Food insecurity, defined as limited availability of nutritionally adequate and safe foods or an uncertain ability to acquire acceptable foods in socially acceptable ways(Reference Andersen1), affects 12 % of US households(Reference Coleman-Jensen, Rabbitt and Gregory2). Additionally, there are racial/ethnic disparities in food insecurity, with US Latinx households experiencing a higher prevalence of food insecurity (19 %) than non-Latinx White households (10 %)(Reference Coleman-Jensen, Rabbitt and Gregory2). Unless new and efficient policies are adopted, food insecurity is likely going to be magnified over time due to continued demographic shifts whereby Latinxs are expected to constitute one-third of the US population by 2065(Reference Cohn3).
Food insecurity has been consistently associated with nutrition-related health conditions (i.e. obesity, pre-diabetes and type 2 diabetes)(Reference Gooding, Walls and Richmond4–Reference Larson and Story6). Emotional eating (EE), which is a dysfunctional eating behaviour characterized by eating due to an inability to resist negative emotions(Reference Karlsson, Persson and Sjostrom7), also has been linked to nutrition-related chronic conditions (i.e. obesity, type 2 diabetes and hypertension)(Reference Lopez-Cepero, Frisard and Lemon8). Few studies have explored mechanisms by which food insecurity may influence behavioural factors, such as EE, particularly among ethnic minority populations disproportionately affected by food insecurity.
Food insecurity is a stressful life experience and studies have documented positive associations between food insecurity and stress(Reference Martin, Maddocks and Chen9,Reference Chung, Kim and Kwak10) . Further, there is evidence that eating behaviours such as EE are triggered during stressful circumstances. In fact, 40 % of US adults report that they change their eating behaviours during times of stress(11). Several studies have shown a positive association between stress and EE(Reference Nguyen-Rodriguez, Chou and Unger12–Reference Belcher, Nguyen-Rodriguez and McClain14). However, only three studies have examined the relationship between food insecurity and EE(Reference Sharpe, Whitaker and Alia15–Reference Lofton and Connell17). These studies focused mainly on women and were not specific to Latinx populations. Because EE and food insecurity affect dietary intake(Reference Leung, Epel and Ritchie18), and both are associated with nutrition-related chronic conditions(Reference Gooding, Walls and Richmond4–Reference Larson and Story6,Reference Lopez-Cepero, Frisard and Lemon8) , it is important to understand the relationship between food insecurity and EE, as well as to uncover potential mediators of this relationship (i.e. perceived stress). Such knowledge would help identify novel intervention targets to alleviate EE and decrease nutrition-related health disparities in Latinxs.
Thus, the objective of the present study was to examine the association between food insecurity and EE and evaluate the mediating role of stress in a sample of US Latinxs. We hypothesized that food insecurity would be positively associated with EE and that this relationship would be mediated, in part, by perceived stress.
Methods
Study design and participants
The current cross-sectional study used data from the Latino Health and Well-being Study(Reference Silfee, Rosal and Sreedhara19). Participants were recruited from the Greater Lawrence Family Health Center, a federally qualified community health centre providing health-care services to 80–85 % of the Latinx population in the city of Lawrence, MA, USA, between September 2011 and May 2013. The sample was stratified by age (21–34, 35–54, 55–85 years) and sex. To be eligible, individuals had to be of Latinx or Hispanic ethnicity, Spanish or English speaking, and between 21 and 85 years of age. Individuals with plans to move out of the area within the study period (12 months), experiencing cognitive or physical impairments for participation, or unwilling to provide informed consent were excluded.
Individuals were contacted via mail with a letter signed by the health centre Chief Medical Officer describing the purpose of the study and stating that a study coordinator would call to provide additional information about the study, assess eligibility and ask about interest in participating. Individuals were provided with a toll-free number to opt out of the study and those who did not opt out were contacted by bilingual trained study personnel. Eligible patients interested in participating were scheduled for a baseline visit. At the baseline visit, study staff obtained written informed consent and conducted face-to-face interviews that included a verbally administered survey assessment and anthropometric measurements. The visits were conducted in the participant’s preferred language (English or Spanish) and lasted approximately 2·5–3 h. This study was approved by the Institutional Review Board of the University of Massachusetts Medical School.
