Introduction

Obesity is increasingly becoming a serious health concern around the world. According to the World Health Organization (WHO), the percentage of children and adolescents (aged 5–19) with obesity or overweight has risen from 4% (1975 est.) to around 18% (2016 est.). By 2016, 13% of adults worldwide were obese while 39% were overweight. Currently, more than one billion people worldwide live with obesity, the majority (650 million) being adults (World Health Organization, 2022). The United States of America (U.S.) has one of the highest rates of obesity in the world– currently approximately 42.4% of U.S. adults have obesity, while the obesity rate among children and adolescents is currently 19.7% (CDC, 2020; Stierman et al., 2021). Obesity rates are particularly high among non-Hispanic Black (49.6%) and Hispanic (44.8%) adults compared to the non-Hispanic White (42.2%) and Asian (17.4%) adults (Flegal et al., 2016; Hales et al., 2020). Similarly, among children and adolescents aged 2–19 years in 2017–2020, Hispanic children (26.2%) and non-Hispanic Black children (24.8%) have the highest rates of obesity, followed by non-Hispanic White children (16.6%), and lastly non-Hispanic Asian children (9.0%) (Stierman et al., 2021).

The recent growing rates of obesity are linked mostly to obesogenic environmental factors (Lee et al., 2019). The Ecological Systems Theory (EST), applied to the study of obesity by Davison and Birch (2001), posits that obesity predictors are interconnected. Similarly, the reasons for racial disparities in obesity rates are complex, and include genetic, behavioral, cultural, and environmental factors (Kumanyika et al., 2007; Kumanyika, 2008). One behavioral factor that may contribute to obesity rates is eating patterns or behavior, as research demonstrates differences in eating behaviors among racial groups (Gluck & Geliebter, 2002; Li et al., 2017; Simone et al., 2022; Thompson et al., 2015). Racial differences in eating behavior or dietary intake are due to a combination of several factors such as economic, environmental, and cultural (Emilien & Hollies, 2017; Ghosh-Dastidar et al., 2014; Mensah et al., 2022; Satia, 2009). For instance, African American families are more likely to live in “food deserts” with less access to healthy food choices which might lead to higher obesity rates (Ghosh-Dastidar et al., 2014).

Eating behavior is associated with obesity and diet-related illnesses such as heart disease, stroke and type 2 diabetes (Micha et al., 2017; Mizia et al., 2021). Specifically, self-regulation around eating or restricting food intake, and emotional eating are related to obesity (Porter et al., 2014). Having culturally sensitive validated eating behavior instruments is critical to understanding factors contributing to obesity disparities and designing effective interventions for especially Black populations.

Current study

Existing measures of eating behaviors such as the 35-item Adult Eating Behavior Questionnaire (AEBQ; Hunot et al., 2016) have largely been developed and validated in predominantly Caucasian or European or non-Black populations (Bjørklund et al., 2024; Guzek et al., 2020; He et al., 2021; Hunot et al., 2016; Jacob et al., 2022; Malan et al., 2017; Warkentin et al., 2022). Given potential cultural differences in eating behaviors or patterns (Gluck & Geliebter, 2002), existing instruments may not fully capture relevant behavioral constructs in different groups the same way. Existing eating behavior or diet-related questionnaires often fall short in capturing the full range of behaviors and constructs across diverse cultural groups due to various factors such as differing dietary habits, food availability, cultural norms, and socioeconomic conditions. Moreover, most eating behavior or diet-related questionnaires or measures are developed with mostly single groups or specific groups. For instance, the Eating Disorder Examination Questionnaire (Habashy et al., 2023) was developed with a predominately White female sample. Similarly, some validation studies of the Three-Factor Eating Questionnaire have used majority White samples (Papini et al., 2022). Additionally, some AEBQ validation studies (Cohen et al., 2021) and other eating behavior questionnaire studies (Klimek et al., 2021) have suggested the validation and development of eating behavior questionnaires with more diverse samples, especially testing for measurement invariance, that is, if the measures/questionnaires work the same across groups.

