Content area
Purpose
The goal of this study was to assess changes in eating self-efficacy after participating in a brief psychoeducational group intervention, grounded in the cognitive-behavioral model, delivered by dieticians in community-based health facilities.
Design/methodology/approach
The study was conducted using a quasi-experimental, pre-post design. A total of 110 program participants took part in the study. They were asked to complete the Eating Self-Efficacy Scale before the start of the intervention, at the end of the intervention, and three months after the intervention ended. Data were analyzed using the Linear Mixed Model.
Findings
Participants’ personal sense of control over their eating behaviors significantly increased after they completed the program and continued to increase up to the three-month follow-up. The effect of the intervention remained significant after controlling for differences in age and whether participants had access to other forms of individual support or completed the follow-up during the COVID-19 general lockdown.
Practical implications
By promoting participants’ sense of eating self-efficacy, this intervention could lead to positive dietary changes, which in turn could promote better health and healthy aging.
Social implications
This community intervention is readily accessible and represents a cost-effective approach to promote healthy eating, reducing the risk of chronic disease and the need for medical care, thereby cutting costs for the healthcare system.
Originality/value
(1) This study addresses a gap in the scientific literature as there was limited published research to date that investigated this intervention. (2) The three-month follow-up made it possible to evaluate whether changes in eating self-efficacy were maintained over time. (3) Potential confounding variables, including age, having access to other forms of individual support and the COVID-19 general lockdown, were taken into account.
Introduction
Adherence to dietary guidelines involves eating habits that conform to recommended nutrients or food intake. Although guidelines may vary, this typically includes a higher intake of fruits, vegetables, nuts, legumes, and whole grains, and a lower intake of sodium and fat (World Health Organization, 2020). Compliance with these guidelines significantly lowers the risk of cardiovascular disease, cancer, diabetes, heart failure, and mortality due to chronic diseases (Hu et al., 2019; Russell et al., 2013; Struijk et al., 2014; Yu et al., 2014). In addition, better adherence to these guidelines appears to be related to healthy aging (Gopinath et al., 2016). In turn, it can reduce the need for medical care, thereby cutting costs for the healthcare systems (Anekwe and Rahkovsky, 2013; Collins et al., 2013; Springmann et al., 2021).
Nevertheless, statistics highlight that adherence to dietary recommendations remains relatively low to moderate, although this can vary based on certain factors, including gender, education level, and socioeconomic status (Gopinath et al., 2016; Leme et al., 2021). A similar trend was observed within the Canadian population (Brassard et al., 2018; Kolahdooz et al., 2018). Furthermore, a recent systematic review of 23 longitudinal studies revealed a decrease in adherence to healthy diets after the onset of the COVID-19 pandemic (González-Monroy et al., 2021). These findings suggest that developing healthy eating habits may require more than simply understanding nutritional information (El Ansari et al., 2015), and emphasize the importance of improving our comprehension of the factors that may influence dietary behaviors.
Theoretical models from the field of psychology, particularly the cognitive-behavioral theoretical framework, can shed light on the factors that influence eating habits.
Theoretical framework
The cognitive-behavioral approach does not refer to a single theory, but rather a set of theories that were introduced in the 1960s. They share a number of theoretical assumptions, the main one being the constant interaction between behaviors, emotions, and cognitions as well as their influence on each other (Dozois et al., 2019). Indeed, behaviors, including eating behaviors, are influenced by emotional states and ways of thinking (Aoun et al., 2019; Wehling and Lusher, 2019). For instance, eating may represent a way to regulate negative emotions, such as depression, and may even serve as an emotional support tool (Evers et al., 2010; Goodspeed Grant, 2008). From a cultural perspective, eating is also strongly connected to particular events, ceremonies, and traditions, and can elicit positive feelings and memories associated with food (Evers et al., 2010; Goodspeed Grant, 2008; Konttinen et al., 2019; Ogden, 2009; Wehling and Lusher, 2019). Attitudes and thought patterns can also play an important role in the development of eating habits (Ogden, 2009). For example, studies have shown that maladaptive cognitions or core beliefs related to themes of failure to achieve, social isolation, and abandonment are significantly related to eating disorders, such as binge eating and bulimia nervosa (Jones et al., 2005; Waller et al., 2002). In addition, specific thought patterns concerning loss of control and dietary restraint can be established as predictors of binge eating episodes (Legenbauer et al., 2018).
Within the cognitive-behavioral approach, Bandura’s Social Cognitive Theory can shed light on how health-related behaviors can be changed and maintained over time (Abraham et al., 2016; Abusabha and Achterberg, 1997). In addition to cognitions and emotions, Bandura also considers the impact of social influences on behaviors (Bandura, 2001; Feist et al., 2018). As it relates to eating behaviors, these may be affected by the presence of other individuals and by social norms. A study by Higgs et al. (2022) revealed that people tend to eat more with family and friends (social facilitation of eating), while they tend to eat less in the presence of strangers (social inhibition of eating). Other studies have found that food selection and consumption are highly influenced by modeling, meaning that what we eat and how much is eaten is shaped by social norms. These norms are established based on other people’s eating behaviors, and adherence to these norms may occur based on the perceived importance of engaging in appropriate behaviors and the need for social acceptance (Cruwys et al., 2015; Higgs, 2015; Higgs and Thomas, 2016).
