:
The current study evaluated the psychometric properties, including factor structure, criterion validity, and measurement invariance of the ELOC and LOCES across two groups: (a) a clinical sample of individuals with eating disorders (n = 106) and (b) a non-clinical sample of college students (n = 321).
Results:
Confirmatory factor analyses indicated that the 16-item version of the ELOC and 7-item brief version of the LOCES provided good fit to the data in both samples. These measures were highly correlated (r = .83–.87) and associated with binge-eating and related psychopathology. The ELOC demonstrated partial invariance between men and women and between the clinical and non-clinical samples. The LOCES-brief demonstrated full invariance between men and women and partial invariance between the clinical and non-clinical samples.
Discussion:
Findings suggest that the 16-item ELOC and 7-item LOCES are reliable measures of severity of loss of control eating in clinical and non-clinical samples. Given the brevity of the LOCES-brief and evidence for measurement invariance across sex, it is recommended over the ELOC in heterogeneous samples. Future research is needed to confirm the validity of these measures across individuals with and without eating disorders.
Keywords: assessment, binge eating, eating loss of control scale, loss of control, loss of control over eating scale, measurement invariance, self-report
1 |. INTRODUCTION
The Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 2013) defines binge eating as the consumption of a definitely large amount of food in a short period of time accompanied by the feeling of loss of control (LOC) while eating. Numerous studies highlight the clinical significance of subjective feelings of LOC over eating regardless of the amount consumed, including the impact of LOC eating on the development and course of eating disorders (EDs) and obesity (Goldschmidt, 2017). However, research only recently has focused on the measurement of this construct.
Most studies either have evaluated the presence/absence of LOC eating or determined the combined frequency of objectively and subjectively large binge-eating episodes. The first method does not capture potential influences of degree or severity of LOC on psychopathology and distress. Although the second method captures dimensionality, it confounds the construct of LOC with binge-episode size (Forney et al., 2016). Furthermore, episode frequency does not account for the presence and/or severity of other factors that may contribute to the overall experience of LOC eating, such as negative emotionality. Indeed, clinical experience suggests that the degree of LOC (e.g., a little to extremely) can vary significantly within and between individuals, supporting the need for a dimensional measure focused on the severity of LOC.
To better assess severity of LOC over eating, two self-report questionnaires were developed, the Eating Loss of Control Scale (ELOC; ) and the Loss of Control over Eating Scale (LOCES; ). The ELOC was developed and validated using a treatment-seeking sample of individuals with binge-eating disorder (BED), but only one study has examined its reliability and validity in an independent sample (). In the initial study, an 18-item uni-dimensional version of the ELOC demonstrated high internal consistency, convergent validity with other measures of eating pathology, and discriminant validity with non-significant associations with restraint (Blomquist et al., 2014).
However, the ELOC may perform differently in clinical and non-clinical samples. Hopwood et al. (2018) examined measurement invariance (MI) of a 16-item ELOC, which indicates that the same observed score across groups reflects the same level of the underlying trait and is essential for meaningful group comparisons. Full MI is met when the construct (e.g., LOC eating) is measured by the same items (configural invariance) and the factor loadings (metric invariance) and intercepts (scalar invariance) of these items are equivalent across groups. Findings from their tests of invariance supported that the factor structure of the ELOC was similar across individuals with BED and college students (configural invariance), but metric invariance was not achieved (Hopwood et al., 2018). Thus, comparing means and standard deviations across groups may be problematic. No study to date has examined the factor structure of the ELOC in an independent sample of individuals with EDs, examined convergent or discriminant validity of the ELOC in a non-clinical sample, or examined its test–retest reliability.
The LOCES was validated in a sample of college students (Latner et al., 2014) and further evaluated in students from the same university (). In the initial scale development study, 74 items were refined to a final 24-item scale and a brief 7-item screening instrument (Latner et al., 2014). Results from an exploratory factor analysis indicated three underlying factors consisting of 13-items, which were labeled as (a) behavioral aspects of LOC eating (seven items);(b) cognitive/dissociative aspects of LOC eating (four items); and (c) positive/euphoric aspects of LOC eating (two items). In both college samples, the 24-item scale demonstrated concurrent validity with eating-disorder related constructs. Stefano et al. (2016) also found higher LOCES scores among individuals at high- compared to low-risk for ED pathology, suggesting that this measure may discriminate between clinical and non-clinical samples. Finally, Vannucci and Ohannessian (2018) evaluated the psychometric properties of the LOCES-brief (7-item scale) in a community sample of adolescents. Results provided support for the uni-dimensionality of the LOCES-brief, MI across gender and weight status, and convergent and discriminant validity. Thus, preliminary data support the LOCES and LOCES-brief in non-clinical samples; however, its utility in individuals with EDs remains unknown.