A questionnaire that measures salient eating behaviors for instance among Black adults could help identify targets for promoting healthy eating. For example, Thompson et al. (2015) found significant differences in food addiction, with African Americans having higher symptom scores than White Americans. Moreover, longitudinal evidence shows differences in eating behaviors. Simone and colleagues (2022) reported linear increases in unhealthy weight control behaviors from adolescence to adulthood among Black/African American women relative to White women. A scale that reliably captures this and other relevant constructs could assist in developing strategies to regulate emotion-driven eating. This highlights the need for rigorous validation of existing eating behavior measures. As such, this study seeks to evaluate the psychometric properties, and validate the AEBQ scale using a predominately Black young adult cross-sectional data sample from a Historically Black University (HBCU). First, by evaluating the internal consistency, and the factor structure of the AEBQ. Secondly, by assessing the associations and direction between the different AEBQ subscales. Lastly, examining the associations between the different AEBQ subscales, and BMI, and depression controlling for age and sex.

Methods

Procedure and data

The data on eating behavior and demographic information were collected as part of a larger study entitled, “The Neuropsych Study”. To be included in the study, participants needed to be between the ages of 18–25, and currently enrolled in the University. Participants were recruited for this study via Blackboard posts by professors and posted physical advertisements for young adult (18–25 years) undergraduate students attending the University. The physical advertisements were posted in a large undergraduate classroom building that housed courses for Psychology, Mathematics, Engineering, Biology, and Chemistry departments at the University. Upon entry, the participant was asked to sit at a round table, as the research assistant sat across from them in the clinical assessment room. The trained research assistant explained the study-related activities to the participant and addressed any questions the participant posed. After receiving informed consent, a trained research assistant began the research protocol, which included participants completing paper and pencil self-report measures including a demographic questionnaire, the Adult Eating Behavior Questionnaire (AEBQ), Beck Depression Inventory-II (BDI-II), and other study measures such as the Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) and Neuropsychological measures. Each of these were completed by the participant in the laboratory’s clinical assessment room with the research assistant present for any questions or concerns. The research protocol administration took approximately 45 min to complete. After completing the study protocol, the participants’ names were collected and provided to course professors for extra credit as compensation. The study was carried out from 02/2018 to 05/2020.

This cross-sectional study received annual approval from the University Institutional Review Board. A Monte Carlo simulation power analysis for Factor Analysis was performed in MPLUS to determine how much sample was needed for the analysis. In order to achieve statistical power at 0.8 or higher (Cohen, 1992, 2013), a sample size of n ≥ 100 is needed when there are two factors, and a sample of n > 200 is needed when there are more than two factors. In addition, a meta-analysis study (Kyriazos, 2018) for power analysis or the needed or required sample size for Factor Analysis reported that a sample of 200 is adequate, specifically a Factor Analysis sample of 50 very poor, 100 as poor, 200 as fair, 300 as good, 500 very good and 1000 excellent. For this study, the cross-sectional data sample size (n = 229) consisted of mostly females (73.4%) with an average age of (20.97 ± 4.31), and average BMI of (29.37 ± 8.27) (see Table 7 in the appendix).

Measures

Body mass index (kg/m2) (BMI)

BMI was calculated based on self-reported height and weight data. Body mass index (BMI) was calculated by dividing weight in pounds (lbs) by height in inches (in) squared and multiplying by a conversion factor of 703 (CDC, 2022).

Adult Eating Behaviors Questionnaire (AEBQ) (Hunot et al., 2016)

AEBQ is a 35-item self-report questionnaire that uses a five-point Likert scale. Items are divided into 4 Food Approach subscales and 4 Food Avoidance subscales. The Food Approach subscales include Hunger (5 items; e.g. I often feel hungry), Food Responsiveness (4 items; e.g. I am always thinking about food), Emotional Overeating (5 items; e.g. I eat more when I’m upset), and Enjoyment of Food (3 items; e.g. I enjoy eating). The Food Avoidance subscales include Satiety Responsiveness (3 items; e.g. I get full up easily), Emotional Undereating (5 items; e.g. I eat less when I’m worried), Food Fussiness (5 items including 3 reverse coded items; e.g. I refuse new foods at first), and Slowness in Eating (4 items including 1 reverse coded item; e.g. I eat slowly). Mean scores were calculated for each subscale. The rest of the item description and subscale descriptive statistics are found in Tables 6 and 7 in the appendix.