A total of 3067 individuals were screened for eligibility; 284 individuals were ineligible (i.e. age, cognitively impaired, moving, residing outside the study area and deceased) and 1547 were unreachable (i.e. wrong number, number disconnected and invitation letter returned). Of the remaining potential participants, 484 refused to participate and 150 were scheduled for a study visit but did not attend. In all, 602 individuals enrolled in the study. The current analysis excluded fourteen participants who had incomplete data on covariates (perceived income, n 4; household size, n 9; country of birth, n 1). Because EE may be higher during pregnancy(Reference Sui, Turnbull and Dodd20), we excluded eight pregnant women. The final sample size was 580.
Measures
Trained bilingual study personnel conducted standardized interviews that included food insecurity, psychosocial factors, eating behaviours, sociodemographics and anthropometric measures.
Food insecurity
Food insecurity was measured with the six-item US Department of Agriculture Household Food Security Scale(Reference Blumberg, Bialostosky and Hamilton21). This instrument was derived from the eighteen-item US Department of Agriculture Household Food Security questionnaire(Reference Carlson, Andrews and Bickel22). The scale measures household food insecurity using the previous 12 months as the reference period and asks about financial means to obtain food and food-conserving behaviours. The food insecurity score was calculated by adding the responses of all six items. The total score ranged from 0 to 6, with higher scores indicating greater food insecurity. As per guidelines, scores were categorized as food secure (score = 0 or 1; referent group) and food insecure (score ≥ 2)(Reference Blumberg, Bialostosky and Hamilton21). This scale has been shown to have good sensitivity and specificity relative to the eighteen-item US Department of Agriculture Household Food Security questionnaire and has been validated in Spanish(Reference Harrison, Stormer and Herman23).
Perceived stress
Perceived stress was measured with the ten-item Perceived Stress Scale(Reference Cohen, Kamark and Mermelstein24). This scale has been widely used and shown to have good reliability, with Cronbach’s α ranging between 0·7 and 0·9(Reference Cohen, Kamark and Mermelstein24). The original version of this scale, which consists of fourteen items, has also been validated in Spanish(Reference Gonzalez Ramirez and Landero Hernandez25). For the present study, we used the translated Spanish items of the original scale, and we cognitive tested the items with community staff to ensure clarity and language adequacy for our target population. The Perceived Stress Scale measures general stress perceptions without reference to the source and asks about uncontrollability and inability to cope. For example, using the previous month as the reference period, participants were asked: ‘How often have you been upset because of something that happened unexpectedly?’, ‘How often have you felt that you were unable to control the important things in life?’ and ‘How often have you felt nervous and stressed?’ Response options were: ‘never’, ‘almost never’, ‘sometimes’, ‘fairly often’ and ‘very often’. To calculate the score, responses to all items were summed. The final score ranged from 0 to 40, with higher scores suggesting greater perceived stress.
Emotional eating
EE was measured with the Three-Factor Eating Questionnaire (TFEQ)-R18 V2(Reference Cappelleri, Bushmakin and Gerber26). This scale was translated into Spanish by a professional translator and was pre-tested for fidelity and clarity using cognitive interviews with community staff from various Latino heritages representing the study target population. The EE subscale consists of six items that evaluate eating in response to negative emotions (e.g. anxious, wound up, depressed, lonely, sad and nervous). Response options for all items were: ‘definitely false’ (= 1), ‘mostly false’ (= 2), ‘mostly true’ (= 3) and ‘definitely true’ (= 4). To calculate the EE score, responses of all items were summed and divided by 6 to generate a mean score (ranging from 1 to 4, with higher scores indicating greater EE). We created three categories for the total EE score because the score was not normally distributed and no transformation improved the distribution (residuals from regression models with the transformed scores were still not normally distributed). Consistent with previous research(Reference Camilleri, Mejean and Kesse-Guyot27), the following three categories were created: (i) No EE, (ii) Low EE and (iii) High EE. The No EE category was comprised of individuals with a score of 1, thus answering ‘mostly false’ to all items. The Low and High EE groups were created using a median split, with a median that exclude scores of 1·0 (median = 1·83). Thus, a score at or below the median was categorized as ‘Low EE’ and a score above as ‘High EE’. Overall, this scale has shown good reliability and factor structure(Reference Cappelleri, Bushmakin and Gerber26) and its Cronbach’s α coefficient is adequate in this sample of Latinx adults (0·86).
