Enhancing Motivation to Reduce Restrictive Eating: A Novel Approach

Abstract

Objective:

Despite growing interest in leveraging motivational techniques to address restrictive eating, it is not yet clear how to most effectively promote motivation to reduce this behavior. Drawing from a behavioral economic framework, the present study evaluates a novel approach for increasing motivation to address disordered eating by amplifying the potential benefits of reducing dietary restriction and the consequences of maintaining disordered behaviors.

Method:

A sample of 126 undergraduate students engaging in restrictive eating participated in a 7-day online experiment. Participants were randomized to one of four conditions: Amplified Benefits and Consequences, Amplified Benefits, Amplified Consequences, or a control condition. Growth curve models were estimated to examine the extent to which experimental conditions led to changes in eating pathology and motivation over the study period.

Results:

Initial reductions in dietary restraint were observed in conditions where the benefits of reducing restriction were amplified; however, these reductions were not sustained over the 7- day study period. The greatest increases in motivation were observed when both the benefits of reducing restriction and the consequences of maintenance were amplified.

Discussion:

Interventions designed to enhance motivation alone are unlikely to yield sustained reductions in eating disorder symptoms and therefore should be accompanied or followed by targeted interventions which directly address restrictive eating behaviors and maladaptive cognitions about shape and weight.

1 |. Introduction

Eating disorders (American Psychiatric Association 2013) are often chronic (Wonderlich et al. 2012), and full recovery can be difficult to achieve (Eddy et al. 2017; Keel and Brown 2010). Most individuals do not fully recover during short-term treatment (Byrne et al. 2017; Grilo et al. 2008; Lock and Le Grange 2019), and many do not achieve full remission even after 10 or more years (Dobrescu et al. 2020; Von Holle et al. 2008). Some have suggested that ambivalence about treatment may partially account for these low recovery rates (Lindgren et al. 2015; Schmidt and Treasure 2006; Williams and Reid 2010). Accordingly, it has become common to leverage motivational techniques to promote treatment engagement among individuals engaging in disordered eating (Geller et al. 2013; Hötzel et al. 2014; Van Der Kaap-Deeder et al. 2014).

Motivation-focused techniques, typically drawn from motivational interviewing (Macdonald et al. 2012), focus on enhancing readiness for change and improving subsequent treatment outcomes (Bonder and Mantler 2014; Feld et al. 2001). However, reviews and meta-analyses have yielded mixed results about the effectiveness of these interventions for disordered eating, and it is unclear whether increasing motivation reliably translates to symptom reduction. Whereas some interventions appear to effectively reduce binge eating, most do not reduce restrictive eating or other compensatory behaviors (Denison-Day et al. 2018; Fetahi, Søgaard, and Sjögren 2022; Knowles, Anokhina, and Serpell 2013). This has led some to conclude that motivational interventions do not significantly improve existing interventions for disordered eating (Denison-Day et al. 2018; Knowles, Anokhina, and Serpell 2013).

One potential explanation for the modest effectiveness of motivational interventions is that the benefits of reducing engagement in disordered eating are not viewed as sufficiently valuable to warrant the effort required to change. In economic terms, individuals “calculate” the value of every action by estimating the difference between its “cost” (e.g., labor required) and expected reward (Rangel, Camerer, and Montague 2008; Verharen, Adan, and Vanderschuren 2020). Within the context of disor- dered eating, there may be a high reward value associated with maintaining disordered behaviors (Bohon and Stice 2011; Stice and Yokum 2023), such as self-starvation in Anorexia Nervosa (Keating 2010; Keating et al. 2012). Conversely, recovery-oriented behaviors, such as eating more, may feel punishing (Anderson et al. 2021). Additionally, the long-term consequences of disordered eating (Kärkkäinen et al. 2018) may not be perceived as sufficiently “costly” to motivate change.

Notably, how to most effectively foster motivation to reduce disordered eating remains unclear. Qualitative studies have yielded valuable insights into mechanisms that may enhance motivation, such as awareness of cultural forces encouraging disordered behaviors, or the harms caused by weight stigma (Venturo-Conerly et al. 2020; Wasil et al. 2019). Although these qualitative data can inform hypotheses about factors motivating recovery, they cannot establish causality. Experiments are necessary to determine which malleable factors influence motivation to reduce disordered eating.

Given these considerations, this study evaluated a novel approach for increasing motivation to reduce disordered eating in college students. Using an experimental design, we tested whether amplifying either (1) the benefits of reducing disordered eating, (2) the consequences of maintaining disordered eating, or (3) both, increased motivation to reduce disordered behaviors. We hypothesized that amplifying both the anticipated benefits of “recovery” (i.e., reduction or cessation of disordered eating) and consequences of maintenance would most strongly increase motivation to decrease engagement in disordered eating. In economic terms, we expected that this manipulation would increase the salience of both the reward associated with behavior change and the cost associated with maintaining current behaviors. Based on evidence indicating that anticipated negative consequences exert a stronger influence on behavior than anticipated rewards, we expected that effects would be slightly weaker when only the negative consequences of maintenance were amplified, and weakest when only the benefits of reducing disordered eating were amplified (Baumeister et al. 2001).

We focused on restrictive eating, which is prevalent on college campuses (Barrack et al. 2019) and is among the strongest predictors of subsequent eating pathology (Schaumberg et al. 2016). Individuals who engage in restrictive eating, even in the absence of other symptoms, may experience comparable levels of functional impairment to individuals with full-syndrome Anorexia Nervosa (Johnson-Munguia et al. 2024). Given the normalization of dieting among young people (Samman, Petocz, and Samman 2012), college students may not be intrinsically motivated to address potentially dangerous restrictive eating, and thus represent a particularly important group to target with preventive interventions.