Depression

Depression symptoms were assessed using Beck Depression Inventory-II (BDI-II) which is a 21-item self-report measure used to determine the severity of depression. Each item is rated on a 4-point Likert-type scale ranging from 0 to 3, with the total scores ranging from a low value of 0 to a high value of 63 (Beck et al., 1996). For this study, the items in the inventory were summed to get a total score, and higher scores imply higher symptomatology and severity of depression (Beck et al., 1996). Studies (García-Batista et al., 2018) have indicated Beck Depression Inventory-II (BDI-II) to have high reliability (α = 0.89). The Beck Depression Inventory-II (BDI-II) has high reliability (α = 0.90) in a predominantly African American sample (Grothe et al., 2005).

Demographics

Participants self-reported age and sex. Age was self-reported by the participant in years. Sex was dummy coded into “0” for male, and “1” for female.

Statistical analysis

The analyses were performed in MPLUS (Muthen & Muthen, 1999–2016) and R statistical language (R Core Team, 2021) for data exploration. First, a frequency of all the variables was obtained for visual observation of any data abnormalities such as outliers and missing values. A boxplot was then used to identify any values considered outliers. Values more than 1.5 times the interquartile range, above the third quartile, or below the first quartile on the boxplot were considered outliers (extreme values). There were no detectable outliers for all the variables used in the analysis. From the simple descriptive analysis of frequencies, missing values were identified and then recoded to a numerical value (9999) for easy identification in the analysis process. A small amount of AEBQ data was missing (0.46%). Further probing revealed that this data was missing from five AEBQ items. Little’s MCAR test and Chi-Squared Test for categorical variables did not support the assumption that data were missing completely at random (MCAR). Also, due to the confidential nature (i.e., non-disclosure identifying information of the participants) of the study, there is a lack of evidence to support that the missing data mechanism was missing not at random (MNAR). As such, data was assumed to be missing at Random (MAR), and handled with both full information maximum likelihood (FIML) for the CFA models and multiple imputations for multivariate multiple linear regression analyses. To account for the non-normality of the data, a sandwich estimator was used in analyses to adjust for the standard errors in the models (Freedman, 2006). For the CFA models, a sandwich estimator implemented in MPLUS using MLR (Maximum likelihood estimator with robust standard errors) was used instead of the weighted least square estimator because of missing data which was handled using FIML in MPLUS. Because of the amount of missing data in the covariate of BMI, a multiple imputation with 20 datasets was performed in MPLUS to handle missing values during the multivariate multiple linear regression analyses. The female participants were dummy coded to one while male participants were dummy coded to zero. BMI, depression, and age were left as continuous variables in the analyses.

Reliability (Cronbach’s alpha estimation) and Confirmatory Factor Analyses were first performed to test for internal consistency and construct validity of the AEBQ, respectively. For this study, standardized factor loadings of ≥ 0.40 were considered adequate for the cutoff point. Only statistically significant indicators were considered acceptable (Wang & Wang, 2019).

A correlation analysis among the 8 subscales was performed to assess for convergent validity. Additionally, in order to further test validity, a multivariate multiple linear regression model to predict each AEBQ subscale using BMI as the main predictor, controlling for sex and age was performed. Another multivariate multiple linear regression model which included depression as another predictor was performed to further assess validity. Sex and age were controlled for because previous studies have found differences in eating behavior based on sex/gender and age (Feraco et al., 2024; Hunot-Alexander et al., 2019). Additionally, other AEBQ validation studies (Hunot-Alexander et al., 2019; Mallan et al., 2017) controlled for sex/gender and age in the statistical analyses.

We used BMI and depression as a proxy for other existing measures or tests. Therefore, a multivariate multiple linear regression was used to establish the degree to which the AEBQ subscales related to the already established measures like BMI and depression to further assess the validity of the AEBQ. A multivariate multiple linear regression model is a correlational statistical method i.e. measures the linear association between the dependent and independent variables. Given that this study aimed to estimate the associations of the different AEBQ subscales as outcome variables, and the associated single set of covariates, utilizing multivariate multiple regression models allowed for the simultaneous inclusion of all the variables of interest in the models, as well as accounting for covariance/correlation among the AEBQ subscale outcomes/dependent variables. Additionally, performing multivariate multiple regression allowed easy comparison of the model coefficients across the AEBQ subscale outcomes.