At the heart of Bandura’s theoretical model, the notion of self-efficacy is conceived as playing a key role in the modification and long-term maintenance of new behavioral patterns (Bandura, 1978, 2004). This concept refers to the belief in one’s abilities to execute and control certain behaviors in order to achieve a task or a goal (Bandura, 1978, 2001). In turn, self-efficacy is believed to influence whether one chooses to initiate health behavior changes, how much effort is invested in changing one’s behaviors, and the extent to which one perseveres when encountering obstacles (Buckworth, 2017). Past research has revealed that self-efficacy is a robust predictor of eating behaviors and weight management (Bektas et al., 2021; Berman, 2006; Doerksen and McAuley, 2014; Kitsantas, 2000; Sniehotta et al., 2005; Vuillier et al., 2021). Bandura’s Social-Cognitive Theory also suggests that self-efficacy is malleable, meaning that it can be targeted and improved. The most influential factors to enhance personal self-efficacy involve mastery experiences or successful performance (Bandura, 1997, 2008; Buckworth, 2017; Kleppang et al., 2023; Warner et al., 2018). Other strategies to promote self-efficacy involve social modeling or vicarious learning, social persuasion or encouragement, as well as learning to manage certain physical and emotional states that could interfere with the behavior (Bandura, 1997, 2008; Buckworth, 2017).
Interventions to promote healthy eating
Bandura’s theoretical underpinnings have greatly contributed to the cognitive-behavioral approach to psychotherapy, which is a proven intervention method in the clinical field. Cognitive Behavioral Therapy (CBT) has been successfully used to change eating behaviors, manage weight, and treat eating disorders (Alimoradi et al., 2016; Corstorphine, 2006; Munsch et al., 2007; Wade et al., 2017). However, the cost associated with CBT or other traditional self-efficacy interventions, and the limited access to specialized services by trained psychologists or psychiatrists, imply that these types of interventions are not readily available to everyone (Bouwman et al., 2020; Grilo et al., 2014; Jakicic et al., 2012; von Ranson et al., 2010; Wilson et al., 2000). This underscores the importance of developing cost-effective and accessible interventions to promote healthy eating (Bouwman et al., 2020). An alternative approach may involve the use of a brief intervention offered in group settings by frontline healthcare workers, such as dieticians.
Few studies have investigated the effect of a brief group intervention on eating self-efficacy and eating behaviors. Results demonstrate that interventions ranging from two to 12-week sessions, that include a psychoeducational component, were effective in reducing problematic eating behaviors, increasing healthy food intake, and improving eating self-efficacy (Ashton et al., 2009; Jones and Clausen, 2013; Smith et al., 2020; von Ranson et al., 2010). In addition, interventions that offer psychoeducation as well as personalized instructions or feedback were more effective than education programs alone (Whatnall et al., 2018). Finally, brief interventions appeared to demonstrate similar results whether delivered in person or via online chat rooms (Gollings and Paxton, 2006). However, in most of these studies, the intervention was still conducted by licensed therapists.
The present study
Given the low level of adherence to dietary guidelines among the Canadian population, and the limited availability of specialists offering CBT-based interventions, there is a growing issue concerning the supply and demand for such support services. In response to this need, Craving Change® was developed in Canada by a clinical psychologist and a dietician. It is a brief, structured, licensed intervention rooted in the cognitive-behavioral model. Unlike conventional nutritional counseling, it takes an innovative approach by addressing psychological factors, such as thoughts and emotions, as well as social factors that influence eating habits. One of the program’s main objectives is to promote a sense of control over one’s eating behaviors (Cannon and Shah, 2024a). This program is delivered through community-based health facilities by primary healthcare providers, including dieticians, making it more accessible compared to other interventions conducted by licensed therapists in a clinical setting. It is currently one of the most widely used programs in Canada for promoting healthy eating and is rapidly gaining recognition internationally (Cannon and Shah, 2024b). Despite its widespread use, there is limited published research to date investigating the effectiveness of this intervention, except for one study by von Ranson et al. (2010) that explored a previous iteration of the program.
The objective of the study was to evaluate whether participation in the Craving Change group program was associated with an increase in eating self-efficacy and whether this increase was sustained over time.
Method
The study was conducted using a quasi-experimental, pre-post design, with a follow-up three months after the intervention. Due to logistical constraints, access to a control group was not possible, thus preventing the use of a randomized controlled trial. Despite their limitations in establishing causality, quasi-experimental designs can be useful in estimating the effect of an intervention, providing preliminary evidence of its potential benefits (Harris et al., 2006; Miller et al., 2020).
Setting and participants
The study took place in a large city (>800,000 residents) located in central Canada. It was conducted in partnership with all six community-based health facilities, located in different parts of the city, where the Craving Change group intervention was led by registered dieticians (program facilitators). These facilitators have been trained and they were licensed to lead Craving Change individual and group interventions. For the purposes of this study, only the group intervention was included in the evaluation. Registration for the group program was open to anyone in the community aged 18 years or older, who was seeking support in managing their eating habits, but who did not meet the diagnostic criteria for an eating disorder (the latter were referred to individual counseling services). Participation in the program was mandatory for people undergoing bariatric surgeries.