Notably, the present study leverages an online format. Lack of transportation and availability of providers prevents the largest portion of people from engaging in care, but stigma also delays many from seeking help (Dockery et al. 2015). Additionally, individuals who do not meet full criteria for a specific eating disorder may struggle to find treatment that is covered by their insurance, creating additional barriers to treatment (Thompson and Park 2016). Scalable online interventions provide a potential solution to these problems (Rahman et al. 2020), as they do not necessarily rely on provider interaction, can be completed from the privacy of one’s home, and typically cost substantially less than in-person treatment.

In sum, using a brief, online experimental design, this study tested whether amplifying the benefits of reducing disordered eating and/or the consequences of maintenance increased motivation to address disordered (i.e., restrictive) eating among college students. We reasoned that if our hypotheses were supported, this would lend further support to the notion that anticipated consequences are powerful causal drivers of behaviors (DeWall et al. 2016; Huang et al. 2020). It would also represent an important step towards developing efficient, scalable interventions which increase motivation to reduce engagement in disordered eating.

2 |. Methods

2.1 |. Participants and Recruitment

Participants were 126 undergraduate students recruited from psychology courses who were at least 18, fluent in English, and engaging in dietary restrictions without binging. Participants were eligible if they reported at least two episodes of restrictive eating within the last month, as evaluated by the Dietary Restriction Screener (DRS; Haynos and Fruzzetti 2015). Because many measures of dietary restriction do not differentiate between problematic and unproblematic restriction, the DRS was developed specifically to identify individuals engaging in maladaptive restrictive eating. The DRS was also designed to capture objective restriction, rather than perceived or relative restriction (e.g., eating less than one prefers). Prior research suggests that individuals who screened positive for restrictive eating using the DRS, compared to those who screened negative, reported greater eating disorder symptoms, and consumed fewer calories during a test snack provided in a laboratory setting (Haynos and Fruzzetti 2015). Although the original DRS requires only one episode of restrictive eating for a positive screen, we adopted a slightly more conservative threshold of two episodes.

To further ensure the presence of clinically significant dietary restriction, participants were also required to endorse preoccupation with either (1) food, eating, or calories, or (2) shape or weight, within the last 4 weeks. To this end, they responded to two items drawn from the Eating Disorders Examination Questionnaire (Fairburn and Bèglin 1994) which assessed the extent to which thinking about eating or weight interfered with their ability to concentrate within the last 28 days. These items were intended to evaluate the extent to which eating pathology may be responsible for impairment in other areas of functioning (e.g., social, occupational). For screening purposes, responses were dichotomized, and preoccupation was categorized as either present or absent.

Lastly, to ensure that dietary restriction was the primary clinical concern, rather than binge eating, participants responded to single items assessing overeating and a concomitant sense of loss of control. They were only excluded if both were present within the last month.

2.2 |. Procedures

Procedures were approved by the Institutional Review Board at Florida State University. Data were collected anonymously via Qualtrics. Participants provided informed consent prior to beginning the study. Participants were enrolled in the study for a total of 7 days. Baseline measures were completed on their first day of study enrollment, following randomization to experimental condition (please see Section 2.2.1, below). Over the subsequent 6 days, participants received daily surveys via email every 24 h and were given 24 h to complete each survey. Compensation was provided via course credit. Upon debriefing, participants were provided with a list of relevant mental health resources.

2.2.1 |. Experimental Design

Participants were randomized to one of four conditions: Amplified Benefits and Consequences (n = 31), Amplified Benefits (n = 30), Amplified Consequences (n = 32), or a control condition (n = 33). Participants were blind to condition. In every condition, participants completed daily self-report measures. Depending on condition, participants also completed the following manipulations.

2.2.1.1 |. Amplified Benefits and Consequences.

Prior to completing surveys, on the first day of the study, participants in the Amplified Benefits & Consequences condition were prompted to write a brief, detailed narrative describing a day in their future if they did not restrict their caloric intake and were not concerned about their body weight or shape (see Supporting Information). Due to the possibility that some participants might associate cessation of dietary restriction with undesirable outcomes (e.g., dissatisfaction with changes in weight), they were encouraged to consider the positive aspects of this day as much as possible. Participants typed their responses and were then prompted to provide a recording of themselves reading their narrative aloud. On each subsequent day, participants were provided with their written narrative and asked to create a new recording of themselves reading it aloud.

After completing their recording and surveys, participants completed a modified version of the Pros and Cons of Anorexia Scale (see Section 2.3), which was intended to amplify the negative consequences of restrictive eating. Next, they read a brief informational blurb about the negative health consequences of restrictive eating (see Supporting Information). A new blurb was presented each day, with the order counterbalanced across participants. Participants subsequently completed True/False comprehension questions.

2.2.1.2 |. Amplified Benefits.

Participants in the Amplified Benefits condition received the same narrative prompt and recording instructions as participants in the Amplified Benefits and Consequences condition. They were not asked to complete the modified PCAN or presented with blurbs about the consequences of restrictive eating.

2.2.1.3 |. Amplified Consequences.

Participants in the Amplified Consequences condition were not prompted to write or record a narrative. After completing daily surveys, they completed the modified PCAN and were presented with blurbs about the consequences of restrictive eating, followed by True/False comprehension questions, as described above.

2.2.1.4 |. Control Condition.

Participants in the control condition wrote a brief narrative nearly identical Amplified Benefits manipulation; however, their prompt instead asked them to consider a future free of concern about finances. This manipulation was intended to mimic the 0 without influencing responses to subsequent eating- or weight-related questions. As with the active conditions, participants typed their narrative and recorded themselves reading it aloud. Each day, participants created a new recording. Participants in the control condition did not complete the modified PCAN and were not provided with informational blurbs.

2.3 |. Measures

2.3.1 |. Demographics

Participants provided their gender, age, race, ethnicity, and sexual orientation.