Results

Reliability analysis

Except for the Hunger subscale (α = 0.67), all Cronbach’s alphas for the other 7 subscales were acceptable (≥ 0.70) (see Table 1). In line with prior AEBQ validation studies (Mallan et al., 2017), combining the Hunger and Food Responsiveness items improved the Cronbach’s alpha to (α = 0.77), hence all the alphas were acceptable (≥ 0.70) (George & Mallery, 2003; Hair et al., 2010), indicating good internal consistency of the items in the scale.

Table 1 Reliability analysis
Full size table

Confirmatory Factor Analysis (CFA)

The mode fit indices of a Chi-Square Test (𝜒2), Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR) were used to assess model fit and to compare the estimated models. CFI or TLI values > 0.95, RMSEA values < 0.06, and SRMR < 0.08 are considered acceptable or close fit (Hu & Bentler, 1999). A small 𝜒2 corresponds to a good fit, 𝜒2 value of 0 corresponds to a perfect fit, and the test is expected not to reject the null hypothesis i.e. it should be non-significant (Wang & Wang, 2019).

The first Confirmatory Factor Analysis (CFA) model had all 8 subscales, and their associated indicators (35 items in total). For model identification, one indicator/measure for each latent variable of the eating behavior subscale was fixed to one, and the common latent variable means fixed to 0. To improve the model’s fit (root mean square error of approximation), the number of parameters and paths were adjusted based on the modification indices (below 10) and normalized residuals for covariances (significant or not). A covariance between the standard errors for some items was determined. For this model, the RMSEA was 0.038, < 0.060. (CI [0.030, 0.045]), CFI (0.944) and TLI (0.936), SRMR (0.065), and Chi-Square Test (P <.05) (see Table 2). The standardized factor loadings for all of the indicator variables were significant and ranged from 0.33 to 0.93. The Hunger subscale had the lowest factor loadings (see Table 8 in the appendix). Suggesting that some items of the Hunger subscale could have been non or less contributing to the general factor structure of the AEBQ.

Table 2 Model fit indices
Full size table

In line with prior AEBQ validation studies (Mallan et al., 2017), a second CFA with the items of Food Responsiveness and Hunger combined was performed. This model had 7-subscales, with associated indicators (35 items in total). RMSEA increased to 0.040, < 0.06. (CI [0.033, 0.047]), the Chi-Square Test was still significant, P <.05, CFI (0.936) and TLI (0.928) and SRMR (0.067) (see Table 2). The standardized factor loadings for all the indicator variables were significant and ranged from 0.26 to 0.93 (see Table 8 in the appendix).

Lastly, based on prior ABEQ validation studies (Bjørklund et al., 2024; Hunot-Alexander et al., 2022; Mallan et al., 2017) and due to the performance of Hunger subscale based on poor factor loadings, and reliability (Cronbach’s alpha < 0.70), a third Confirmatory Factor Analysis with 30 items was performed excluding Hunger subscale. Like the 8-factor model, except for the Chi-Square Test (P <.05), the rest of the model fit indices (RMSEA (0.040), (CI [0.031, 0.049]), CFI (0.949) and TLI (0.940), SRMR (0.061) were satisfactory (see Table 2). The standardized factor loadings for all of the indicator variables were significant and ranged from 0.47 to 0.93 (see Table 8 in the appendix).

In summary, for all three models, RMSEA (< 0.06) and SRMR (< 0.08) were satisfactory. However, the Chi-Square Tests were significant, both CFI and TLI were < 0.95 for two of the models. Thus, the model fit may not have been adequate enough (marginal fit). The 8-factor model had the lowest RMSEA but the 7-factor model without hunger had higher CFI, TLI, lower SRMR and Chi-Square Test value, as well as higher factor loadings overall. This implies that the 7-factor model without the Hunger subscale seems to be a better model overall.

Correlations and multivariate multiple linear regressions

The average scores of responses to item questions for each subscale of the AEBQ were used in the correlation analysis and multivariate multiple linear regression analyses.