In sum, although a subjective feeling of LOC is the key feature of binge eating (Goldschmidt, 2017), limited attention has been given to how this construct is assessed. Both the ELOC and LOCES represent improvements over traditional assessments of LOC eating, but further validation in clinical populations is warranted. Indeed, neither the ELOC nor LOCES has been evaluated in individuals with a range of EDs—the groups for whom such assessments are needed. Additionally, tests of measurement invariance are needed when investigating mean differences across groups (). Finally, the ELOC and LOCES have not been administered concurrently in the same sample, so the question remains as to which provides the most useful assessment of severity of LOC eating across different samples. Thus, the aims of the current study were (a) to evaluate the underlying factor structure and reliability and validity of the ELOC and LOCES in clinical and non-clinical samples and (b) to explore measurement invariance across these groups.
2 |. METHOD
2.1 |. Participants
2.1.1 |. Eating-disorder participants (clinical sample)
Participants were 106 individuals either seeking or receiving inpatient or day hospital treatment from an academic medical center in the Midwestern United States (n = 48) or individuals participating in a larger study examining cognitive flexibility in EDs (n = 58). ED diagnosis was determined by chart review (treatment-seeking patients) or by the Structured Clinical Interview for DSM-IV-TR Axis I Disorders () (non-treatment seeking participants). Demographic and diagnostic information appear in Table 1.
TABLE 1
Demographic variables and eating-disorder diagnoses
| Clinical sample | Non-clinical sample | |||
|---|---|---|---|---|
| Demographics | n | % | n | % |
| Gender | ||||
| Female | 95 | 89.6 | 239 | 74.5 |
| Male | 11 | 10.4 | 82 | 25.5 |
| Agea | 28.0 | 9.8 | 19.6 | 2.4 |
| Race | ||||
| White | 88 | 83.0 | 257 | 80.1 |
| Black | 8 | 7.5 | 34 | 10.6 |
| Asian/Pacific islander | 4 | 3.8 | 21 | 6.5 |
| Multi-racial/other | 6 | 5.7 | 9 | 2.8 |
| Ethnicity | ||||
| Hispanic | 8 | 7.5 | 79 | 24.6 |
| Non-Hispanic | 98 | 92.5 | 242 | 75.4 |
| Marital status (never married) | 77 | 72.6 | 315 | 98.1 |
| Education (have college degree) | 38 | 35.8 | – | – |
| Eating-disorder diagnosis | n | % | ||
| AN-R | 18 | 17 | ||
| AN-B/P | 26 | 24.5 | ||
| BN | 31 | 29.3 | ||
| BED | 8 | 7.5 | ||
| OSFED | 23 | 21.7 | ||
Open in a separate window
AN-R = anorexia nervosa-restricting subtype; AN-B/P = anorexia nervosa binge-eating/purging subtype; BN = bulimia nervosa; BED = binge-eating disorder; OSFED = other specified feeding or eating disorder.
aAge reported as mean and standard deviation.
2.1.2 |. Student participants (non-clinical sample)
Participants were 483 undergraduate psychology students from a large southeastern state university who completed a survey on disordered eating and affect. The validity of participant responses was checked by examining (a) the length of time to complete the survey and (b) response consistency among three duplicate reverse-worded items. Participants with invalid response styles were excluded from analyses (n = 162, 33.5%). The final sample consisted of 321 participants (see Table 1 for demographic information).
Data were collected with approval of the local Institutional Review Boards, and all participants provided informed consent prior to participation.
2.2 |. Procedure and measures
Participants completed a variety of questionnaires to assess eating and mood related psychopathology. The clinical sample completed these measures at one time point; however, a subset of the non-clinical sample (n = 263; 81.9%) completed questionnaires 2 weeks later to assess test–retest reliability. Of those who completed the second assessment, 90 (34%) were excluded due to invalid responding. Participants received either monetary compensation (clinical sample) or course credit (non-clinical sample).
2.2.1 |. Eating loss of control scale
The final 18-item ELOC (Blomquist et al., 2014) was administered in both samples. Participants are instructed to think about their eating over the past 4 weeks and indicate responses accordingly. Items are scored from 0 (not at all) to 10 (extremely), with higher scores indicating greater severity of LOC over eating.
2.2.2 |. Loss of control over eating scale
The original 24-item LOCES (Latner et al., 2014) was administered in both samples. Participants are instructed to indicate how often they have had a variety of experiences while eating over the past 4 weeks. Items are scored on a 5-point scale (1 = never, 2 = rarely, 3 = occasionally, 4 = often, 5 = always), with higher scores indicating greater severity of LOC over eating.
2.2.3 |. Binge eating scale
The BES () is a 16-item questionnaire that identifies binge eating-related psychopathology and severity, and it was used in the current study to assess concurrent validity of the ELOC and LOCES. Item responses are summed to yield a continuous score ranging from 0 to 46. Internal consistencies for the clinical (α = .92) and non-clinical (α = .93) samples Internal consistencies for the clinical (α = .92) and non-clinical (α = .93) samples were good.
2.2.4 |. Eating disorder examination questionnaire
The EDE-Q () is a questionnaire adapted from the EDE interview to assess ED psychopathology. The frequency item for binge-eating episodes (total number of episodes over the past 28 days) was used to assess concurrent validity, and the 5-item EDEQ restraint subscale was used to assess discriminant validity. Internal consistencies for this subscale were adequate in both samples (α’s = .81–.84).