Demographics and covariates
Confounders were selected a priori because of their known association with food insecurity and dysfunctional eating behaviours and included: sex, age, education, marital status, household size and country of birth(Reference Coleman-Jensen, Rabbitt and Gregory2,Reference Pigeyre, Duhamel and Poulain28–Reference Mulders-Jones, Mitchison and Girosi31) . Age was used as a continuous variable. Education was defined as the highest level of education attained and was categorized as ‘less than high school’, ‘high-school graduate’ and ‘some college/college graduate’. Marital status was a dichotomous variable categorized as ‘currently married or living with partner as married’ and ‘not married’ (i.e. single, separated, divorced and widowed). Household size was assessed for the total household, including children and other family members. Lastly, country of birth was self-reported by participants and was categorized as ‘born in Puerto Rico’, ‘born in the mainland USA’ (in any state of the USA) and ‘born in other country’ (outside the USA).
Statistical analysis
Descriptive statistics include mean and standard deviations for continuous variables and frequencies for categorical variables. EE and covariates were contrasted by food security status using the χ 2 test (for categorical variables) or the t test (for continuous variables). Due to violation of the proportional odds assumption, multinomial logistic regression models were performed to evaluate the association between food insecurity and EE, and the association between perceived stress and EE. We performed mediation analyses using the Baron and Kenny method(Reference Baron and Kenny32), which consists of four steps. For the first step, we evaluated the association between food insecurity and EE using multinomial regression models (adjusting for covariates; path c). The second step examined the association between food insecurity and perceived stress using a linear regression model (because perceived stress was used as a continuous measure) adjusting for covariates (path a). For the third step, we evaluated the association between perceived stress and EE (taking food insecurity into account) using multinomial regression and adjusting for covariates (path b). In the fourth and final step, we examined the association between food insecurity and EE adjusting for perceived stress and all covariates using a multinomial regression model (path c′). The mediated effect (a × b) and the mediated proportion were calculated (a × b/c)(Reference Breen, Karlsson and Holm33,Reference Karlson and Holm34) . A 95 % CI was also calculated for the mediated proportion using the formula for a confidence interval around a proportion: 
$(\widehat p \pm 1\!\cdot\!96\sqrt {{{\widehat p(1 - \widehat p)}/n}})$, where n is the sample size and p is the calculated mediated proportion. Statistical significance was set at P < 0·05. The statistical software package Stata version 14 was used for all analyses.
Results
Sample characteristics are presented by food security status in Table 1. Overall, mean age was 47 years and about half of the sample were women. Approximately 75 % self-identified as Dominicans and the majority were born outside the mainland USA. Mean household size was 3 and 43 % were married or living with a partner. Over a third of the sample experienced food insecurity. A larger proportion of food-insecure individuals had a lower education level than food-secure individuals. In addition, a larger proportion of food-insecure individuals reported High EE than food-secure individuals.
Table 1 Characteristics of the sample by food security status: Latinx individuals (n 580) aged 21–84 years from the Latino Health and Well-being Study, Lawrence, MA, USA, September 2011–May 2013
EE, emotional eating.
Continuous variables (age, household size) are presented as mean and standard deviation; categorical variables are presented as numbers and percentage.
* Data missing for one participant.
