Risk and Protective Factors Associated with Adolescent Food Insecurity in Baltimore

Abstract

While emerging evidence is highlighting a growing problem of food insecurity among adolescents in disadvantaged neighborhoods, very little is known about the factors that may either protect or place adolescents at higher risk for food insecurity. The primary objective for this analysis, therefore, was to examine the associations between individual-, family-, and neighborhood-level risks and protective factors and food insecurity among 452 adolescents in Baltimore, Maryland. Results show that nearly 30% of our sample were food insecure (29.4%). Food insecure youth were more likely to be unstably housed (OR 5.17, 1.24–21.62), live in larger households (OR 1.14, 1.08–1.20), and perceive their neighborhoods unsafe (OR 2.37, 1.47–3.83). Protective factors included perceiving both male and female adult support (OR 0.55 and 0.47, respectively), having a higher sense of community belonging (OR 0.91, 0.32–0.95) and having positive perceptions of their neighborhood’s physical environment (OR 0.93, 0.88–0.98). These results suggest that strengthening family and neighborhood relations and resources may promote the health of adolescents in disadvantaged urban areas.

Introduction

Food insecurity, defined as limited or uncertain access to adequate food, is a widespread, but often hidden problem among adolescents in low-income neighborhoods in the USA [1–3]. Food insecurity among young people is particularly concerning, as adolescence is a critical developmental stage [4]. The onset of puberty to adulthood involves hormonal changes that stimulate structural and functional shifts in the adolescent brain [5]. During this period of immense change, proper nutrition and caloric intake are particularly important [6]. However, there is growing evidence that shows food-insecure youth not only have poor nutrition and reduced levels of physical activity but are much more likely to have impaired cognitive development and mood, anxiety, and behavior disorders [7, 8]. Recent qualitative studies have also shown that food insecure adolescents often have to engage in high risk behaviors to mitigate the effects of food insecurity [1–3]. For example, in Baltimore, adolescent girls reported that it was easier for them to trade sex for money to purchase food than it was for them to visit food charity centers. For boys, it was common for them to sell drugs or shop lift to get money for food [1].

While the vast majority of research has examined food insecurity among parents and young children, very little research has been conducted among adolescents. Part of the reason for this is that most national surveys only collect data from the adult head of household. For example, both the annual USDA report on household food insecurity and the National Health and Nutrition Examination Study collect data at the household level rather than at the individual level. Many adolescents in disadvantaged neighborhoods, however, often act as the “head of household” by taking on responsibilities such as paying the household bills or caring for younger siblings; adolescents are also more likely than younger children within a family to be unstably housed or homeless [9]. As a result, the experience and means of coping with food insecurity may be quite different, and less likely to be recorded, among adolescents compared with younger children or parents within a household.

One relatively new area of research that may be particularly relevant for adolescents is the influence of the family and neighborhood contexts on food insecurity. Recently, Jackson and colleagues found three neighborhood-level factors independently associated with household food insecurity: physical disorder, low social capital, and exposure to violence [10]. Yet, it was the confluence of both low social capital and violence/perceived danger in the neighborhood that resulted in the strongest association with household food insecurity. Indeed, sociologists have long recognized the importance of neighborhoods as structural conditions that shape individual lives and opportunities [11, 12]. In public health, it has also been argued that the physical and social attributes of a neighborhood are important for understanding the distribution of health outcomes [13]. Given that food insecurity is a measure of food access in the context of financial constraints, it makes theoretical sense that the neighborhood would have a particularly large influential role on food insecurity. Yet, research has also shown that certain protective factors, particularly within a family, can attenuate the link between neighborhood risk factors and adolescent health outcomes. For example, parental support and connection have been shown to be protective against substance use and misuse, suicidal ideation or attempts, and high-risk sexual behavior among adolescents [14–17]. Whether these same processes are also protective against adolescent food insecurity remains to be seen.