2.3.2 |. Eating Disorder Examination Questionnaire—Q7 (EDE-Q7)

The EDE-Q is a self-report assessment of the key features of eating disorders. We used a brief version containing seven items assessing dietary restraint, shape/weight overvaluation, and body dissatisfaction (Grilo et al. 2013, 2015) which is suitable for student populations (Jenkins and Davey 2020). Items are rated on a seven-point Likert scale assessing frequency or intensity. For this study, items were anchored to the last 24 h. Internal reliability ranged from “good” to “excellent” for each subscale (α = 0.89 for dietary restraint, 0.94 for shape/weight overvaluation and body dissatisfaction).

2.3.3 |. Readiness and Motivation Questionnaire for Eating Disorders (RMQ)

The RMQ (Geller et al. 2013) is a self-report measure assessing motivation to change disordered attitudes and behaviors (i.e., restriction, cognitive symptoms, binging, and compensatory strategies). Respondents rate each item (e.g., “If you were to reduce your restriction, how much of this would be for you [versus for others]?”) on a percentage scale ranging from zero to 100. The original RMQ was designed to yield both global and symptom-specific scores across four domains of motivation: precontemplation, action, internality, and confidence. Global scores are calculated for each category by averaging scores across all endorsed eating disorder symptoms. Notably, scores are not combined or average across motivational domains, only across eating disorder symptoms (i.e., within motivation domains). For the present study, the RMQ was modified to include only items assessing dietary restriction, with one item assessing each of the aforementioned categories, and items were anchored to the last 24 h.

2.3.4 |. Pros and Cons of AN—Modified (PCAN)

The PCAN (Serpell et al. 2004) is a self-report questionnaire assessing the pros and cons of AN. Respondents rate their agreement with each statement on a five-point scale. We used the PCAN as an experimental manipulation to amplify the consequences of restrictive eating; therefore, we included only items assessing the cons of AN and replaced instances of the word “anorexia” with “restrictive eating.” Internal reliability was determined to be “excellent” for the Modified PCAN (α = 0.91).

2.4 |. Analyses

Analyses were conducted using R statistical software (R Core Team 2020). One-way ANOVAs compared experimental groups on baseline demographics, eating pathology, and motivation. The rmcorr package was used to compute repeated measures correlations (Bakdash and Marusich 2023).

Our primary analyses were conducted using the lme4 (Bates et al. 2015) and lmerTest (Kuznetsova, Brockhoff, and Christensen 2017) packages. Specifically, growth curve models examined whether experimental conditions were associated with changes over time in eating pathology and motivation. Experimental conditions were dummy-coded such that the control group served as the reference group; therefore, analyses examined the effects of each condition relative to the control condition. For each outcome, a series of unconditional models were first computed to determine the best form of change over time. Residual and predicted values plots were examined, and a chi-square difference test was used to determine improvements in model fit across iterative models. After establishing the best-fitting form of change, conditional models were estimated to evaluate differences in rates and form of change across conditions. Specifically, interactions between time (linear and quadratic terms, when applicable) and condition were modeled. Time was centered such that baseline was equal to zero. When interaction effects were significant, simple slopes of time for each outcome were modeled across each condition.

In the Section 3 and corresponding tables, we provide information regarding the linear and quadratic effects of time on each outcome, the main effects of experimental condition on each outcome (i.e., the mean difference between each condition and the control condition), and the interaction effects of time and condition on each outcome (i.e., whether the linear and quadratic effects of time were significantly different across the intervention versus control conditions). Differences in intervention effects between active conditions are summarized descriptively. For interpretation purposes, the quadratic effect for time refers to the rate of change in the slope as time increases, such that the quadratic regression coefficient is equivalent to half of the rate of change in time for each one-unit increase in time. The linear regression coefficient for time refers to the slope of the tangent line when time = 0 (i.e., baseline); if time is recentered, the regression coefficient for the linear effect changes to match the tangent line of the growth curve at the new recentered level (i.e., when time = 0). Meanwhile, the intercept is the predicted score for each outcome variable when time = 0. In the case of significant interaction effects, a comprehensive description of simple slopes is provided in Supporting Information.

3 |. Results

3.1 |. Descriptive Statistics

Average age across participants was 19.72 years. Most participants were female (92.06%), heterosexual (76.98%), and White (78.57%). Full demographics are available in Table 1. Repeated measures correlations of key variables are available in Table S1.

TABLE 1 |.

Sample demographics.

Characteristic	M (SD) or n (%)
Age	19.72 (1.84)
Gender
Female	116 (92.06%)
Male	7 (5.56%)
Non-binary/third gender	2 (1.59%)
Prefer not to say	1 (0.79%)
Race
White/Caucasian	99 (78.57%)
Asian/Pacific Islander	11 (8.73%)
Black/African American	8 (6.34%)
Multiracial/other	8 (6.34%)
Ethnicity
Non-Hispanic	93 (73.81%)
Hispanic	33 (26.19%)
Sexuality
Heterosexual/straight	97 (76.98%)
Bisexual or pansexual	23 (18.25%)
Gay or lesbian	4 (3.17%)
Prefer not to say	1 (0.79%)
Questioning/unsure	1 (0.79%)

At baseline, average scores on the EDE-Q subscales were 3.95 for dietary restriction (median = 3.67; SD = 1.56), 4.64 for body dissatisfaction (median = 5.00, SD = 1.55), and 4.83 for shape/weight overvaluation (median = 5.00, SD = 1.55). These scores were comparable to those of clinical eating disorder populations (Calugi et al. 2017). Baseline scores on the RMQ were 47.61 for precontemplation (mean = 40.00, SD = 25.69), 22.31 for action (median = 10.00, SD = 23.94), 41.37 for internality (median = 40.00, SD = 26.09), and 46.67 for confidence (median = 40.00, SD = 24.95). Experimental conditions did not significantly differ based on demographics, baseline eating pathology, or motivation (all p’s > 0.05).

Despite each experimental condition involving different manipulations, one-way ANOVAs revealed that participants generally spent a similar amount of time completing experimental procedures regardless of condition. The only exception to this occurred on day six of the study, when participants in the Amplified Consequences spent approximately six fewer minutes (i.e., 346.71 s less) completing experimental procedures when compared to the Amplified Benefits and Consequences condition; no other significant differences in duration were detected between conditions.