The study aimed to include all individuals aged 18 years or older enrolled in the Craving Change group program at one of the six community-based health facilities between 2018 and 2020, which amounted to a pool of 154 potential participants. All 154 individuals were invited to participate in the study. Of those, 110 agreed to participate.
Intervention
Craving Change® is a licensed, Canadian program that was developed in 2008 by clinical psychologist Colleen Cannon and dietitian Wendy Shah. It is intended to be delivered by primary healthcare providers, such as dietitians. Professionals who learn how to administer this program are trained to incorporate cognitive-behavioral strategies into their interventions. Gaining international recognition, it is now one of the most widely used programs in Canada for managing weight and chronic disease as well as promoting healthy eating (Cannon and Shah, 2024b). Despite its widespread use, more research is needed to examine the effects of this program with participants.
Inspired by the cognitive-behavioral model, the program is based on the premise that people often eat in response to certain thoughts or emotions (e.g. all or nothing thinking, stress, boredom). Therefore, unlike traditional nutritional counseling, the program does not focus on what people should or should not eat, but rather on the reasons that motivate them to engage in certain eating behaviors. It uses the analogy of an iceberg to explain the reasons why people eat. The visible part of the iceberg is the behavior, while the reasons underlying eating behaviors, such as negative thoughts and emotions, lay hidden under the surface (Cannon and Shah, 2024a). Thus, program participants are encouraged to explore and identify their personal eating triggers. In turn, they are taught step-by-step techniques and strategies to help them control food cravings and to self-regulate their eating behaviors. The program also emphasizes the importance of living a balanced lifestyle by learning to set realistic goals, while occasionally indulging in certain things that provide pleasure. It encourages participants to make healthy food choices 80% of the time instead of striving for perfection (Cannon and Shah, 2024b).
The program offers 6 h of facilitated curriculum, structured into four modules. The first module—Why it’s difficult to change eating behaviors—examines how eating behaviors are tied to factors that are often beyond awareness, including environmental, biological, and behavioral conditioning factors. The second module—What needs to be changed—is aimed at helping participants identify their respective eating triggers and distinguish between different types of hunger, namely stomach hunger (the physical need for food), mouth hunger (cravings related to the sensory pleasure of certain foods), and heart hunger (eating in response to emotions). The third module—How to change—teaches various cognitive-behavioral techniques and strategies to help participants monitor and manage their personal triggers in order to regulate their eating behaviors. Finally, the fourth module—Keep the change—focuses on using research-based techniques to promote maintenance of the positive changes made in the program and to prevent relapse. Each session involves a combination of instructions based on a psycho-educational approach, group discussions, and assignments to be completed in the provided workbook (Cannon and Shah, 2024b).
The overarching goal of the program is to stimulate participants’ sense of control or agency, so they can learn to self-regulate their eating behaviors, while encouraging self-compassion.
Measure
Research data were collected with the use of a questionnaire. The first section contained questions measuring demographic variables, including, age, gender, ethnic identity, annual household income, and highest level of education achieved. One question also asked whether they had access to other forms of individual support to help them with their eating habits. The second section contained the Eating Self-Efficacy Scale (ESES) (Glynn and Ruderman, 1986). This measure was chosen because it addresses one of the intervention’s core objectives, namely, to promote a greater sense of control or agency over one’s eating behaviors. It contains two subscales with 15 items measuring whether participants believe they are able to control their consumption of food in response to negative emotions or affect (e.g. feeling tense or irritable) and 10 items measuring eating self-efficacy in social circumstances where consuming food is considered socially acceptable (e.g. with friends, around the holiday time or after an argument). Each item was answered on a Likert scale that varied from 1 (No difficulty controlling eating) to 7 (Most difficulty controlling eating). Responses to each item were reversed for ease of interpretation, with higher scores indicating a greater sense of eating self-efficacy. The ESES has been proven to have a seven-week test-retest reliability of 0.70 and has been demonstrated to be a valid scale (Glynn and Ruderman, 1986). Item scores for the two ESES subscales were aggregated by computing mean values to create two factors, namely eating self-efficacy in response to negative emotions or affect (NA) and eating self-efficacy in circumstances where consuming food is considered socially acceptable (SAC). NA and SAC aggregated scores were computed for each of the three measurement times, namely pre-intervention, post-intervention, and 3-month follow up.
Procedure
Ethics approval was obtained from the Research Ethics Boards of the Université de Saint-Boniface and of the Regional Health Authority. Group interventions involved three face-to-face group meetings of two hours conducted over a three-week period. There was a $10 registration fee to cover the cost of the workbook. However, copies were made available free of charge to low-income participants upon request. Participants were asked to complete the ESES at three different points in time. Program facilitators asked participants to complete the ESES at the beginning of the first Craving Change session (pre-intervention) and again at the end of the last session (post-intervention). Three months after completing the Craving Change program, participants received an email from a member of the research team, inviting them to complete the questionnaire one last time through SurveyMonkey (3-month follow-up). The first page of the online survey contained the consent form, in which the objectives of the study were outlined as well as the nature of their participation in this study. Instructions on the form indicated that by clicking a button at the bottom of the page, they agreed (1) to take part in the study, and (2) for their program facilitators to disclose to the research team a copy of their first two completed questionnaires. Participants who completed the follow-up online survey received a $10 gift card.