To fill in this knowledge gap, we use data from the Well-Being of Adolescents in Vulnerable Environments (WAVE) study, which was a five-city global study of urban adolescents in disadvantaged neighborhoods. For this paper, we concentrate on adolescents in Baltimore as it was the primary site where food insecurity was explicitly measured. Specifically, the main objective of the present analyses was to examine the associations between individual-, family-, and neighborhood-level risks and protective factors and food insecurity among adolescents in Baltimore, Maryland.

Methods

Data

The WAVE study used respondent-driving sampling (RDS) to conduct a cross-sectional survey of 15–19-year-olds in five economically disadvantaged urban sites. RDS is a snowball sampling technique that relies on peer to peer recruitment and statistical adjustments that approximate random sampling with the advantage of accessing typically hidden populations [18]. Details of the recruitment and sample characteristics of the overall WAVE study are available in Decker et al. [19] The data presented here are from Baltimore only and were collected between March and June 2013. Inclusion criteria were youth aged 15–19 living or spending a majority of their time in five zip codes that comprise Historic East Baltimore, a geography that accounts for approximately 20% of Baltimore’s total population. All eligible participants completed a survey using Audio Computer-Assisted Self-Interview (ACASI) software. The Johns Hopkins Bloomberg School of Public Health Institutional Review Board approved all study protocols.

Measures

Dependent Variable

1. Food insecurity: Food insecurity was defined as a dichotomous variable indicating whether a participant was food secure or food insecure based on whether respondents answered Yes or No to the following three questions based on the previous 30 days: Did you eat small portions of food than you would have liked to eat? Was there a time in which there was no food in your home because of lack of resources to purchase food? Was there a time in which you or some other members of your household slept without eating because there was no food to eat? These questions were adapted from the USAID Household Food Insecurity Access Scale [20]. Participants who responded Yes to two or three questions were categorized as having high food insecurity, while those who responded No to one or no questions were categorized as having low food insecurity.

Individual Demographic Factors

1. Age and gender: Participants were categorized into two age groups: 15–16-year-olds and 17–19-year-olds based on the latter group being more likely to have completed schooling and live independently. In addition, respondents were categorized as male or female based on their reported gender.

2. Perceived relative economic status: Participants were asked how their family or household compares with other families in their neighborhood in terms of income and were classified into one three-response categories based on whether they replied that they were better off, about the same, or worse off than most families using a categorical variable.

3. Out of school: A dichotomous variable representing whether the participant prematurely left school was created based on age, reported school enrollment, and highest level of education. Participants who were age 15–16 and reported that they were not currently enrolled in school, and participants who were age 17–19, reported that they were not currently enrolled in school, and reported their highest level of education as less than graduating from high school were recorded as being out of school. Participants who were age 15–16 and reported being enrolled in school, participants who were age 17–19 and reported being enrolled in school, and participants who were age 17–19, not currently enrolled in school, and reported their highest level of education as high school graduation or higher were recorded as not being out of school.

Household Factors

1. Unstably housed: Unstably housed was a dichotomous variable indicating whether a participant did not have a regular place to stay OR stayed an average of 3–4 nights a week or less in their regular place of stay over the last 30 days; AND stayed overnight in more than one place in the last 7 days.

2. Household size: Household size was a continuous variable based participants’ response of how many people live in their household, including themselves. The majority of responses ranged from 1 to 17, with one response of 56. The response of 56 was treated as missing due to it being an extreme outlier.

3. Adult support in household: Support from adults presently in participants’ households was represented by two variables, adult male support and adult female support. Participants were asked to respond to the following statements about adults presently in their home: “There is a (fe)male adult who believes you will be successful.” “There is a (fe)male adult who you can turn to when you have a problem.” “There is a (fe)male adult who listens to you when you have something to say.” The four response options ranged from never true (0) to always true (3). Responses were summed across the three questions, presenting a total maximum possible score of 9. For both male and female household support, responses of 5 or higher were considered high support, whereas responses of less than 5 were considered low support.