3.2 |. Intervention Effects

For all outcomes, visualizations of growth curve trajectories are depicted in Figure 1 (Section 3.2.1) and Figure 2 (Section 3.2.2).

FIGURE 1 |.

Growth curve plots by experimental condition: motivation. Note: For all plots, red = anticipated benefits and consequences, green = anticipated benefits, blue = anticipated consequences, purple = control condition.

FIGURE 2 |.

Growth curve plots by experimental condition: eating pathology. Note: For all plots, red = anticipated benefits and consequences, green = anticipated benefits, blue = anticipated consequences, purple = control condition.

3.2.1 |. Motivation

3.2.1.1 |. Precontemplation.

Comparison of iterative unconditional models indicated that the inclusion of a random slope for time was superior to a fixed slope alone (χ²[1] = 19.30, p < 0.001), but that a fixed quadratic term for time was not superior to a random slope alone (χ²[1] = 0.75, p = 0.387); therefore, the random intercept random slope model was retained. There was no significant condition-by-time interaction effects; how- ever, there was a main effect for condition, with participants in the amplifying benefits condition demonstrating lower average precontemplation scores than those in the control condition (B = −14.20, SE = 6.78, p = 0.038; Table 3).

TABLE 3 |.

Growth curve models: associations between experimental condition and eating disorder symptoms over a 1-week interval.

Model: dietary restraint
Fixed effects	B	SE	95% CI
Intercept	12.52***	0.84	10.87, 14.17
Day	0.37	0.31	−0.23, 0.98
Day2	−0.08	0.05	−0.17, 0.02
Condition: both	−1.07	1.21	−3.46, 1.33
Condition: benefits	−1.11	1.20	−3.48, 1.27
Condition: consequences	−0.72	1.16	−3.01, 1.58
Day × both	−1.05*	0.43	−1.89, −0.21
Day × benefits	−1.27**	0.45	−2.15, −0.38
Day × consequences	−0.71	0.41	−1.52, 0.11
Day2 × both	0.15*	0.07	0.02, 0.29
Day2 × benefits	0.22**	0.07	0.08, 0.36
Day2 × consequences	0.12	0.07	−0.005, 0.25
Random effects	Variance	SD
Intercept	17.67***	4.20
Day	0.22***	0.47
Residual	4.05	2.01
Model: shape/weight overvaluation
Fixed effects	B	SE	95% CI
Intercept	10.18***	0.57	9.06, 11.30
Day	−0.39	0.23	−0.84, 0.06
Day2	0.03	0.04	−0.04, 0.11
Condition: both	−0.96	0.83	−2.59, 0.67
Condition: benefits	−1.08	0.82	−2.70, 0.54
Condition: consequences	−0.55	0.79	−2.11, 1.01
Day × both	0.18	0.32	−0.44, 0.81
Day × benefits	−0.48	0.34	−1.14, 0.18
Day × consequences	0.14	0.31	−0.46, 0.75
Day2 × both	−0.02	0.05	−0.12, 0.08
Day2 × benefits	0.09	0.05	−0.02, 0.19
Day2 × consequences	0.03	0.05	−0.12, 0.07
Random effects	Variance	SD
Intercept	7.82***	2.80
Day	0.10***	0.32
Residual	2.30	1.52
Model: body dissatisfaction
Fixed effects	B	SE	95% CI
Intercept	9.80***	0.53	8.75, 10.84
Model: body dissatisfaction
Fixed effects	B	SE	95% CI
Day	−0.09	0.08	−0.24, 0.06
Condition: both	−1.08	0.77	−2.59, 0.43
Condition: benefits	−0.83	0.76	−2.34, 0.68
Condition: consequences	−0.50	0.73	−1.95, 0.95
Day × both	0.06	0.11	−0.16, 0.28
Day × benefits	−0.13	0.12	−0.36, 0.10
Day × consequences	0.09	0.10	−0.12, 0.30
Random effects	Variance	SD
Intercept	7.31***	2.70
Day	0.07***	0.27
Residual	1.92	1.38

Note: The control condition served as the reference group for the condition variable. “Both” refers to the amplified benefits and consequences condition.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

3.2.1.2 |. Action.

The random slope model had significantly better fit than the fixed slope model (χ²[1] = 24.44, p < 0.001), and the fixed quadratic model had significantly better fit than the random slope model (χ²[1] = 6.94, p = 0.008). Because the random quadratic model did not converge, the fixed quadratic model was retained.

There were significant interactions between condition and time (linear and quadratic) in predicting action (Table 2). See Table S2 and Figure S1 for a comprehensive description of simple slopes. At baseline (i.e., when time = 0), the linear effect of time on action was significant and negative in the control condition but non-significant for the remaining conditions. A significant positive quadratic effect of time on action was found in the Amplified Benefits and Consequences and control but not the Amplified Benefits or Amplified Consequences conditions. Specifically, the slope of the relationship between time and action became more positive over time in these two conditions (by 1.12 and 1.40 per day, respectively). Notably, the quadratic effect was stronger in the Amplified Benefits condition than in the control condition (B = −1.21, SE = 0.43, p = 0.005); there were no differences between the other two conditions and the control condition.

TABLE 2 |.

Growth curve models: associations between experimental condition and motivation over a 1-week interval.