In order to assess convergent validity among the subscales, a correlation analysis was performed, and correlation estimates of the subscales ranged from small (-0.035) to large (0.574) (see Table 3). The Food Approach and Food Avoidance subscales were negatively associated (see Table 3). The Food Approach subscales had higher positive correlations than the Food Avoidance subscales (see Table 4). Except for the small correlation estimate between Enjoyment of Food and Emotional Overeating (0.110), most Food Approach subscales’ correlation estimates ranged from moderate (0.318) to large (0.574). Except for the moderate correlation estimate between Slowness in Eating and Satiety Responsiveness (0.457), most Food Avoidance subscale correlation estimates were small ranging from 0.055 to 0.243.

Table 3 Correlations among the subscales
Full size table
Table 4 Food approach subscales vs. food avoidance subscales
Full size table

In line with other AEBQ validation studies (He et al., 2021; Hunot-Alexander et al., 2019; Mallan et al., 2017; Warkentin et al., 2022) and to further assess validity, a multivariate multiple linear regression model containing all 8 subscales as outcome variables was performed with BMI as a main predictor controlling for age and sex. There were slight differences between the multivariate multiple linear regression analyses estimates of the non-imputed data and that of the imputed data. The multivariate multiple linear regression model estimates of the imputed data were reported and interpreted in this study. Multiple imputation imputes the missing data points based on the statistical characteristics of the data. As such, multiple imputation provides accurate estimates among the associations of variables of interest (Li et al., 2015). Additionally, multiple imputation was used for a sensitivity analysis to assess the robustness or stability of the estimates (Héraud-Bousquet et al., 2012). Only the models in which at least one covariate was statistically significant were used for result interpretations, and the rest of the models are in Table 10 in the appendix. For all the models, the variance explained in each outcome ranged from 4.2 to 12.0%. BMI was statistically significant and positively related to Emotional Overeating and Food Fussiness, and negatively related to Emotional Undereating (B = 0.03, β = 0.24, p <.05; B = 0.03, β = 0.20, p <.05 and B=-0.03, β=-0.19, p <.05), respectively (see Table 5). The results indicate that a one standard deviation increase in BMI is expected to increase Emotional Overeating by 0.24 standard deviations and Food Fussiness by 0.2 standard deviations after controlling for age and sex. Similarly, a one standard deviation increase in BMI is expected to decrease Emotional Undereating by 0.19 standard deviations after controlling for age and sex.

Table 5 Multivariate multiple linear regression models (n = 229)
Full size table

Among the controls, sex was statistically significant and positively related to Emotional Overeating, Emotional Undereating, Food Responsiveness, Satiety Responsiveness, and Hunger, (B = 0.39, β = 0.19, p <.05; B = 0.45, β = 0.19, p <.05; B = 0.28, β = 0.14, p <.05; B = 0.62, β = 0.31, p <.05 and B = 0.34, β = 0.18, p <.05), respectively (see Table 5). The results indicate that being female compared to being male is expected to increase Emotional Overeating, Emotional Undereating, Food Responsiveness, Satiety Responsiveness, and Hunger by 0.19, 0.19, 0.14, 0.13, and 0.18 standard deviations, respectively, after controlling for BMI and age. Likewise, age was statistically significant and negatively related to Emotional Undereating, Food Responsiveness and Hunger, (B= -0.04, β=-0.16, p <.05; B= -0.04, β=-0.17, p <.05 and B= -0.03, β=-0.16, p <.05), respectively (see Table 5). The results indicate that a one standard deviation increase in age is expected to decrease Emotional Undereating, Food Responsiveness, and Hunger by 0.16, 0.17, and 0.16 standard deviations, respectively, after controlling for BMI and sex.

Several studies have examined the association of depression and eating behavior (Cifuentes et al., 2022; Paans et al., 2018; Sims et al., 2014). Therefore, a second multivariate multiple linear regression model containing all 8 subscales as outcome variables and depression as another covariate (in addition to age and sex) was performed. Only the model in which depression was statistically significant was reported (see Table 9). Depression was statistically significant and positively related to Emotional Undereating, (B = 0.03, β = 0.22, p <.05) (see Table 9 in the appendix). The results indicate that a one standard deviation increase in depression total score is expected to increase Emotional Undereating by 0.22 standard deviations after controlling for BMI, sex and age. BMI still significantly predicted Emotional Overeating, Emotional Undereating and Food Fussiness.