Neighborhood Factors

1. Physical environment: An eight-item, four-response category continuous scale asking about attributes of participants’ neighborhoods, such as whether abandoned or rundown buildings are common, whether there is a lot of trash, and whether there is poor lighting, was used to measure the quality of participants’ physical environments. The items used were informed by the Extent of Neighborhood Problems Scale, a similar scale used to measure physical environment [21]. Scores ranged from 0 to 24, with a higher score denoting poorer quality physical environment. For individuals with missing scale items, scale totals were imputed using multiple imputation. Cronbach’s alpha was 0.88.

2. Neighborhood safety: Perceived neighborhood safety was measured as a dichotomous variable based on whether, in terms of violence, participants reported that they consider the community they live in “very safe” or “safe,” versus “unsafe” or “very unsafe.” Responses were collapsed into very safe/safe versus unsafe/very unsafe.

3. Community trust: Participants’ trust in their communities was a continuous variable measured based on their responses of how much they trust their family, neighbors, and friends [22]. The four response options ranged from no trust at all (0) to a lot of trust (3). Responses were summed to produce a community trust score which ranged from 0 to 9; Cronbach’s alpha was 0.76.

4. Community belonging: Participants’ sense of connection to their communities was a continuous variable measured using a three-question, four-response category scale that asked if participants’ feel connected to their neighborhoods, if they know most of the people, and if they feel part of a community in their neighborhood. Scores ranged from 0 to 9, with a Cronbach’s alpha of 0.80. Higher scores denoted higher sense of belonging.

Analysis

Descriptive statistics of the study sample as well as bivariate analyses of food insecurity by each covariate were produced. The unadjusted odds of food insecurity for each covariate was estimated, followed by three multivariable logistic regression models: Model 1 contained individual demographic variables (gender, age group, perceived relative economic status, and being out of school), Model 2 contained individual and household-level variables (being unstably housed, household size, adult male support, adult female support), and Model 3 contained individual, household, and neighborhood-level variables (perceived physical environment score, perception of neighborhood safety, community trust, and community belonging). We conducted analyses using Stata’s complex design procedures to accommodate the non-independence of observations, and weights were generated via the RDSII estimator to account for the bias inherent in the network size of peer recruiters and were used in the analyses presented in Tables 2, 3, and 4. A multi-collinearity check of the covariates produced variance inflation factors all below 1.5, suggesting no significant collinearity.

Table 2.

Sample characteristics by food insecurity status (n = 452)*

Variables	Low food insecurity	High food insecurity	p value
Individual-level variables
Gender (%)			0.803
Female	47.9%	47.3%
Male	52.1%	52.8%
Age group (%)			0.019
15–16	57.4%	49.6%
17–19	42.7%	50.4%
Perceived relative economic status (%)			0.068
Better off than most	41.0%	27.1%
About the same as most	53.2%	61.2%
Worse off than most	5.8%	11.7%
Out of school (%)	4.6%	5.0%	0.540
Household-level variables
Unstably housed (%)	0.7%	5.9%	<0.001
Household size (mean)	4.5	5.6	<0.001
High adult male support (%)	70.4%	50.1%	<0.001
High adult female support (%)	87.3%	77.8%	0.032
Neighborhood-level variables
Perceived neighborhood as unsafe (%)	29.4%	60.5%	<0.001
Physical environment (mean)	13.5	9.3	<0.001
Community trust (mean)	4.9	3.9	0.009
Community belonging (mean)	3.6	3.9	0.075

*Gender, age group, being out of school, being unstably housed, adult male support, adult female support, and perceived neighborhood safety are dichotomous variables. Perceived relative economic status is a categorical variable. Household size, physical environment score, community trust, and community belonging are continuous variables

Table 3.