Data analysis
Data were analyzed with the use of SPSS version 24.0 (IMB Corp, 2016). The dataset was first inspected for missing values on the two outcome measures, namely NA and SAC. Figure 1 shows the flow of participant attrition and missing values. Out of the 110 program participants who initially agreed to take part in the study, 72 provided complete data for all three measurement times, while 29 participants provided data for two measurement times, and 9 participants provided data for only one measurement time. No follow-up was made to determine why they failed to complete the survey during the post-intervention or the 3-month follow-up. Descriptive analyses of sociodemographic variables were conducted for the whole sample (N = 110) as well as for the 38 participants with missing values and the remaining 72 participants with no missing values. These analyses included frequencies for categorical variables (gender, language, education, employment, annual income, and ethnicity) as well as means and standard deviations for the continuous variable (age). Participants with missing values were compared to those with no missing values in terms of their age (independent t-test) and in terms of gender, language, education, employment status, and annual income (chi-square tests). Descriptive statistics were also obtained on the outcome variables, namely NA and SAC scores, for each of the three measurement times. Mean scores and standard deviations were computed using all available data points for the whole sample.
Given that 38 participants had missing values, the data were analyzed using the Linear Mixed Model (LMM), which is an extension of the linear regression model. LMM is becoming increasingly common in psychology and in health research (Meteyard and Davies, 2020; Schober and Vetter, 2021). One of the advantages of LMM, compared to a repeated-measure (or within-subject) ANOVA, is that it allows for the inclusion of all available data points, including participants with missing values (Bell and Rabe, 2020; Krueger and Tian, 2004; Xi et al., 2018). This ensures greater statistical power while avoiding bias introduced by the loss of participants, as LMM uses all observations to predict participants’ trajectories over time by incorporating random effects to account for variance across participants (Walker et al., 2019).
The construction of a LMM entails several decisions, beginning with the inclusion of random effect terms into the model, followed by fixed effect terms (Meteyard and Davies, 2020). To mitigate the loss in statistical power due to the relatively small sample size, the model was gradually developed by progressing from most parsimonious to more complex model structures, while assessing whether the inclusion of each additional term significantly improved model fit, as recommended by Matuschek et al. (2017). A Likelihood Ratio Test was performed on each successive nested model to compare relative fit, computing the difference in restricted maximum likelihood between models. This difference follows a chi-square distribution, with the difference in the number of parameters in both models equating to degrees of freedom (Matuschek et al., 2017; Meteyard and Davies, 2020). The significance level was set at 0.01. The process of building and comparing models is outlined in Table 1.
The simplest model (Model 1) included a random intercept for participants to account for the non-independence in participants’ values across the three measurement times, given the repeated-measure design. A random slope was then added into the model (Model 2) to account for variance in slope between outcome measures and Time across participants. In this model (and all subsequent models) Time was specified as a repeated factor, using Compound symmetry: Correlation metric for the covariance structure. This covariance structure assumes homogenous variance across repeated measures and homogenous covariance between measurement times, aligning with the characteristics of the dataset. However, Model 2 failed to converge after the inclusion of a random slope, and it was consequently removed from the model. Fixed effect terms were then added, starting with the predictor, namely Time (Model 3). The inclusion of Time significantly improved the model fit for both outcome measures. Covariates were then added as fixed effects. The first covariate was Age given there was a significant difference in the age of participants with missing values when compared to those with no missing values (Model 4). This inclusion significantly improved model fit for both outcome measures. Since 23 participants completed the three-month follow-up during the COVID-19 general lockdown in the spring of 2020, this factor was considered as another potential covariate. However, the inclusion of the COVID-19 factor into the model (Model 5) did not significantly improve model fit for NA or for SAC and was thus removed. Another consideration was that 28 participants reported having access to other forms of individual support during the three-month follow-up. While including Individual support in the model (Model 6) improved model fit, it reduced sample size to 82 as 28 participants had missing data on that variable. Consequently, it was decided to remove Individual support from the final model to avoid loss in statistical power. Nevertheless, it is important to note that when testing Model 6 on the 82 participants with available data on Individual support, a significant difference in NA scores was observed, F (1, 79.15) = 5.95, p = 0.017, with those who had access to individual support having a higher general mean across the three measurement times (4.67) than those who did not (3.97). Despite this difference, the effect of Time on NA remained significant even when controlling for Age and Individual support, F (2, 149.56) = 26.58, p < 0.001. With respect to SAC scores, the effect of Individual support was not significant, F (1, 79.11) = 1.32, p = 0.254.
The final model (Model 4) for both outcome variables included a random intercept for participants with the predictor (Time) and covariate (Age) identified as fixed factors. Complete syntax using the SPSS MIXED command for this final model is provided as a supplementary file (Appendix). The model was performed on a total of 108 participants (as 2 participants did not report their age), contributing to 278 data points for each outcome measures. Since this model was performed twice, once for each outcome variable, the significance level was set at 0.025, using a Bonferroni correction. A pseudo-R2 (or estimated variance explained) for fixed effects was calculated using the formula provided by Snijders and Bosker (2012). To obtain an estimate of the effect size of Time, in relation to variance explained, f2 was also computed based on Cohen’s guidelines (Cohen, 1992).