2.2.5 |. Clinical impairment assessment
The clinical impairment assessment (Bohn et al., 2008) is a 16-item questionnaire that evaluates interpersonal, occupational, and emotional domains of impairment attributable to the presence of an ED (if any). This measure was used to assess concurrent validity of the ELOC and LOCES. Internal consistencies for this measure were good (α’s > .94).
2.2.6 |. Three factor eating questionnaire (only administered in the non-clinical sample)
The three factor eating questionnaire (TFEQ) () is a 51-item self-report questionnaire that assesses three dimensions of eating behavior, Cognitive Restraint of Eating, Disinhibition, and Hunger. The Restraint subscale was used to assess divergent validity of the ELOC and LOCES whereas the Disinhibition and Hunger subscales were used to assess concurrent validity. Internal consistencies for all subscales were good (αs = .82–.87).
2.2.7 |. Positive affect-negative affect schedule-expanded form (only administered in the non-clinical sample)
The positive affect-negative affect schedule-expanded form (PANAS-X; ) is a 60-item questionnaire that assesses positive and negative affect. Items are scored on a five-point scale from 1 (very slightly or not at all) to 5 (extremely). Given associations between negative affect and binge eating (), the negative and positive affect scales were used to examine concurrent and discriminant validity, respectively. Internal consistencies in the current study were good (αs > .88).
2.3 |. Statistical analyses
Mplus Version 8.0 () was used to conduct power analyses and confirmatory factor analyses (CFA) to test the underlying factor structure of the ELOC and LOCES. Following the recommendation of Muthén and Muthén (2002), Monte Carlo analyses were conducted to determine power of 0.80 for each possible CFA (i.e., different structure in ELOC and LOCES in the different samples) with a sample size of 106 and missing data patterns matching missing data in the current study. Based on these results, both the 18-item (Blomquist et al., 2014) and 16-item (Hopwood et al., 2018) ELOC were examined to assess uni-dimensionality and determine best fit to the data in the clinical and non-clinical samples. Four separate CFAs were conducted for the LOCES (Latner et al., 2014), examining the uni-dimensionality of the 24-, 13-, and 7-item versions and the multi-dimensionality of the 13-item version to determine underlying structure and best fit to the data. Only the uni-dimensional 13-item and 7-item LOCES were examined in the clinical sample due to inadequate power to examine the 24-item or multi-factor CFA (). A comparative fit index (CFI) >0.90 and standardized root mean square residual (SRMR) <0.10 were used as indicators of adequate fit (Hu and Bentler, 1998).
Based on results from CFAs, scale scores and Cronbach’s alphas were calculated in SPSS 24.0 (SPSS, 2016). Criterion validity was examined by correlating total mean scores of the ELOC and LOCES with measures of eating pathology, impairment, and affect (De Von et al., 2007) using Pearson’s correlations. All variables in the non-clinical sample except positive affect and TFEQ Restraint and Hunger subscales underwent square root transformation to correct for non-normality.1 Only binge-eating frequency was positively skewed in the clinical sample and transformed using square root transformation for correlational analyses. We hypothesized that concurrent validity would be evidenced by Pearson correlations >.45 with binge-eating severity, binge-eating frequency, disinhibition, hunger, impairment, and negative affect, and that discriminant validity would be evidenced by correlations with absolute values <.45 with dietary restraint and positive affect (De Von et al., 2007).
Next, exploratory multi-group CFAs were conducted to test the MI of the factor structure of the ELOC and LOCES between the clinical and non-clinical samples and sex (within the non-clinical sample) using structural equation modeling (SEM) (Bryne, 2012; ). To test MI, a sequential pattern of more restrictive models were compared. First, configural invariance was tested to determine whether the same items are indicators for the same factor(s) across groups with no constraints on item loadings or intercepts. Next, metric invariance was assessed by constraining item loadings to be equal to determine whether the strength of the relationship between each item and LOC severity was similar across groups. Lastly, scalar invariance was tested by constraining item loadings and intercepts to be equal across groups to determine whether levels of the underlying items (i.e., intercepts) would be similar across groups. Improvement in model fit between the unconstrained and constrained models was tested with a χ2 difference test. A non-significant χ2 difference test indicates MI at the level tested. If the less restrictive test was not supported (e.g., metric), then stricter tests of invariance were not conducted (e.g., scalar).
2.3.1 |. Missing data
For the clinical sample, 20 participants (18.9%) had missing data on the ELOC or LOCES, including one individual who did not respond to any LOCES items. There were no differences between individuals with versus without missing data on any variables of interest. For the non-clinical sample, 14 participants (2.9%) had missing data on the ELOC or LOCES, including five participants who did not respond to any ELOC items and three individuals who did not respond to any LOCES items. There were no differences between individuals with or without missing data in the non-clinical sample. As such, data from both groups were assumed to be missing at random, and full information maximum likelihood was used in analyses to account for missing data (Enders, 2001).