Univariate association of food insecurity with each variable*

Variables	OR of food insecurity	95% CI	p value
Individual-level variables
Gender (n = 452)
Male	1.00 (Ref)	--	--
Female	1.03	0.80–1.32	0.803
Age group (n = 452)
15–16	1.00 (Ref)	--	--
17–19	1.37	1.07–1.75	0.019
Perceived relative economic status (n = 443)
Better off than most	1.00 (Ref)	--	--
About the same as most	1.74	1.01–3.00	0.047
Worse off than most	3.07	0.80–11.82	0.090
Out of school (n = 451)	1.08	0.82–1.41	0.540
Household-level variables
Unstably housed (n = 452)	8.43	6.53–10.87	<0.001
Household size (n = 441)	1.17	1.13–1.21	<0.001
High adult male support (n = 445)	0.42	0.38–0.47	<0.001
High adult female support (n = 445)	0.51	0.28–0.93	0.033
Neighborhood-level variables
Perceived neighborhood as unsafe (n = 449)	3.68	2.39–5.67	<0.001
Physical environment (n = 442)	0.89	0.87–0.92	<0.001
Community trust (n = 452)	0.87	0.79–0.95	0.009
Community belonging (n = 446)	1.04	1.00–1.08	0.075

*Gender, age group, being out of school, being unstably housed, adult male support, adult female support, and perceived neighborhood safety are dichotomous variables. Perceived relative economic status is a categorical variable. Household size, physical environment score, community trust, and community belonging are continuous variables

Table 4.

Multivariable association of food insecurity with individual, household, and neighborhood factors (n = 452)*

Covariates	Model 1			Model 2			Model 3
	OR	95% CI	p value	OR	95% CI	p value	OR	95% CI	p value
Individual-level variables
Gender
Male	1.00 (Ref)	--	--	1.00 (Ref)	--	--	1.00 (Ref)	--	--
Female	0.95	0.76–1.19	0.605	0.82	0.69–0.99	0.039	0.92	0.74–1.13	0.328
Age group
15–16	1.00 (Ref)	--	--	1.00 (Ref)	--	--	1.00 (Ref)	--	--
17–19	1.27	0.92–1.75	0.117	1.33	1.05–1.69	0.026	1.10	0.84–1.43	0.414
Perceived relative economic status
Better off than most	1.00 (Ref)	--	--	1.00 (Ref)	--	--	1.00 (Ref)	--	--
About the same as most	1.70	0.90–3.21	0.086	1.63	0.90–2.96	0.088	1.40	0.67–2.89	0.299
Worse off than most	3.04	0.70–13.16	0.111	2.20	0.42–11.49	0.281	1.78	0.25–12.66	0.493
Out of school	0.80	0.43–1.50	0.372	1.27	0.79–2.06	0.241	1.31	0.80–2.14	0.206
Household-level variables
Unstably housed				5.42	1.81–16.20	0.012	5.17	1.24–21.62	0.032
Household size				1.17	1.13–1.21	0.001	1.14	1.08–1.20	0.004
High adult male support				0.55	0.37–0.81	0.010	0.70	0.52–0.93	0.025
High adult female support				0.47	0.26–0.41	0.025	0.56	0.26–1.20	0.108
Neighborhood-level variables
Perceived neighborhood as unsafe							2.37	1.47–3.83	0.005
Physical environment							0.93	0.88–0.98	0.020
Community trust							0.93	0.81–1.06	0.219
Community belonging							0.91	0.32–0.95	0.004

*Gender, age group, being out of school, being unstably housed, adult male support, adult female support, and perceived neighborhood safety are dichotomous variables. Perceived relative economic status is a categorical variable. Household size, physical environment score, community trust, and community belonging are continuous variables

Participants in the 456-person sample who were missing any of the food insecurity outcome items were dropped from the analysis (4 participants). Eight of the twelve covariates had small amounts of missing data in the remaining 452-person sample, namely, relative economic status (9 values), household size (11 values) perceived neighborhood safety (3 values), whether respondents were out of school (1 value), adult male support (7 values), adult female support (7 values), physical environment score (10 values), and community belonging score (6 values). Data were assumed to be missing at random based on associations between missingness and the remaining covariates for each variable. Prior to multivariable regression analyses, missing values were multiply imputed using chained equations that produced 20 imputations. Relative economic status was imputed using unweighted multinomial logistic regression imputation, and perceived neighborhood safety, whether respondents were out of school, and adult male and female support were imputed using unweighted logistic regression. Household size, physical environment score, and community belonging were imputed using unweighted predictive mean matching based on the three nearest neighbors, which provided better fit with the original data than imputation using linear regression. The fit of imputed data was assessed using the midiagplots command in Stata (version 15.0, College Station, TX USA) to compare imputed and observed data using cumulative distribution functions for continuous variables and tables of proportions for dichotomous and categorical variables [23].