Model: precontemplation
Fixed effects	B	SE	95% CI
Intercept	54.85***	4.69	45.60, 64.08
Day	−0.67	0.75	−2.16, 0.82
Condition: both	−10.10	6.77	−23.46, 3.26
Condition: benefits	−14.20*	6.78	−27.57, −0.84
Condition: consequences	−3.81	6.48	−16.58, 8.97
Day × both	0.44	1.07	−1.69, 2.55
Day × benefits	0.40	1.13	−1.85, 2.61
Day × consequences	−0.02	1.02	−2.03, 1.98
Random effects	Variance	SD
Intercept	481.51***	21.94
Day	6.01***	2.45
Residual	202.45	14.23
Model: action
Fixed effects	B	SE	95% CI
Intercept	27.46***	3.98	19.61, 35.32
Day	−4.46*	1.80	−8.01, −0.92
Day2	0.72*	0.29	0.15, 1.29
Condition: both	−8.61	5.77	−19.99, 2.78
Condition: benefits	−4.77	5.74	−16.08, 6.54
Condition: consequences	−7.21	5.53	−18.12, 3.70
Day × both	2.11	2.51	−2.83, 7.04
Day × benefits	7.28**	2.64	2.09, 12.48
Day × consequences	4.28	2.43	−0.50, 9.06
Day2 × both	−0.18	0.40	−0.97, 0.61
Day2 × benefits	−1.21**	0.43	−2.05, −0.36
Day2 × consequences	−0.63	0.38	−1.38, 0.13
Random effects	Variance	SD
Intercept	359.38***	18.96
Day	7.13***	2.67
Residual	141.88	11.91
Model: internality
Fixed effects	B	SE	95% CI
Intercept	45.61***	4.37	37.00, 54.24
Day	0.39	0.81	−1.21, 1.98
Condition: both	−9.79	6.31	−22.26, 2.66
Condition: benefits	−1.36	6.32	−13.85, 11.10
Condition: consequences	−3.50	6.04	−15.42, 8.41
Model: internality
Fixed effects	B	SE	95% CI
Day × both	0.04	1.15	−2.23, 2.31
Day × benefits	−0.02	1.20	−2.39, 2.35
Day × consequences	−0.20	1.09	−2.35, 1.96
Random effects	Variance	SD
Intercept	415.62***	20.39
Day	9.22***	3.04
Residual	177.64	13.33
Model: confidence
Fixed effects	B	SE	95% CI
Intercept	47.32***	4.18	39.09, 55.57
Day	−0.03	0.70	−1.41, 1.35
Condition: both	−5.08	6.04	−16.99, 6.82
Condition: benefits	−3.35	6.05	−15.29, 8.57
Condition: consequences	4.77	5.77	−6.62, 16.16
Day × both	2.44*	0.99	0.47, 4.40
Day × benefits	0.24	1.04	−1.82, 2.30
Day × consequences	−0.64	0.94	−2.51, 1.21
Random effects	Variance	SD
Intercept	365.99***	19.13
Day	4.86***	2.20
Residual	181.33	13.47

Note: The control condition served as the reference group for the condition variable. “Both” refers to the amplified benefits and consequences condition.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

3.2.1.3 |. Internality.

Inclusion of a random slope for time was superior to a fixed slope alone (χ²[1] = 47.56, p < 0.001), but the inclusion of a fixed quadratic term for time was not superior to a random slope alone (χ²[1] = 2.16, p = 0.142). Thus, the random slope model was retained. There were no significant interactions between condition and time in predicting internality, nor were there significant main effects for condition or time (Table 2).

3.2.1.4 |. Confidence.

Unconditional model comparisons indicated that inclusion of a random slope for time was superior to a fixed slope alone (χ²[1] = 21.94, p < 0.001), but that the inclusion of a fixed quadratic term for time did not result in a better model fit than a random slope alone (χ²[1] = 1.34, p = 0.247). Therefore, the random slope model was retained. There was a significant interaction between condition and time in predicting confidence (Table 2). See Table S2 and Figure S2 for a depiction of simple slopes. Specifically, the relation between time and confidence was significant and positive in the Amplified Benefits and Consequences condition but was nonsignificant in the other conditions. Notably, this relationship was stronger in the Amplified Benefits & Consequences condition than in the control condition (B = 2.44, SE = 0.99, p = 0.016); there were no differences in the relation between time and confidence between the other two conditions and the control condition.

3.2.2 |. Eating Pathology

3.2.2.1 |. Dietary Restraint.

Comparison of iterative unconditional models indicated that the inclusion of a random slope for time was superior to a fixed slope alone (χ²[1] = 37.73, p < 0.001), and that the inclusion of a fixed quadratic term for time was superior to a random slope alone (χ²[1] = 4.23, p = 0.040); the random slope random quadratic model did not converge, so the random slope fixed quadratic model was retained.

There were significant interaction effects between the condition and both the linear and quadratic terms for time (see Table 3). See Table S3 and Figure S3 for a comprehensive depiction of simple slopes. At baseline, there was a significant negative linear effect of time on dietary restraint in the Amplified Benefits and Consequences and Amplified Benefits conditions, indicating that these interventions initially led to reductions in dietary restraint. No significant linear effect of time on dietary restraint was detected for either the Amplified Consequences or control conditions. Importantly, the linear effect of time on dietary restraint was stronger in both the Amplified Benefits and Consequences (B = −1.05, SE = 0.43, p = 0.015) and the Amplified Benefits (B = −1.27, SE = 0.45, p = 0.005) conditions compared to the control condition. There were no differences between the Amplified Consequences condition and the control condition (B = −0.71, SE = 0.41, p = 0.088).

A significant quadratic effect was only found in the Amplified Benefits condition and not in the other three conditions. The quadratic effects for time were significantly stronger in the Amplified Benefits and Consequences (B = 0.15, SE = 0.07, p = 0.025) and Amplified Benefits (B = 0.22, SE = 0.07, p = 0.003) conditions compared to the control condition; however, there were no differences between the Amplified Consequences and Control conditions (B = 0.12, SE = 0.07, p = 0.059).

3.2.2.2 |. Shape/Weight Overvaluation.

Unconditional model comparisons indicated that the inclusion of a random slope for time was superior to a fixed slope alone (χ²[1] = 27.54, p < 0.001), and that the inclusion of a fixed quadratic term for time was superior to a random slope alone (χ²[1] = 4.79, p = 0.029) but that the random slope random quadratic model did not exhibit improved fit over a fixed quadratic model (χ²[1] = 0.67, p = 0.412). Thus, the random slope fixed quadratic model was retained. There were no significant condition-by-time (linear or quadratic) interactions, indicating that the study condition did not differentially impact rate or strength of change over time. Main effects for condition and time were also non-significant (Table 3).