3 |. RESULTS
3.1 |. EATING LOSS OF CONTROL SCALE
3.1.1 |. Confirmatory factor analysis and reliability
Results from CFAs supported a one-factor model of the ELOC. Both the 18- and 16-item ELOC demonstrated adequate fit to the data, but the 16-item version demonstrated slightly better fit (Table 2). Factor loadings for the 16-item version ranged from .53 to .92 in the clinical sample and .30 to .84 in the non-clinical sample. Similar to prior studies, internal consistencies were excellent (αs = .92–.96), supporting the uni-dimensionality of the construct being measured. The ELOC also demonstrated high test–retest reliability in the non-clinical sample (r = .84, p < .001).
TABLE 2
Confirmatory factor analysis fit indices
| Clinical sample | Non-clinical sample | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Eating loss of control scale (ELOC) | ||||||||||
| Model | χ2 | df | CFI | SRMR | Alpha | χ2 | df | CFI | SRMR | Alpha |
| 1 | 286.48 | 135 | 0.91 | 0.051 | 0.96 | 423.79 | 135 | 0.90 | 0.052 | 0.93 |
| 2 | 183.59 | 104 | 0.94 | 0.041 | 0.96 | 314.75 | 104 | 0.91 | 0.051 | 0.92 |
| Loss of control over eating scale (LOCES) | ||||||||||
| Model | χ2 | df | CFI | SRMR | Alpha | χ2 | df | CFI | SRMR | Alpha |
| 3 | – | – | – | – | – | 1,501.69 | 252 | 0.77 | 0.072 | 0.96 |
| 4 | 242.79 | 65 | 0.83 | 0.086 | 0.94 | 571.62 | 65 | 0.78 | 0.079 | 0.91 |
| 5 | – | – | – | – | – | 221.23 | 62 | 0.93 | 0.053 | 0.72–0.81 |
| 6 | 35.71 | 14 | 0.96 | 0.030 | 0.94 | 101.75 | 14 | 0.94 | 0.046 | 0.91 |
Open in a separate window
Model 1 = 18-item ELOC; Model 2 = 16-item ELOC; Model 3 = 24-item LOCES; Model 4 = 13-item LOCES uni-dimensional; Model 5 = 13-item LOCES multi-dimensional; Model 6 = 7-item LOCES “brief version”; χ2 = Chi-square; CFI = comparative fit index; df = degrees of freedom; SRMR = standardized root mean square residual.
3.1.2 |. Criterion validity
Means and standard deviations of the clinical variables and their associations with the ELOC appear in Tables 3 and and4.4. Among both samples, convergent validity was demonstrated by positive correlations with binge-eating severity and frequency. The ELOC also was correlated significantly with hunger, disinhibition, impairment, and negative affect in the non-clinical sample; however, it was not associated with impairment in the clinical sample (p = .23). With regard to discriminant validity, the ELOC was not associated with restraint in the clinical sample (p = .97) and demonstrated smaller statistically significant correlations with restraint (r = .21–.42, ps < .001) and positive affect (r = −.31, p < .001) in the non-clinical sample.
TABLE 3
Pearson correlations among loss of control measures and eating-disorder related psychopathology in the clinical sample of individuals with eating disorders (N = 106)
| 1 | 2 | 3 | 4 | 5 | 6 | |
|---|---|---|---|---|---|---|
| 1. Eating loss of control scale (16-item) | – | |||||
| 2. Loss of control over eating scale (brief ) | .87** | – | ||||
| 3. EDEQ binge-eating frequency | .64** | .73** | – | |||
| 4. Binge-eating scale | .86** | .83** | .66** | – | ||
| 5. Clinical impairment assessment | .12 | .18 | −.02 | .21* | – | |
| 6. EDEQ restraint | .004 | .01 | −.12 | .07 | .56** | – |
| Means (SD) | 4.59 (3.04) | 2.64 (1.15) | 9.06 (11.38) | 24.25 (11.14) | 31.66 (11.49) | 3.81 (1.64) |
Open in a separate window
EDEQ = Eating Disorders Examination Questionnaire.
*p < .05;;