Multivariable multinomial models adjusted for age, sex, education, marital status, household size and country of birth showed that food insecurity was associated with almost twice the odds of High EE (Table 2). Food insecurity was also associated with higher perceived stress scores in a multivariable linear regression model adjusted for covariates (Table 2). In turn, perceived stress was associated with almost 10 % higher odds of High EE but not Low EE (Table 2). When perceived stress was added in the main effects model, food insecurity was no longer associated with High EE (Table 2). Estimates from Table 2 were used to calculate the mediated proportion. Perceived stress explained 69·9 (95 % CI 66·2, 73·6) % of the association between food insecurity and High EE.
Table 2 Adjusted associations from meditation analysis for food insecurity, perceived stress and emotional eating (EE) among Latinx individuals (n 580) aged 21–84 years from the Latino Health and Well-being Study, Lawrence, MA, USA, September 2011–May 2013
OR are presented for all models except for food insecurity–perceived stress, for which the β coefficient is presented.
All models are adjusted for age, education, sex, household size, marital status and country of birth.
Discussion
To our knowledge, the present study is the first to examine the association between food insecurity and EE and the mediating role of perceived stress in a sample of US Latinxs. Overall, our results show that food insecurity was associated with EE and that perceived stress mediated a large proportion of this association.
Only three previous studies have examined the association between food insecurity and EE. Our finding of food insecurity being associated with high levels of EE is in line with two of the previous studies conducted with samples of predominantly African-American women(Reference Sharpe, Whitaker and Alia15,Reference Lofton and Connell17) ; and found that food-insecure individuals had higher EE scores than food-secure individuals. The third study, conducted with a diverse sample of US men and women, failed to find an association between food insecurity and EE(Reference Myles, Porter and Johnson16). This discrepancy may be due to differences in sample size and race/ethnicity of study participants as the third study was considerably smaller (n 118(Reference Myles, Porter and Johnson16), v. 202 and 632 in the previous two studies, respectively(Reference Sharpe, Whitaker and Alia15,Reference Lofton and Connell17) ) and had a larger proportion of White participants (57 % African Americans(Reference Myles, Porter and Johnson16), v. 84 and 87 % African Americans, respectively(Reference Sharpe, Whitaker and Alia15,Reference Lofton and Connell17) ). Methodological differences could also account for the difference in findings. Specifically, the approach taken to categorize EE dichotomously may have hindered the ability to detect a difference.
The high prevalence of food insecurity in this sample reflects insufficient income and insufficient benefits among Latino residents in Lawrence. Compared with the overall national estimate(Reference Coleman-Jensen, Rabbitt and Gregory2), the prevalence of food insecurity was approximately three times higher, which shows that food insecurity is a major public health issue in this population. The population in Lawrence, which is comprised largely of Latinos (74 %)(35), experiences socio-economic disadvantage compared with the rest of the State of Massachusetts: median household income is low and poverty is high(35). Thus, all of these factors may increase the risk of food insecurity in the Lawrence population and highlight the need to address this public health issue in this vulnerable group.
Although the mechanisms by which food insecurity may be associated with EE are likely complex, our results suggest that perceived stress mediates a large proportion of this relationship. It is possible that food insecurity is associated with EE due to the emotional distress caused by lack of access to adequate and nutritious meals, or due to the correlation between food insecurity and numerous other stressors associated with low socio-economic status. Food insecurity has been linked to stress(Reference Martin, Maddocks and Chen9,Reference Chung, Kim and Kwak10) , which is a known predictor of EE(Reference Camilleri, Mejean and Kesse-Guyot27,Reference van Strien, Konttinen and Homberg36–Reference Ostrovsky, Swencionis and Wylie-Rosett38) . Thus, EE may serve as a coping mechanism to relieve the emotional distress caused by food insecurity and other stressors related to low socio-economic status.