3.2.2.3 |. Body Dissatisfaction.

For body dissatisfaction, the random slope model was the best-fitting model, as it outperformed the fixed slope model (χ²[1] = 28.02, p < 0.001) but not the fixed quadratic model (χ²[1] = 0.05, p = 0.816). There were no significant condition-by-time interactions, suggesting that study condition did not differentially impact change over time in body dissatisfaction, nor were there significant main effects for condition or time (Table 3).

4 |. Discussion

Drawing from a behavioral economic framework, we evaluated the impact of amplifying (1) the benefits of reducing restrictive eating, (2) the consequences of maintaining restrictive eating, and (3) both, on motivation to reduce restrictive eating. Because motivation does not necessarily correlate with changes in eating disorder symptoms (Waller 2012), we also evaluated effects on dietary restraint, shape and weight overvaluation, and body dissatisfaction.

4.1 |. Motivation

When evaluating the effects of experimental condition on motivation, we considered precontemplation, action, internality, and confidence independently. We did not detect the effects of any experimental condition on internality, nor did we detect meaningful condition-by-time interactions on precontemplation. However, notable differences emerged between conditions with regards to action, or the extent to which participants were actively working to reduce restriction. Although action scores increased most quickly when both the benefits of reducing restriction and the consequences of maintenance were amplified, pre- to post-intervention changes did not reach statistical significance. Nevertheless, because the relationship between time and action became significantly more positive over time in the Amplified Benefits and Consequences condition, it appears that amplifying both the benefits of reducing restriction and consequences of maintenance may have a cumulative, positive effect on attempts to reduce dietary restriction. Over a longer intervention period, these changes may have reached statistical and clinical significance.

Participants’ confidence in their ability to reduce restriction significantly increased in the Amplified Benefits and Consequences condition. Although we expected to detect the strongest effects in this condition, it is notable that no significant effects were detected in conditions which amplified the benefits of reducing restriction or consequences of maintenance alone. This highlights the importance of attending to both the cost of current behaviors and the reward associated with behavior change; when both are made sufficiently salient, it may be easier to appreciate the value of adopting a new behavior. This approach is consistent with traditional motivational techniques, which facilitate the exploration of ambivalence by weighing pros and cons of change (Miller and Rose 2015; Prochaska et al. 1994). We extended this concept by adding an element of repeatedly and vividly imagining the long-term outcome of addressing disordered thoughts and behaviors, making the pros of change more salient and tangible.

It is also important to note that the Amplified Benefits and Consequences condition involved the greatest number of active intervention components when compared to the other experimental conditions. It is therefore possible that participants in this condition simply benefited from a higher “dose” of treatment. However, it is notable that minimal differences were detected in the amount of time participants spent completing experimental procedures each day across conditions. Nevertheless, future research may benefit from continuing to explore potential differences in the effectiveness of motivational interventions based on their intensity (or the “dose” of treatment provided).

4.2 |. Eating Pathology

Although we evaluated the effects of experimental conditions on three dimensions of eating pathology, significant effects were detected only for dietary restraint. One possibility for the specificity of effects on dietary restraint is that the present intervention centered on the importance of changing behavior. Whereas dietary restraint is primarily a behavioral construct, shape/weight overvaluation and body dissatisfaction typically manifest cognitively. Because we did not directly target thoughts or beliefs, it is possible that individuals who benefited from this intervention were able to reduce restriction while still experiencing maladaptive cognitions about weight and shape.

Initial effects on dietary restraint were most pronounced in the amplified benefits condition, followed by the amplified benefits and consequences condition; no significant effects were detected in the amplified consequences condition. One possible explanation for this pattern of findings draws from evidence suggesting that caloric deprivation can lead to alterations in reward sensitivity (Avena, Murray, and Gold 2013; Carr 2002; Holsen et al. 2012). If individuals engaging in restrictive eating do indeed experience a heightened response to reward, perhaps due to increased sensitivity of the dopaminergic system (Cassidy and Tong 2017), this could partially explain why participants did not respond to experimental conditions which only amplified the consequences (i.e., punishment) associated with disordered behaviors, whereas conditions that amplified the benefits of reducing restriction (i.e., reward) were more successful. This hypothesis is speculative, however, as some evidence suggests reward sensitivity may be highly state-dependent, rather than a stable trait of individuals with restrictive eating disorders (Neuser et al. 2020). It is also notable that the effects of restriction on reward sensitivity are not universal; individuals with a history of restrictive eating disorders appear to demonstrate different patterns of reward responsivity than healthy individuals engaging in restriction (Kaye et al. 2020). Because the preponderance of evidence in this domain centers on response to food-related rewards, caution is warranted when attempting to generalize the results to rewarding experiences more broadly.

It is also possible that amplifying the benefits of reducing restriction positively influenced participants’ hope for the future. Individuals who consistently engage in dietary restriction, such as chronic dieters or those with restrictive eating disorders, may struggle to experience hope for a future where they are no longer preoccupied with their caloric intake (Fox 2018; Malson et al. 2011). By prompting participants to vividly imagine their futures as unencumbered by thoughts about food or weight, the narrative component of the intervention may have provided an opportunity for participants to visualize their own versions of “recovery,” thereby enhancing motivation to begin reducing restriction.

Although participants in the Amplified Benefits and Amplified Benefits and Consequences conditions demonstrated initial reductions in dietary restraint, this response was not sustained, perhaps because no behavioral changes were prescribed that would lead to increased caloric intake. Whereas increases in motivation may have prompted initial self-directed efforts to reduce dietary restraint, these efforts may have proven unsustainable. Interventions designed to enhance motivation may be best suited as a first stage in treatment and should be followed by interventions directly targeting restrictive eating and maladaptive cognitions.