**p < .001.
TABLE 4
Pearson correlations among loss of control measures and eating-disorder related psychopathology in a non-clinical sample of undergraduate college students (N = 321)
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. | Eating loss of control scale (16-item) | - | ||||||||||
| 2. | Loss of control over eating scale (brief ) | .81** | – | |||||||||
| 3. | EDEQ binge-eating frequency | .68** | .64** | – | ||||||||
| 4. | Binge-eating scale | .80** | .78** | .62** | – | |||||||
| 5. | TFEQ hunger | .68** | .68** | .45** | .72** | – | ||||||
| 6. | TFEQ disinhibition | .72** | .74** | .57** | .75** | .70** | – | |||||
| 7. | Negative affect | .45** | .47** | .33** | .49** | .39** | .38** | – | ||||
| 8. | Clinical impairment assessment | .66** | .68** | .51** | .69** | .47** | .56** | .58** | – | |||
| 9. | TFEQ restraint | .21** | .24** | .16* | .24** | .03 | .20** | .14* | .47** | – | ||
| 10. | EDEQ restraint | .42** | .43** | .32** | .47** | .24** | .35** | .31** | .61** | .71** | – | |
| 11. | Positive affect | −.31** | −.27** | −.25** | −.30** | −.23** | −.29** | −.24** | −.35** | .03 | −.03 | – |
| Means (SD) | 2.64 (1.81) | 1.97 (0.87) | 1.70 (3.45) | 12.66 (8.77) | 6.66 (3.65) | 6.41 (3.81) | 21.10 (7.01) | 9.26 (9.44) | 9.63 (5.31) | 1.73 (3.45) | 32.17 (7.55) | |
Open in a separate window
TFEQ = Three Factor Eating Questionnaire; SD = standard deviation.
*p < .05;;
**p < .01; EDEQ = Eating Disorders Examination Questionnaire.
3.1.3 |. Exploratory tests of invariance
Full MI for sex was not supported (Table 5). Although configural and metric invariance were found, comparison of scalar and metric models resulted in a significant χ2 difference test. Elevated intercepts for women indicated that means for the ELOC items were higher among women than men (i.e., scalar non-invariance).
TABLE 5
Tests of measurement invariance
| Δχ2 (df ) | P | CFI | SRMR | |
|---|---|---|---|---|
| Eating loss of control scale (ELOC) | ||||
| Sex | ||||
| Configural | -– | – | 0.91 | 0.06 |
| Metric | 20.59 (15) | .15 | 0.90 | 0.07 |
| Scalara | 41.69 (12) | <.001 | 0.89 | 0.08 |
| Sample | ||||
| Configural | – | – | 0.92 | 0.05 |
| Metricb | 28.40 (12) | .005 | 0.92 | 0.06 |
| Scalar | – | – | – | – |
| Loss of control over eating scale (LOCES) | ||||
| Sex | ||||
| Configural | – | – | 0.94 | 0.05 |
| Metric | 9.67 (6) | .14 | 0.94 | 0.06 |
| Scalar | 6.28 (6) | .39 | 0.94 | 0.06 |
| Sample | ||||
| Configural | – | – | 0.95 | 0.04 |
| Metricc | 5.17 (4) | .27 | 0.94 | 0.05 |
| Scalard | 4.65 (3) | .17 | 0.94 | 0.05 |
Open in a separate window
Note. Tests of invariance were conducted using the 16-item version of the Eating Loss of Control Scale [ELOC] and 7-item brief version of the Loss of Control over Eating Scale; Δχ2 = change in chi-square; df = degrees of freedom; CFI = comparative fit index; SRMR = standard root mean square residual.
aIntercepts for “feel helpless to control eating urges,” “give in to an impulse to eat even though not hungry,” “feel upset by the feeling that couldn’t stop eating,” “hard to stop thinking about food you were craving” allowed to vary.
bLoadings for “afraid of losing control over eating,” “feel helpless…,” “hard to stop thinking…” allowed to vary.
cLoadings for “…ball rolling…,” “could not do anything other than eat” allowed to vary.
dIntercepts for “…ball rolling…,” “could not do anything…,” “helpless about controlling my eating” allowed to vary.
There also was limited support for MI between samples (Table 5). Configural and metric models showed adequate fit, but comparison of these models resulted in a significant χ2 difference test (Δχ2 [df] = 41.48 [15], p < .001). Several items had large loading differences between clinical and non-clinical groups; thus, the constraint on three items (afraid of losing control over eating, feel helpless to control eating urges, hard to stop thinking about food you were craving) was relaxed sequentially. Although fit indices for this modified metric model were adequate, the χ2 difference test remained significant, indicating metric non-invariance.
3.2 |. Loss of Control over Eating Scale
3.2.1 |. Confirmatory factor analysis and reliability
Results from CFAs of the uni-dimensional 24- and 13-item scales indicated poor fit to the data (Table 2). When the 13 items were loaded onto the three factors found by Latner et al. (2014), this model fit significantly better than the 13-item uni-dimensional model (Δχ2 [df] = 117.33 [3], p < .001) in the non-clinical sample. Finally, the 7-item version provided excellent fit to the data in both samples. For parsimony and consistency with prior research (), the LOCES-brief was examined in subsequent analyses. Factor loadings were high (r = .47–.81), and internal consistencies were good (αs = .91–.94) across samples. The LOCES-brief also demonstrated high test–retest reliability in the non-clinical sample (r = .79, p < .001).
3.2.2 |. Criterion validity
Means and standard deviations of the clinical variables and their associations with the LOCES-brief appear in Tables 3 and and4.4. Convergent validity was demonstrated by positive correlations with binge-eating severity and frequency in both samples. The LOCES-brief also demonstrated large associations with hunger, disinhibition, impairment, and negative affect in the non-clinical sample. Contrary to hypotheses, the LOCES-brief demonstrated a small, non-significant correlation with impairment (r = .18, p = .07) in the clinical sample. As evidence of discriminant validity, the LOCES-brief was not associated with restraint (r = .01) in the clinical sample and demonstrated smaller correlations with restraint (r = .24–.43) and positive affect (r = −.27) in the non-clinical sample.