The fact that food insecurity and EE have both been linked with nutrition-related health conditions such as obesity, pre-diabetes and type 2 diabetes, which are highly prevalent among US Latinx populations(Reference Gooding, Walls and Richmond4,Reference Murillo, Reesor and Scott5,Reference Strings, Ranchod and Laraia39–Reference Hainer, Kunesova and Bellisle42) , points to a need to ameliorate food insecurity in this population. Reducing and eliminating the problem of food insecurity is a national priority(43). At the policy level, there is room for improvement of benefits and eligibility of such benefits, as well as a need for efforts to reduce the racial/ethnic wage gap. In addition, at the community level, improving access to healthy foods may decrease stress and EE by alleviating food insecurity overall. For example, community-level Interventions should incorporate efforts to improve access to healthy foods such as through mobile markets (to decrease transportation barriers) and community gardens and chain supermarkets (to eliminate food deserts)(Reference Evans, Banks and Jennings44,Reference Haynes-Maslow, Auvergne and Mark45) . Lastly, at the individual level, there is a need for tailored interventions targeting stress and EE in food-insecure Latinxs. Although the aforementioned suggestions are sensible, they may not be sufficient to alleviate food insecurity and their efficiency has not been vigorously tested.
The study results should be considered with limitations and strengths in mind. One limitation is the cross-sectional nature of the study, thus causality cannot be determined. Another limitation of our study is the potential existence of selection bias in our cohort. Out of 2783 potentially eligible participants, only 602 were enrolled. Comparison between participants v. non-participants on key demographic characteristics was not feasible as the study was not able to collect data on non-participants. In addition, the generalizability of our findings may be limited to Latinxs of Caribbean decent residing in the north-east of the USA. However, this is also a study strength as Caribbean Latinxs, the largest Latinx group in the north-east USA, experience disparities in both food insecurity(Reference Coleman-Jensen, Rabbitt and Gregory2) and nutrition-related health conditions(Reference Daviglus, Pirzada and Talavera46) but have been under-represented in research. To our knowledge, the present study is the first of its kind to evaluate the relationship between food insecurity, perceived stress and EE in Latinxs.
Conclusion
In conclusion, the present study found that food insecurity was associated with high levels of EE and this association was largely mediated by perceived stress. Identifying modifiable factors associated with nutrition-related health conditions will help to design interventions to reduce health disparities in this vulnerable population. Longitudinal studies are needed to examine the association between food insecurity, perceived stress and EE, and to test interventions that improve access to healthy foods and include stress reduction strategies for food-insecure Latinxs.
Acknowledgements
Acknowledgements: The authors thank the City of Lawrence Mayor’s Health Task Force, Lawrence Senior Center, YWCA of Northeastern Massachusetts, Greater Lawrence Family Health Center, and the University of Massachusetts Medical School colleagues, students and staff, who collaborated in this research. Financial support: Research reported in this manuscript was supported by the National Institute of Mental Health (grant number R01 MH085653). In addition, S.C.L. and M.C.R. received funding from the National Institute of Minority Health and Health Disparities (grant number P60 MD006912) and Centers for Disease Control and Prevention (grant number U48 DP005031); A.L.-C. received funding from the University of Massachusetts Center for Clinical and Translational Sciences (grant number TL1 TR001454); and G.B. received funding from a T32 pre-doctoral fellowship of the National Heart, Lung, and Blood Institute at the University of Massachusetts Medical School (grant number 5T32HL120823-03). The funding agencies had no role in the design, analysis or writing of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Conflict of interest: The authors declare that they have no conflict of interests. Authorship: A.L.-C. conceptualized and carried out the analysis and drafted the manuscript. C.F. contributed to the statistical analysis. G.B. contributed to the conceptualization of the analysis and revised the manuscript. S.C.L. and M.C.R. were responsible for the design of the parent study and its data collection. S.C.L. and M.C.R. also contributed to the conceptualization of the analysis and revised the manuscript critically. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving research study participants were approved by the Institutional Review Board at the University of Massachusetts Medical School. Written informed consent was obtained from all participants.