It is also possible that the effects of the amplified benefits and amplified benefits and consequences conditions diminished over time due to habituation. Whereas the amplified consequences manipulation involved presenting participants with new information each day, the amplified benefits manipulation required participants to re-read the same narrative daily. The repetitive nature of this manipulation may not have provided sufficient novelty to produce an emotional response, particularly after several days of completing identical experimental procedures. If our amplified benefits manipulation had introduced more novelty, it is possible that intervention effects may have been sustained over a longer duration.

4.3 |. Limitations

Eating disorders affect nearly all demographic groups; (Calzo et al. 2017; Cheng et al. 2019; Raevuori, Keski-Rahkonen, and Hoek 2014), however, our sample was predominantly female, White, and heterosexual. Findings may have differed in a more diverse sample. Relatedly, all participants were undergraduate students. The college environment is characterized by a distinct social context and set of stressors (Hurst, Baranik, and Daniel 2013; Kroshus, Hawrilenko, and Browning 2021), which may uniquely influence how eating disorder symptoms manifest and how well certain interventions perform. Our use of this sample may limit the generalizability of findings.

Our inclusion criteria required the presence of restrictive eating, as well as functional impairment related to eating pathology. To screen participants for inclusion, we used the DRS to assess for the presence of maladaptive restrictive eating, and two items drawn from the EDE-Q to assess preoccupation with eating and weight (i.e., functional impairment). Because these measures were anchored to the last month, and the duration of our study was only 1week, it is possible that we captured some participants whose symptoms were not sufficiently frequent or intense to detect changes over the study period. Additionally, we dichotomized functional impairment as either present or absent. Our aim was to exclude individuals whose restrictive eating was clearly not causing impairment in other areas of functioning, while still including individuals experiencing a level of impairment consistent with a nonclinical population. Taken together, it is possible that our screening approach was not sufficiently conservative to ensure that all included participants were engaging in clinically significant restrictive eating. We reason this is unlikely to be the case, however, as we found that baseline EDE-Q scores within our sample were comparable to those found in clinical eating disorder populations.

Relatedly, because we did not assess for DSM-5 eating disorders, we are unable to draw conclusions about the effectiveness of the present intervention for individuals whose symptoms meet the threshold of a diagnosable eating disorder, such as Anorexia Nervosa. Our findings are also most applicable to individuals with restrictive eating, and it is unclear whether the present findings would generalize to populations experiencing binge eating, or compensatory behaviors beyond restriction. It remains possible that stronger effects may have been detected in a clinical sample, whereas there may have been a “ceiling” on improvement in our student sample. Conversely, it is also possible that we may have detected weaker effects in a more severe sample, as newer-onset or subthreshold eating pathology may be more malleable than its chronic or full-threshold counterpart.

We administered self-report measures, which are susceptible to certain limitations, including recall bias (Schwarz 1994). We also modified the EDE-Q7 and RMQ by anchoring all items to the last 24 h, though both scales continued to demonstrate adequate psychometric properties despite these modifications. In addition to modifying the timeframe assessed by the RMQ, we administered a version of the measure which only assessed a single symptom: dietary restriction. Because RMQ scores are not intended to be averaged or combined across motivational domains (rather, they are averaged only across symptoms), scores for each domain of motivation were therefore based on single items. Although the developers of the RMQ recommend that the scale only be used to assess the symptoms that are relevant for a particular individual, we are not aware of any other studies which constrained the RMQ to only assess dietary restriction. Additional research is needed to establish the validity of this approach.

In this study, we relied on participants to generate narratives about their imagined futures; however, we were unable to shape responses if they lacked certain components. Nevertheless, most respondents followed instructions satisfactorily, and a post hoc manipulation check revealed that our experimental manipulations appeared to have the intended effects. Among participants asked to write a narrative describing a future free from concern about eating or weight (i.e., participants in the amplified benefits and consequences and amplified benefits conditions), 98% provided a detailed account of a day in the future spanning from morning to night. Although only about one-third of narratives (31%) included an explicit mention of recovery-oriented thoughts and behaviors (e.g., “I am able to eat without thinking about the caloric intake I am consuming”), nearly all narratives (95%) referenced engagement in valued activities unrelated to eating or weight (e.g., spending time with friends and family, building a career). Only one participant explicitly mentioned weight loss, suggesting that they may have attributed their reduced concern about eating and weight to achieving a desired body size. A more robust version of this intervention may involve interaction with an experimenter to ensure comprehension and strengthen adherence to study protocols.

Our experimental design allowed us to examine the impact of amplifying the anticipated benefits of reducing restriction, the consequences of maintenance, and their combined effects on eating pathology and motivation. However, whereas our control condition included a narrative component which resembled the amplified benefits portion of two of our experimental conditions (i.e., asking participants to imagine a future free from concern about finances), it did not directly mimic the information presented in the amplified consequences manipulation (e.g., providing participants with statistics about the negative consequences of financial indiscretion). Therefore, we are unable to draw conclusions about the necessity of considering the negative consequences of restrictive eating specifically; it is possible that simply being presented with negatively valenced information about the consequences of current behavior may be sufficient to influence motivation and change behavior.

Our assessments of dietary restraint did not capture changes in eating behavior. Instead, the EDE-Q asks respondents to remark upon the extent to which they have been consciously limiting the amount they eat. Another approach would be to monitor food consumption in a lab environment, which would ensure that self-reported reductions in restriction corresponded with increases in caloric intake. Alternatively, future research may wish to leverage ambulatory assessment methods (Smith et al. 2019) to estimate caloric intake in a naturalistic setting.

Although we monitored changes in participants’ motivation and eating pathology over a 7-day period, we did not administer subsequent follow-up measures. Therefore, we were unable to establish whether changes which occurred during the study were maintained over time. Given the relatively high likelihood of relapse in restrictive eating disorders (Khalsa et al. 2017), it is important to further clarify the extent to which the effects of motivational interventions are sustained over time. We look forward to future research which continues to evaluate the long-term outcome of brief motivational interventions for restrictive eating.