Discussion

There is limited evidence that shows the validation of the AEBQ in minority groups in the United States. As such, this study aimed to evaluate the psychometric properties of AEBQ (explore the structure, consistency, and validity of the AEBQ) using a predominately Black young adult sample attending an HBCU in the United States. Consistent with other studies (Mallan et al., 2017), the Hunger subscale had the lowest Cronbach’s alpha (below 0.70) compared to the other subscales which had acceptable Cronbach’s alphas (above 0.70). The seven-factor CFA model without the Hunger subscale seems to be performing better than the 8- factor model and 7-factor model with Hunger and Food Responsiveness items combined. The 7-factor CFA model without Hunger had the lowest SRMR, higher CFI, TLI, and factor loadings overall compared to the other two models. This finding is consistent with other studies (Bjørklund et al., 2024; Guzek et al., 2020; Hunot-Alexander et al., 2019) which found a 7-factor CFA model without Hunger to be the most adequate model. Studies such as Hunot-Alexander and colleagues (2022) suggest the removal of the Hunger subscale from future studies. Moreover, combining the Hunger and Food Responsiveness items did not improve the CFA model fit. This result is consistent with Mallan and colleagues (2017) whose model achieved a worse fit when the Food Responsiveness and Hunger items were combined. Additionally, in the 8-factor CFA model, two items of the Hunger subscale (AEBQ 6, and AEBQ 34) (see Table 6 in the appendix) consistently had the lowest factor loadings compared to other items. This could be due to the individual and cultural differences in the perception and interpretation of physical hunger verses appetitive trait (Hunot-Alexander et al., 2022) in this study’s sample. Two items of Satiety Responsiveness (AEBQ 23 and AEBQ 30) (see Table 6 in the appendix) had lower factor loadings throughout all the models. This may be because these specific items in this sample were not capturing the same eating behaviors as the rest of the items in the Satiety Responsiveness subscale or the items were not culturally relevant to this specific sample. To this point, scholars have found cultural influences on the eating patterns or behaviors among Black or African Americans (Airhihenbuwa et al., 1996; Hargreaves et al., 2002). Moreover, community and culture impact eating behavior and dietary habits among African Americans with regard to consumption of certain foods that may be deemed as important to cultural heritage (James, 2004).

The correlation analysis results indicated the Food Approach subscales to have more convergent validity than the Food Avoidance subscales. This finding is consistent with Molitor and colleagues (2021) whose findings demonstrated Food Approach subscales to have more convergent validity than the Food Avoidance subscales. Food Approach subscales demonstrating more convergent validity than the Food Avoidance subscales could be related to the specific items of each subscale vis a vis how they measure each behavior. This could also imply that the Food Approach subscales may be explained by a common underlying behavioral target that could benefit from a shared targeted intervention. The poor convergent validity of mostly Food Avoidance subscales as demonstrated by weak correlation estimates may be attributed to one’s ability to sufficiently self-report a specific food avoidance behavior. The poor convergent validity of the Food Avoidance subscales could also be due to the cultural irrelevance of specific food avoidance behavior. Prior research (Oney et al., 2015) indicated stronger Black or African American cultural orientations to be associated with less eating disordered behaviors and attitudes of dieting, bulimia, and oral control.

Contrary to other AEBQ validation studies (Hunot-Alexander et al., 2019; Mallan et al., 2017), BMI significantly predicted Emotional Overeating, Emotional Undereating, and Food Fussiness after statistically adjusting for age and sex. This study’s findings suggest that an increase in BMI increased Emotional Overeating while it decreased Emotional Undereating. This finding is consistent with prior studies that have found associations between one’s BMI or obesity/overweight status, and emotional eating (Dakanalis et al., 2023). The current findings also suggest increases in BMI are associated with increases in Food Fussiness, which does not align with previous recent literature (He et al., 2021; Mallan et al., 2017; Warkentin et al., 2022). Picky eating in adults may be related with unhealthy eating, which can lead to higher BMI (Hunot et al., 2016; Kauer et al., 2015). Specifically, when picky eating is done in excess, it might hinder the maintenance of a healthy weight (Zickgraf et al., 2016). It could also be that some picky eaters gravitate towards higher energy dense and easy to consume foods (Dubois et al., 2022) which might eventually lead to excessive calorie intake and weight gain.