3.2.3 |. Exploratory tests of invariance
For the LOCES-brief, results supported full MI for sex in the non-clinical sample (Table 5). Specifically, there was adequate model fit for the configural, metric, and scalar models and non-significant χ2 difference tests between the most and least restrictive models.
In contrast, partial invariance was found between clinical and non-clinical samples (Table 5). Configural and metric models showed adequate fit to the data, but the χ2 difference test was significant (Δχ2 [df] = 22.91 [6], p < .001). Given large loading differences between groups, the constraint on two items (eating felt like a ball rolling down a hill, could not do anything other than eat) was relaxed sequentially. Fit indices for this modified metric model were adequate, and comparison with the configural model resulted in a non-significant χ2 difference test. The scalar model showed acceptable fit (CFI = .92, SRMR = .09); however, a significant χ2 difference test (Δχ2 [df] = 55.43 [6], p < .001) indicated scalar non-invariance. Relaxing the constraints on three items (eating felt like a ball rolling down a hill, could not do anything other than eat, felt helpless about controlling my eating) resulted in adequate fit and a non-significant χ2 difference test, indicating partial scalar invariance.
4 |. DISCUSSION
This study is the first to compare and evaluate the psychometric properties, including test–retest reliability and MI, of the ELOC and LOCES in both clinical and non-clinical samples. Findings suggest that these measures reliably assess severity of LOC eating and support their use in studies examining mechanisms and outcomes related to this construct. Given the brevity of the LOCES-brief and slightly greater evidence for MI across sex, the LOCES-brief is recommended over the ELOC in heterogeneous samples. However, given the preliminary nature of these analyses and relatively small sample sizes, particularly for the clinical sample, additional research is needed to confirm the validity of these measures across individuals with and without eating disorders.
Importantly, results from factor analyses add to the growing literature on the construct validity of the ELOC and LOCES. Consistent with findings from Hopwood et al. (2018), the 16-item ELOC demonstrated good fit to the data in both clinical and non-clinical samples with high internal consistency and test–retest reliability. Thus, this version is recommended over the original 18-item version. Although factor analyses confirmed the uni-dimensionality of the ELOC, the 24- and 13-item uni-dimensional models of the LOCES had poor fit to the data. In contrast, the 13-item three-factor model demonstrated good fit with relatively high internal consistencies, supporting its multi-dimensionality in the non-clinical sample. The one-factor model for the 7-item scale also demonstrated good fit to the data with high internal consistencies and test–retest reliability, indicating stability of LOC eating over a 2-week period. Based on these findings, both the 13-item (multi-dimensional) and 7-item (uni-dimensional) LOCES are recommended over the original 24-item version and may be useful tools for assessing severity of LOC eating.
The ELOC and LOCES-brief demonstrated convergent validity with associations with other assessments of eating pathology. In particular, we found high correlations among the BES, EDEQ binge-eating frequency, ELOC, and LOCES-brief. Item overlap across measures likely contributed to high correlations. Additionally, correlations were slightly greater with the BES than with EDE-Q binge-eating frequency, providing some evidence that LOC severity may be distinct from LOC frequency.
Consistent with prior findings, we found significant associations between the LOC measures and impairment in the non-clinical sample. However, neither measure was correlated with impairment in the clinical sample. This null finding is somewhat surprising, given substantial research demonstrating associations between LOC eating and ED severity (Goldschmidt, 2017). Notably, however, the majority of prior studies have been conducted in community-based samples or in samples of individuals with bulimia nervosa (BN)-spectrum disorders rather than the full spectrum of EDs. Moreover, a large proportion of individuals with EDs in the current study were treatment-seeking, and this heterogeneity likely confounded any associations between severity of LOC eating and impairment. Specifically, individuals with AN were the most likely to be receiving inpatient treatment and to demonstrate the lowest severity of LOC eating.
Finally, evidence of discriminant validity came from non-significant or small associations between ELOC and LOCES scores and dietary restraint. Although dietary restraint has been linked to binge eating in BN (), our findings are consistent with a prior study of the ELOC (Blomquist et al., 2014) and could reflect limitations in the measurement of restraint () or our transdiagnostic sample. Indeed, common measures of restraint focus on attempts to restrict food intake regardless of success. These efforts may increase risk for LOC eating for many individuals, but may decrease risk for others, leading to overall lower or non-significant associations between measures of restraint and severity of LOC eating found in our samples.