The assumptions of LMM for this final model were assessed, which include linearity between outcome variables and the predictor (examined through scatterplots of NA and SAC against Time), constant variance in residuals (assessed via scatterplots of residuals against Time), independence of errors (evaluated using scatterplots of residuals against predicted values), and normality of residuals (examined with histograms of residuals). All assumptions were deemed satisfactory.
Results
Descriptive statistics
The demographic characteristics of the sample are summarized in Table 2. The result of an independent t-test revealed that participants with missing values were significantly older than those with no missing values. No other statistically significant differences were observed for the other demographic variables. Ethnic identity consisted of an open-ended question. A content analysis revealed that a total of 38% of participants described themselves as “Caucasian/white”, while 26% described themselves as Canadians or mixed Canadians (e.g. French-Canadian, Ukrainian-Canadian, English-Canadian, Italian-Canadian), another 7% of participants identified as Aboriginal/First Nations/Metis, and the remaining 29% identified with other heritage/ancestry, with a predominance towards European ancestry.
Descriptive statistics were obtained on NA and SAC scores across the three measurement times. Mean scores and standard deviations are presented in Table 3. A clear trend emerges, showing that both NA and SAC scores increased after the intervention and continued to increase until the three-month follow-up. NA scores were slightly higher than SAC scores across the three measurement times.
Main analyses
The results of the LMM revealed that Age had a significant effect on NA scores, F (1, 109.65) = 6.10, p = 0.015. Model estimates, reported in Table 4, show a negative relationship between Age and NA scores, indicating that older participants tended to have lower NA scores. When controlling for differences in participants’ ages, the effect of Time on NA scores remained significant, F (2, 173.48) = 41.21, p < 0.001. Estimated marginal means based on the model for the three measurement times, along with standard errors and 95% confidence intervals, are reported in the last two columns in Table 3. Post-hoc comparisons using a Bonferroni adjustment revealed that scores on NA significantly increased between pre-intervention and post-intervention (p < 0.001), and between post-intervention and the three-month follow-up (p < 0.001). Pseudo-R2, calculated when Time was added in Model 3, indicates that Time explained an estimated 8.06% of the variance in NA scores. The effect size was estimated to be in the small to medium range, f2 = 0.09. The estimated explained variance increased to 10.71% when age was added as a covariate.
The results of the LMM revealed that Age did not have a significant effect on SAC, F (1, 111.47) = 1.59, p = 0.211. However, the effect of Time was significant, F (2, 176.92) = 27.37, p < 0.001. Estimated marginal means based on the model for the three measurement times, along with standard errors and 95% confidence intervals, are reported in the last two columns in Table 3, while model estimates are provided in Table 4. Post-hoc comparisons using a Bonferroni adjustment showed that scores on SAC significantly increased between pre-intervention and post-intervention (p < 0.001), and between post-intervention and the three-month follow-up (p = 0.014). Pseudo-R2, calculated based on Model 3, reveals that Time explained an estimated 7.82% of the variance in SAC scores. This value remained unchanged after the inclusion of age as a covariate. The effect size of Time on SAC was again estimated to be in the small to medium range, f2 = 0.08.
Discussion
The objective of the study was to evaluate whether participation in the Craving Change group intervention was related to an increase in eating self-efficacy, and whether this increase was maintained over time. Craving Change is a brief, structured, licensed intervention that is founded on the cognitive-behavioral model. As opposed to traditional nutritional counseling, Craving Change is innovative in its approach, focusing less on eating behaviors per se, but on the underlying psychological (e.g. cognitions and emotions) and social factors that can influence eating habits. The program is delivered through community-based health facilities by primary healthcare providers, including dieticians. Although the Craving Change intervention can be offered both individually and in a group setting, only the group intervention was evaluated for the purposes of this study. The target population for the study included individuals aged 18 years or older registered in the Craving Change group program, who sought support in managing eating habits, but who did not meet the diagnostic criteria for an eating disorder.
Four key findings emerged from this study. First, in accordance with past research investigating the effectiveness of psychoeducational group programs to promote healthy eating (Ashton et al., 2009; Jones and Clausen, 2013; Smith et al., 2020; von Ranson et al., 2010), the results of the current study showed that after participants completed the Craving Change program, there was a significant improvement in their reported sense of control or agency over their eating behaviors and this sense of control continued to increase between the end of the program and the 3-month follow-up. It is possible that as participants started practicing their new acquired skills and changing their eating habits, they gradually gained more confidence in their ability to regulate their eating behaviors. According to Bandura (2008), mastery experience is one of the most influential factors to promote self-efficacy. This trend is consistent with a study conducted by Warner et al. (2018), showing that day-to-day mastery and self-efficacy shared a reciprocal positive relationship among smokers who attempted to quit smoking. Another recent study investigated the relationship between mastery experiences, social support, and self-efficacy among adolescents and showed that mastery experiences was the strongest correlate for promoting and strengthening self-efficacy (Kleppang et al., 2023). In turn, self-efficacy is believed to play a major role in initiating health behavior changes and in influencing how much effort and perseverance are invested in maintaining these new behavior patterns (Bandura, 2004, 2008; Buckworth, 2017). This is supported by empirical evidence showing that self-efficacy is a robust predictor of eating behaviors (Bektas et al., 2021; Berman, 2006; Doerksen and McAuley, 2014; Kitsantas, 2000; Sniehotta et al., 2005; Vuillier et al., 2021). Consequently, the results of the present study suggest that the Craving Change intervention may represent a promising avenue for promoting healthy eating habits. However, it is important to note that the effect of participating in the Craving Change program on participants’ eating self-efficacy was not very strong, but rather ranged from small to medium.