Lastly, it is important to note that dietary restriction is not inherently a maladaptive behavior. Individuals with certain medical concerns, such as a Type II diabetes, may be advised by their physicians to engage in caloric restriction to manage the condition (Magkos, Hjorth, and Astrup 2020). Additionally, some individuals may choose to adopt a more restrictive diet due to personal preferences, or a desire to lose weight which does not necessarily occur in the context of eating pathology (i.e., “normative” dieting). Although our screening procedures were intended to identify individuals engaging in maladaptive dietary restriction, it remains possible that this was not the case for all included participants. However, the high baseline EDE-Q scores detected in our sample suggest that clinically significant eating pathology was present for a majority of participants, including symptoms unrelated to dietary restriction (i.e., body dissatisfaction, shape/weight overvaluation).

4.4 |. Future Directions and Clinical Implications

Existing evidence surrounding the effectiveness of motivational interventions for restrictive eating is mixed (Knowles, Anokhina, and Serpell 2013). Our findings lend tentative support to a novel approach which builds upon aspects of traditional motivational interventions (e.g., weighing the pros and cons of change) while incorporating additional techniques intended to enhance the benefits of change and the negative consequences of maintaining disordered behaviors. Although we were encouraged to find that our experimental manipulations yielded increased confidence to reduce restriction, as well as short-term reductions in restrictive eating, their broad impact on motivation and eating pathology was modest. These findings align with prior evidence suggesting that interventions targeting motivation are unlikely to be more effective than established treatments for eating pathology, such as cognitive behavioral therapy. However, when considering the conditions under which the study was conducted (e.g., entirely online format, no clinician interaction, low intensity, brief duration), it is notable that our intervention nevertheless yielded significant changes in certain key outcomes. Although the approach adopted in the present study is not intended to replace mainstream interventions for eating disorders, our findings suggest there may be benefits to leveraging novel motivational techniques to reduce engagement in problematic restrictive eating. We look forward to future research in clinical populations which continues to explore how to most effectively and efficiently enhance motivation to reduce restrictive eating.

Relatedly, balancing effectiveness and efficiency is an important clinical consideration. Our choice of a 7-day study period was somewhat arbitrary; motivational interventions often require at least four weekly sessions (Macdonald et al. 2012), and enhanced Cognitive Behavioral Therapy for eating disorders can include up to 100 sessions (Atwood and Friedman 2020). This may raise questions about whether we could reasonably expect to detect changes in motivation and eating pathology over such a brief period. Whereas short-term motivational interventions are relatively rare in the eating disorders field, they are commonly used to address other clinical concerns, including substance use (DiClemente et al. 2017). In this domain, brief motivation-focused interventions have been shown to effectively reduce substance use, and some evidence even suggests that briefer motivational interventions may be more effective than longer ones (Lindson-Hawley, Thompson, and Begh 2015). Although many of the existing brief motivation-focused interventions for eating pathology comprise a small number of weekly sessions, it is notable that even single-session interventions have shown promise for improving motivation when compared to non-motivational, lower-intensity approaches (Denison-Day et al. 2018). Improvements in motivation resulting from these brief interventions can be detected immediately postintervention (Vella-Zarb et al. 2015), as well as over longer follow-up intervals (Geller, Brown, and Srikameswaran 2011). Future studies may wish to continue exploring the viability of brief interventions as an efficient and scalable alternative to longer or more resource-intensive interventions (Schleider, Smith, and Ahuvia 2023).

Online interventions also play a crucial role in increasing access to scalable eating disorder treatment (Aardoom, Dingemans, and Van Furth 2016), and it is likely that demand for these interventions will continue to increase (Linardon et al. 2020). Our findings suggest that brief online interventions, including ones which do not necessarily rely on provider interaction, may have the potential to increase motivation to reduce engagement in potentially harmful eating practices, including maladaptive restricive eating. Future research may clarify whether existing online interventions for eating disorders could be strengthened through the addition of novel motivational techniques, including those assessed within this study (e.g., vividly imagining recovery).

Lastly, our findings may inform how clinicians and other supports choose to approach conversations about reducing restrictive eating. Publicly available psychoeducational materials about eating disorders, such as those found on the National Eating Disorders Association website, often emphasize the consequences of maintaining disordered behaviors (National Eating Disorders Association n.d.). Individuals supporting friends or family members with disordered eating, as well as healthcare providers, may also be tempted express concern about the dangers of dietary restriction (Akey and Rintamaki 2014). Although this approach is well-intentioned, our results suggest that it is unlikely to inspire behavioral change, unless it is paired with information about the benefits of reducing disordered eating. This information may be particularly impactful when it is shared by trusted others, such as peer mentors (Venturo-Conerly et al. 2020).

4.5 |. Conclusions

Within a 7-day online intervention, amplifying the benefits of reducing dietary restriction and consequences of maintaining disordered behaviors led to initial reductions in dietary restraint and sustained improvements in confidence to reduce restrictive eating; however, overall effects on motivation were modest and effects on and eating pathology were not sustained. Although motivational interventions are not intended to replace mainstream approaches to targeting eating pathology, our findings suggest that brief, online interventions which incorporate novel motivational techniques are worth further investigation. Ultimately, we hope this line of inquiry clarifies the ways in which interventions can be further adapted and implemented to effectively address eating disorders.

Data Availability Statement

The data that support the findings of this study are available upon reasonable request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Summary.

• Eating disorder symptoms are generally considered challenging to treat.

• Individuals with eating disorder symptoms, particularly restrictive eating, often experience low motivation to engage in treatment and change their behavior.

• We found that amplifying the potential benefits of reducing dietary restriction, as well as the negative consequences of continuing to engage in disordered behaviors, modestly improved motivation to reduce engagement in dietary restriction.