In addition to evaluating the reliability and validity of the ELOC and LOCES, our study aimed to explore MI. Consistent with findings from Hopwood et al. (2018), the ELOC demonstrated limited evidence of invariance between clinical and non-clinical samples. Configural invariance suggests that the ELOC is assessing the same general structure (i.e., one-factor model) of LOC severity across individuals with and without EDs. However, direct comparisons of scores may be problematic, given metric non-invariance. Individuals with and without EDs may not be interpreting questions in the same way, and the same observed score on the ELOC may not reflect the same level of LOC severity in these groups. Taken together, the ELOC should not be used as a diagnostic tool in non-clinical samples for assessing pathological LOC eating (Hopwood et al., 2018), and different norms should be established for clinical and non-clinical groups.
Furthermore, the current study was the first to examine invariance of the ELOC across men and women. Configural and metric invariance for sex indicate that men and women conceptualize LOC eating in the same way, but scalar non-invariance reflects that women have higher initial severity of LOC eating than men. In other words, for the same severity of LOC eating, women and men will have different response ratings for the items. This finding is somewhat consistent with findings on gender differences in ED symptoms (; Striegel-Moore et al., 2009), and suggests that results from comparisons of the ELOC across sex may be confounded by differences in the scaling properties of this measure.
In contrast to the ELOC, the LOCES-brief demonstrated full MI between men and women, which replicates and extends prior research conducted in community adolescents (). Current findings suggest that the LOCES-brief performs similarly in male and female college students. Never-theless, direct comparisons of LOCES-brief scores between clinical and non-clinical samples should be interpreted with caution, given evidence of partial invariance. Findings indicate that any differences in LOCES-brief scores across individuals with or without eating disorders could be attributed either to true differences in severity of LOC eating or to differences in the way specific items may relate to the LOC eating construct being measured. For example, findings suggest that the item “my eating felt like a ball rolling down a hill…” does not assess severity of LOC eating in similar ways and may be more indicative of LOC eating severity in individuals with than without EDs. Thus, consistent with the ELOC, caution should be employed when using the LOCES-brief in a non-clinical sample to identify a clinical subgroup, and different norms may be needed for clinical versus non-clinical populations.
Main strengths of the current study include a heterogeneous sample of individuals with EDs and a large, diverse sample of college students. An additional strength includes the independent replication of the psychometric properties of the ELOC and LOCES across two distinct samples. However, there are several limitations to acknowledge. First, we were not adequately powered to examine all proposed CFAs in the clinical sample (i.e., 24-item LOCES) or to examine MI across diagnostic categories. Additionally, only a subset of the clinical sample was diagnosed using a standardized interview; thus, the reliability of the diagnoses may be limited. Finally, because the data from the clinical and non-clinical samples were collected in different regions of the United States, limited support for invariance between these samples could reflect cultural/regional differences in addition to differences between clinical and non-clinical groups. Thus, generalizability may be limited.
In conclusion, findings provide additional support for the use of the ELOC and LOCES-brief as measures of LOC eating severity. The ELOC performs well in hom*ogenous samples (e.g., individuals with BED), but the LOCES-brief may be applied across a wider group of individuals (e.g., men and women). Although future studies are needed to examine further the utility of these assessments across individuals with EDs, these measures have the potential to inform our understanding of mechanisms underlying binge eating and the impact of LOC eating severity on treatment and outcome.
ACKNOWLEDGMENTS
The authors would like to thank Meghan Martinho, Rachel Kolko, and Carina Gobes for their help with data collection and database assistance. They also would like to thank all of the individuals who participated in this study. This study was funded, in part, by National Institute of Mental Health grants T32 MH93311, T32 MH082761, and F31 MH105082 and National Institute on Alcohol Abuse and Alcoholism grant 3R01AA013746-14S1.
Footnotes
1After transformation, the LOCES and EDEQ binge-eating frequency remained positively skewed in the non-clinical sample; thus, Spearman correlations (Rho) also were conducted with these variables. Given very minimal differences between parametric and nonparametric correlations, the parametric correlations are reported for all variables for ease of interpretation.