Second, although eating self-efficacy in socially acceptable circumstances increased after participating in the intervention, scores remained lower across the three measurement times when compared to eating self-efficacy when faced with negative emotions. One explanation may be that participants in the program did not feel as confident in their ability to regulate their eating behaviors in situations where consuming food is considered socially acceptable or even encouraged (e.g. with friends, around the holiday time or after an argument), which highlights a potential area of improvement for the Craving Change program. Another explanation may lie in the measurement tool itself. In the initial validation study of the ESES, researchers reported lower mean scores for SAC compared to NA (Glynn and Ruderman, 1986). The same is true of a validation study for the brief version of the ESES (Lombardo et al., 2021). More research is needed to determine whether this pattern is observed with other samples participating in the Craving Change program.
Third, eating self-efficacy increased even after controlling for potential confounding variables, including age and whether participants completed the 3-month follow-up during the COVID-19 general lockdown that occurred in 2020. In fact, the inclusion of the COVID-19 factor in Model 5 did not improve model fit and showed that this factor had no effect on participants’ eating self-efficacy. This successful outcome is important given that a recent systematic review revealed a decrease in adherence to dietary guidelines following the onset of the COVID-19 pandemic, which could be explained by factors such as food insecurity, work and family related circumstances, and lack of social contact (González-Monroy et al., 2021). Results show that participants in the program were able to remain confident in their ability to regulate their eating behaviors, despite their exposure to a highly stressful situation.
Finally, when testing Model 6 among a reduced sample of 82 participants, the results showed that the effect of the Craving Change group program was positive, whether or not participants had access to other forms of individual support. This finding suggests that the program could be beneficial on its own or when combined with other forms of intervention.
Implications
Past research has shown that adherence to dietary recommendations remains relatively low (Gopinath et al., 2016; Leme et al., 2021), including within the Canadian population (Brassard et al., 2018; Kolahdooz et al., 2018). Non-adherence to a healthy diet increases the risks of cardiovascular disease, cancer, diabetes, and mortality due to chronic illness (Hu et al., 2019; Russell et al., 2013; Struijk et al., 2014; Yu et al., 2014), consequently elevating healthcare costs (Anekwe and Rahkovsky, 2013; Collins et al., 2013; Springmann et al., 2021). These data underscore the need for intervention strategies that can effectively address factors likely to influence dietary practices, including psychological factors (thoughts and emotions) and social factors. CBT-based therapies have been successfully used in the past to change eating behaviors (Alimoradi et al., 2016; Corstorphine, 2006; Munsch et al., 2007; Wade et al., 2017). However, the cost linked with CBT and the restricted availability of specialized services provided by trained psychologists or psychiatrists suggest that such interventions are not always accessible (Bouwman et al., 2020; Grilo et al., 2014; Jakicic et al., 2012; von Ranson et al., 2010; Wilson et al., 2000). This emphasizes the need for affordable and accessible interventions to encourage healthy eating (Bouwman et al., 2020).
Considering the economic advantages and accessibility of the Craving Change program, it effectively addresses a societal need. Administered through community-based health facilities by primary healthcare providers, it offers greater accessibility compared to interventions delivered by licensed therapists in clinical settings. Additionally, the cost of training facilitators is significantly lower than that of specialized services. Consequently, health authorities should continue investing resources in training frontline healthcare providers to deliver the Craving Change intervention. This intervention could be a first course of action for promoting healthy eating. In fact, a stepped care model has been proposed in the past to close the gap between supply and demand for support services (Jakicic et al., 2012; Wilson et al., 2000). This model calls for the use of least intensive and cost-effective interventions before moving on to more specialized and potentially costly care. These efforts could be particularly important among low-income populations and those in remote communities where access to specialized care is scarce. Offering the program through an online platform represents another avenue that could be explored and investigated in the future. Past research has shown comparable results when a group intervention was offered in person or online in synchronous format (Gollings and Paxton, 2006). Overall, this cost-effective intervention could improve the availability of support services providing necessary tools to promote healthy eating habits.
Limitations and strengths
These research findings should be considered in light of certain limitations inherent to the study design. One notable limitation stems from the absence of a control group, which impedes the ability to ascertain a causal relationship between the intervention program and changes in eating self-efficacy. Other confounding factors, such as participation attrition and missing data over the course of the study, could explain why participants’ sense of control over their eating behaviors increased over time. For instance, we observed that participants with missing values tended to be older compared to those with no missing values, which in turn may influence eating self-efficacy. Although a study by Berman (2006) showed no correlation between age and five domains of eating-self efficacy, as measured by the Weight Efficacy Life-Style Questionnaire, the present study found a negative relationship between age and eating self-efficacy when faced with negative emotions (NA subscale). However, the use of LMM to analyze data reduces the risk of bias due to participant attrition by accounting for variances across participants to predict the trajectory of participants with missing values over time, thus providing a more accurate estimate of the sample’s means on outcome variables (Walker et al., 2019). Additionally, age was included as a covariate in the model to assess whether the effect of participating in the program remained significant after considering differences in ages. Other potential confounding factors, such as completing the three-month follow-up during the COVID-19 general lockdown and having access to other forms of support, were also considered.