Sex was entered into the model as a control and significantly predicted Emotional Overeating, Emotional Undereating, Food Responsiveness, Satiety Responsiveness and Hunger. Relative to men, women on average are found to have higher emotional eating habits (Hunot-Alexander et al., 2019; Guerrero-Hreins et al., 2022). In addition, women compared to men are more likely to engage in food avoidance behaviors (Bärebring et al., 2020; Wardle et al., 2004). Interestingly, the results in this study seem to suggest that females compared to males have higher levels of food responsiveness and hunger-related feelings. Prior findings show women to have more restrained eating behavior compared to men (Feraco et al., 2024; Khor et al., 2002) but these findings indicate they might be experiencing more hunger and restraining their eating. As such, these findings should be considered within the context of not only the sample type (predominantly Black), but also the potential sex differences as suggested in our results. It is plausible that men and women differ in their food avoidant behaviors (Katzman et al., 2021).

Additionally, this study indicated that an increase in age was associated with a decrease in Emotional Undereating, Food Responsiveness, and Hunger traits. Research suggests that being older is associated with more restrained eating behavior compared to being younger (Abdella et al., 2019; Ferreira-Pêgo et al., 2020).

To provide further context for the AEBQ validation, we determined the association between eating behavior and depression symptoms. Our results suggested a positive association between depressive symptoms and Emotional Undereating. This finding is not surprising since prior evidence links depression diagnosis to emotional eating (Cifuentes et al., 2022; Paans et al., 2018). In contrast to our findings, Emotional Overeating was associated with higher depressive symptoms than their undereating counterparts (Dixit et al., 2023). The differentiation between Emotional Undereating and Emotional Overeating may be associated with an individual’s parental engagement or family functioning during childhood and adolescent development (Bjorklund et al., 2019). The current findings add to a developing literature elucidating nuance within emotional eating behavior that may vary by not only race/ethnicity, but also childhood family functioning.

Strengths and limitations

It is important to acknowledge the limitations. These limitations include the absence of established norms specific to the population of the phenomenon under investigation. Compared to the other validation studies, this study had a smaller sample size, which might have affected the model fit indices, especially the Chi-Square Test. The majority of the sample was female college students which limit generalizability to the general population. However, this is the first study according to our knowledge to validate the AEBQ using a predominately Black/African American sample. As such, there were no established norms specific to this population from larger sample sizes. Researchers should replicate this study with other samples of predominantly Black samples to ensure the results are generalizable.

Implications and conclusions

Overall, the results provide some evidence to confirm the validity and reliability of the AEBQ in the studied sample. Food Avoidance subscales demonstrated more internal consistency than the Food Approach subscales. The CFA results indicated a 7-factor model in which the Hunger subscale was eliminated to fit the data better overall. The Food Approach subscales demonstrated more convergent validity than the Food Avoidance subscales. BMI significantly predicted both Emotion Overeating and Emotional Undereating, as well as Food Fussiness. Moreover, both sex and age significantly predicted Emotional Undereating, Food Responsiveness, and Hunger. Further, sex significantly predicted Emotional Overeating and Satiety Responsiveness. Lastly, depression predicted Emotional Undereating.

Based on these findings, an implication of this study is that tools geared toward the understanding of eating behaviors should be further evaluated for use in diverse populations. Moreover, some eating behaviors with a common underlying mechanism as demonstrated by convergent validity of Food Approach subscales could benefit from integrated interventions rather than targeting each eating behavior separately. Nevertheless, there is a need to further examine the items that seem to be non-contributing to the factor structure including the re-evaluation of the Hunger sub-scale. Culturally validated tools are needed to advance research on contributors to obesity disparities and interventions tailored to salient eating patterns or behaviors. Large samples of mostly minority groups are needed for future studies, as well as longitudinal studies to examine the continuity and stability of eating behavior or patterns over time. Given that, there were associations between some AEBQ subscales, and sex and age, future studies should examine measurement invariance of the AEBQ scale across sex and age.