REFERENCES
- American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders: DSM-5. Washington, D.C: American Psychiatric Association. [Google Scholar]
- Blomquist K,K, Roberto CA, Barnes RD, White MA, Masheb MA, & Grilo CM (2014). Development and validation of the eating loss of control scale. Psychological Assessment, 26(1), 77–89. 10.1037/a0034729 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Bohn K, Doll HA, Cooper Z, O’Connor M, Palmer RL, & Fairburn CG (2008). The measurement of impairment due to eating disorder psychopathology. Behavior Research and Therapy, 46(10), 1105–1110. 10.1016/j.brat.2008.06.012 [CrossRef] [Google Scholar]
- Byrne BM (2012). Structural equation modeling: Basic concepts, applications, and programming. New York, NY: Routledge. [Google Scholar]
- De Von HA, Block ME, Moyle-Wright P, Ernst DM, Hayden SJ, … Kostas-Polston E (2007). A psychometric toolbox for testing validity and reliability. Journal of Nursing Scholarship, 39, 155–164. [PubMed] [Google Scholar]
- Enders CK (2001). A primer on maximum likelihood algorithms available for use with missing data. Structural Equation Modeling, 8(1), 128–141. [Google Scholar]
- Fairburn CG, & Beglin SJ (2008). Eating disorder examination questionnaire (EDE-Q 6.0). In Fairburn CG (Ed.), Cognitive behavior therapy and eating disorders. New York: Guildford Press. [Google Scholar]
- First M, Spitzer RL, Gibbon M, & Williams JBW (1997). Structured clinical interview for DSM-IV axis I disorders—patient edition (SCID/P). Biometrics Research Department: New York State Psychiatric Institute: New York. [Google Scholar]
- Forney KJ, Bodell LP, Haedt-Matt AA, & Keel PK (2016). Incremental validity of the episode size criterion in binge-eating definitions: An examination in women with purging syndromes. International Journal of Eating Disorders, 49(7), 651–662. 10.1002/eat.22508 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Goldschmidt AB (2017). Are loss of control while eating and overeating valid constructs? A critical review of the literature. Obesity Reviews, 18(4), 412–449. 10.1111/obr.12491 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Gormally J, Black S, Datson S, & Rardin D (1982). The assessment of binge eating severity among obese persons. Addictive Behaviors, 7, 47–55. 10.1016/0306-4603(82)90024-7 [PubMed] [CrossRef] [Google Scholar]
- Haedt-Matt AA, & Keel PK (2011). Revisiting the affect regulation model of binge eating: A meta-analysis of studies using ecological momentary assessment. Psychological Bulletin, 13(4), 660–681. 10.1037/a0023660 [CrossRef] [Google Scholar]
- Herman CP, & Polivy J (1975). Anxiety, restraint, and eating behavior. Journal of Abnormal Psychology, 84, 66–72. 10.1037/0021-843X.84.6.666 [CrossRef] [Google Scholar]
- Hopwood CJ, Nye CD, Blomquist KK, & Grilo CM (2018). Confirmatory validation and measurement equivalence of the eating loss of control scale in binge eating and non-clinical samples. Journal of Psychopathology and Behavioral Assessment, 40, 476–483. 10.1007/s10862-018-9643-0 [CrossRef] [Google Scholar]
- Hoyle RH, & Smith GT (1994). Formulating clinical research hypotheses as structural equation models: A conceptual overview. Journal of Consulting and Clinical Psychology, 62, 429–440. [PubMed] [Google Scholar]
- Hu L, & Bentler PM (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424–453. [Google Scholar]
- Latner JD, Mond JM, Kelly MC, Haynes SN, & Hay PJ (2014). The loss of control over eating scale: Development and psychometric evaluation. International Journal of Eating Disorders, 47(6), 647–659. 10.1002/eat.22296 [PubMed] [CrossRef] [Google Scholar]
- Lipson SK, & Sonneville KR (2018). Eating disorder symptoms among undergraudate and graduate students at 12 U. S., colleges and universities. Eating Behaviors, 24, 81–88. 10.1016/j.eatbeh.2016.12.003 [CrossRef] [Google Scholar]
- Muthén LK, & Muthén BO (2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling, 9(4), 599–620. 10.1207/s15328007sem0904_8 [CrossRef] [Google Scholar]
- Muthén LK, & Muthén BO (1998–2017). Mplus User’s Guide. Los Angeles, CA: Author. [Google Scholar]
- SPSS. (2016). Statistics for Windows [computer program]. Version 24.0. Armonk, NY: IBM Corp. [Google Scholar]
- Stefano EC, Wagner AF, Mond JM, Cicero DC, & Latner JD (2016). Loss of control over eating scale (LOCES): Validation in undergraduate men and women with and without eating disorder symptoms. Eating Behaviors, 23, 137–140. 10.1016/j.eatbeh.2016.09.005 [PubMed] [CrossRef] [Google Scholar]
- Stice E, Sysko R, Roberto CA, & Allison S (2010). Are dietary restraint scales valid measures of dietary restriction? Additional objective behavioral and biological data suggest not. Appetite, 54, 331–339. 10.1016/j.appet.2009.12.009 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Striegel-Moore RH, Rosselli F, Perrin N, DeBar L, Wilson GT, May A, & Kraemer HC (2009). Gender differences in the prevalence of eating disorder symptoms. International Journal of Eating Disorders, 42(5), 471–474. 10.1002/eat.20625 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
- Stunkard AJ, & Messick S (1985). The three-factor eating questionnaire to measure dietary restraint, disinhibition and hunger. Journal of Psychosomatic Research, 29(1), 71–83. [PubMed] [Google Scholar]
- Vandenberg RJ, & Lance CE (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70. [Google Scholar]
- Vannucci A, & Ohannessian CM (2018). Psychometric properties of the brief loss of control over eating scale (LOCES-B) in early adolescents. International Journal of Eating Disorders, 51, 459–464. 10.1002/eat.22845 [PubMed] [CrossRef] [Google Scholar]
- Watson D, & Clark LA (1994). The PANAS-X: Manual for the positive and negative affect schedule-expanded form. Ames: University of Iowa. 10.13072/midss.438 [CrossRef] [Google Scholar]