Another limitation was the use of self-reported data and the focus solely on eating self-efficacy. Thus, actual changes in behavior were not measured as part of this study. Consequently, it is not possible to determine whether the observed changes in eating self-efficacy would translate into changes in eating habits. However, past research has shown self-efficacy to be a robust predictor of eating behaviors (Bektas et al., 2021; Berman, 2006; Doerksen and McAuley, 2014; Kitsantas, 2000; Sniehotta et al., 2005).
Finally, the sample may not fully represent the overall population, as the majority of participants were women, employed or self-employed, and predominantly of European ancestry. Despite this, the sample exhibits good diversity in terms of age and socioeconomic background. To ensure adequate representation, participants were recruited from all six sites where the program was being offered throughout the city, including in more disadvantaged neighborhoods. The composition of the sample appears to reflect the characteristics of attendees who typically participate in the intervention program within the regional health authority. Future studies could build on this preliminary research by employing an experimental design to compare program participants with a control group. They could also assess changes in eating behaviors and evaluate whether these changes are mediated by an increase in eating self-efficacy. Additionally, they could evaluate whether the effect of the program on eating self-efficacy and eating behaviors is moderated by other factors, including sociodemographic variables (e.g. age, ethnicity, socioeconomic background), to determine whether the program is effective across diverse populations. Finally, qualitative research could be conducted to deepen our understanding of participants’ subjective experiences in the program and their own perceptions of how the program impacted them.
Conclusion
This study adds to the existing literature by exploring the potential benefits of the Craving Change group intervention, rooted in the cognitive-behavioral approach, on participants' sense of control over their eating behaviors. While the findings provide preliminary evidence suggesting a positive association between participation in the program and increased sense of control, it is important to note that this study is not experimental in nature. As such, we cannot draw definitive conclusions about the effectiveness of the program or establish causality. However, the observed increase in participants' sense of control suggests that this greater sense of agency could lead to positive changes in eating habits. Thus, despite the limitations of the study design, these preliminary findings suggest that the intervention may represent a cost-effective avenue to help people adopt and sustain healthy eating habits, which is key for reducing the risk of chronic disease and promoting better health and healthy aging. Future research, including controlled experimental studies, is needed to further investigate the impact of the intervention on promoting healthy dietary behaviors and reducing the risk of chronic diseases.
Funding: Internal funding from the Université de Saint-Boniface (grant # 839) was obtained to conduct this study.
Figure 1
Participant attrition and missing data across the three measurement times
[Figure omitted. See PDF]
Table 1
Model building process and model comparison using the Linear Mixed Model
| Model | Term added | N | Number of parameters | Model fit | Nested model | LRT against nested model | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| RLL | BIC | AIC | X2 | df | p | |||||
| NA | ||||||||||
| 1 | Random intercept for participants | 110 | 3 | 950.22 | 961.50 | 954.22 | ||||
| 2 | Random slope | 110 | Convergence warning | 1 | ||||||
| 3 | Fixed effect of Time | 110 | 6 | 882.39 | 899.30 | 888.39 | 1 | 67.83 | 3 | <0.001 |
| 4 | Fixed effect of Age (covariate) | 108 | 7 | 867.59 | 884.43 | 873.59 | 3 | 14.80 | 1 | <0.001 |
| 5 | Fixed effect of COVID-19 (covariate) | 108 | 8 | 864.97 | 881.80 | 870.97 | 4 | 2.62 | 1 | 0.106 |
| 6 | Fixed effect of Individual support (covariate) | 82 | 8 | 711.15 | 727.42 | 717.15 | 4 | 153.82 | 1 | <0.001 |
| SAC | ||||||||||
| 1 | Random intercept for participants | 110 | 3 | 872.66 | 883.94 | 876.66 | ||||
| 2 | Random slope | 110 | Convergence warning | 1 | ||||||
| 3 | Fixed effect of Time | 110 | 6 | 827.02 | 843.93 | 833.02 | 1 | 45.64 | 3 | <0.001 |
| 4 | Fixed effect of Age (covariate) | 108 | 7 | 818.39 | 835.23 | 824.39 | 3 | 8.63 | 1 | 0.003 |
| 5 | Fixed effect of COVID-19 (covariate) | 108 | 8 | 817.50 | 834.33 | 823.50 | 4 | 0.89 | 1 | 0.346 |
| 6 | Fixed effect of Individual support (covariate) | 108 | 8 | 662.52 | 678.78 | 668.52 | 4 | 154.98 | 1 | <0.001 |
Note(s): RLL = Restricted Log Likelihood; BIC = Bayesian Information Criterion; AIC = Akaike Information Criterion; LRT = Likelihood Ratio Test
Source(s): Authors’ own work
Table 2
Sociodemographic characteristics of all participants, participants with missing values and the remaining 72 participants
| Variables | All participants (N = 110) | Participants with missing values (N = 38) | Remaining participants (N = 72) | Comparisons |
|---|---|---|---|---|
| t (p) | ||||
| Age, years | 48.94 (11.60) | 52.35 (12.55) | 47.17 (10.73) | −2.25(0.036) |
| X2 (p) | ||||
| Gender | 0.14(0.705) | |||
| Women | 88.2 | 86.8 | 88.9 | |
| Men | 9.1 | 10.5 | 8.3 | |
| Missing | 2.7 | 2.6 | 2.8 | |
| Language | 3.73(0.292) | |||
| English | 71.8 | 63.2 | 76.4 | |
| English and French | 17.3 | 18.4 | 16.7 | |
| English and other | 9.1 | 15.8 | 5.6 | |
| English, French and other | 1.8 | 2.6 | 1.4 | |
| Education | 9.84(0.080) | |||
| Did not finish high school | 3.6 | 2.6 | 4.2 | |
| High school diploma | 22.7 | 18.4 | 25 | |
| College/technical diploma | 38.2 | 44.7 | 36.1 | |
| Undergraduate university diploma | 8.2 | 13.2 | 5.6 | |
| Graduate university diploma | 17.3 | 5.3 | 23.6 | |
| Other | 10 | 15.8 | 5.6 | |
| Employment | 7.44(0.190) | |||
| Employed/self-employed | 61.8 | 47.4 | 70.8 | |
| Retired | 16.4 | 26.3 | 11.1 | |
| Student | 1.8 | 2.6 | 1.4 | |
| On sick/disability leave | 5.5 | 2.6 | 6.9 | |
| Unemployed | 5.5 | 5.3 | 5.6 | |
| Other | 7.3 | 10.5 | 4.2 | |
| Annual household income | 6.60(0.252) | |||
| Under $20K | 8.2 | 13.2 | 5.6 | |
| $20K to $40K | 13.6 | 15.8 | 12.5 | |
| $40K to $60K | 23.6 | 13.2 | 29.2 | |
| $60K to $80K | 41.8 | 36.8 | 44.4 | |
| $80K to $100K | 1.8 | 2.6 | 1.4 | |
| Above $100K | 2.7 | 5.3 | 1.4 | |
| Missing | 8.2 | 13.2 | 5.6 |
Note(s): This table presents the demographic characteristics of the 110 participants enrolled in the study, including the 38 participants with missing values and the remaining 72 participants. Participants with and without missing values were compared using a t-test for age and a series of chi-square analysis for categorical variables. Results revealed that participants with missing values were significantly older than those with no missing values
Source(s): Authors’ own work
Table 3
Means (standard deviations) and estimated marginal means (standard errors and 95% confidence intervals) on eating self-efficacy when faced with negative emotions (NA) and in socially acceptable circumstances (SAC) at pre-intervention, post-intervention, and 3-month follow-up
| Descriptive statistics | Estimated marginal means | ||
|---|---|---|---|
| Means (SD) | Means (SEM) | 95% CI | |
| Negative affect (NA) | |||
| Pre (N = 105) | 3.69 (1.50) | 3.72 (0.14) | [3.44, 4.00] |
| Post (N = 91) | 4.34 (1.44) | 4.31 (0.14) | [4.03, 4.60] |
| 3-month follow-up (N = 87) | 4.76 (1.44) | 4.78 (0.15) | [4.49, 5.07] |
| Socially acceptable circumstances (SAC) | |||
| Pre (N = 105) | 3.29 (1.08) | 3.29 (0.12) | [3.06, 3.52] |
| Post (N = 91) | 3.82 (1.32) | 3.80 (0.12) | [3.56, 4.04] |
| 3-month follow-up (N = 87) | 4.14 (1.21) | 4.14 (0.12) | [3.90 4.39] |
Note(s): This table presents the means and standard deviations on the two outcome variables at each measurement time. The last two columns also present estimated marginal means, with standard errors and 95% confidence intervals based on the model
Source(s): Authors’ own work
Table 4
Fixed and random effect estimates of the final model (Model 4) for both outcome measures
| Fixed effects | Estimate | SE | t (df) | p |
|---|---|---|---|---|
| NA | ||||
| Intercept | 6.09 | 0.56 | 10.97 (110.30) | <0.001 |
| Time 1 | −1.06 | 0.12 | −9.01 (174.42) | <0.001 |
| Time 2 | −0.47 | 0.12 | −3.85 (172.80) | <0.001 |
| Time 3 | 0 | 0 | ||
| Age | −0.03 | 0.01 | −2.47 (109.65) | 0.015 |
| SAC | ||||
| Intercept | 4.68 | 0.45 | 10.52 (112.86) | <0.001 |
| Time 1 | −0.85 | 0.12 | −7.29 (178.16) | <0.001 |
| Time 2 | −0.34 | 0.12 | −2.86 (176.21) | 0.005 |
| Time 3 | 0 | 0 | ||
| Age | −0.01 | 0.01 | −1.26 (111.47) | 0.211 |
| Random intercept for participants | Variance | SE |
|---|---|---|
| NA | 1.28 | 11.14 |
| SAC | 0.78 | 3.64 |
Note(s): This table presents model estimates for Model 4, using Linear Mixed Model for both outcome measures, namely eating self-efficacy when faced with negative emotions (NA) and in socially acceptable circumstances (SAC)
Source(s): Authors’ own work
© Emerald Publishing Limited.