Results

Sample Characteristics

There were a total of 452 adolescents in the analytical sample, with slightly more females than males (52% vs. 48%) and a higher proportion of younger vs. older adolescents (55% vs. 45%, Table 1). The majority (55%) perceived their relative wealth was about the same as most in their neighborhood, and only about 6% of the sample was out of school. Very few were unstably housed (2%), and the mean household size was 4 individuals. A higher percentage of adolescents perceived there to be high adult female support in the households (85%) compared with high male support in the households (65%). Over one third of the sample perceived their neighborhood was unsafe (37%), and the mean neighborhood physical environment score was 13, with a range between 0 and 24. The average neighborhood belonging score was relatively low (3, from a range between 0 and 9), and the average community trust score reported by adolescents was 4.7, from a range of 0–9. Finally, in regard to food insecurity status, over one quarter of the sample had high food insecurity (24.9%).

Table 1.

Sample characteristics (n = 452)∗†

Variables	Proportion(95% CI)	Mean/Median(SD/IQR)∗	Range
Individual-level demographic variables
Gender
Female	52.2% (41.3–63.0)	--	--
Male	47.8% (37.0–58.7)	--	--
Age
15–16	55.4% (41.3–68.7)	--	--
17–19	44.6% (31.3–58.7)	--	--
Perceived relative economic status
Better off than most	37.5% (30.9–44.7)	--	--
About the same as most	55.2% (48.8–61.5)	--	--
Worse off than most	7.2% (4.3–11.9)	--	--
Out of school	4.7% (3.5–6.4)	--	--
Household-level variables
Unstably housed	2.0% (1.2–3.3)	--	--
Household size	--	4 (3–6)	1–17
High adult male support	65.5% (61.3–69.4)	--	--
High adult female support	85.0% (81.6–87.8)	--	--
Neighborhood-level variables
Perceived neighborhood as unsafe	37.2% (57.9–67.4)	--	--
Physical environment	--	13 (8–17)	0–24
Community trust	--	4.7 (2.7)	0–9
Community belonging	--	3 (1–6)	0–9
Outcomes
High food insecurity	24.9% (22.0–28.0)	--	--
Ate smaller portions than would like in past 30 days	48.4% (45.6–51.2)	--	--
Experienced a time with no food in home in past 30 days	27.4% (24.4–30.7)	--	--
Self or family slept without eating in past 30 days	22.5% (19.9–25.4)	--	--

*Gender, age group, being out of school, being unstably housed, adult male support, adult female support, and perceived neighborhood safety are dichotomous variables. Perceived relative economic status is a categorical variable. Household size, physical environment score, community trust, and community belonging are continuous variables.

†Mean and standard deviation are provided for continuous variables that are normally distributed, and median and interquartile range are provided for continuous variables that are not normally distributed. Normality of distribution was determined using Shapiro–Wilk test. Community trust was determined to be normally distributed, and household size, physical environment score, and community belonging were determined to be non-normally distributed

Food Insecurity Status by Sociodemographic Characteristics

In Table 2, we present the bivariate analyses of sociodemographic characteristics between adolescents of low and high food insecurity status. Adolescents with high food insecurity were more often male and perceived themselves to be of lower relative socioeconomic status compared with their neighbors. Compared with adolescents with low food insecurity, those with high food insecurity were more commonly unstably housed, resided in larger households, and had lower levels of adult male and female support in their household. At the neighborhood level, adolescents with high food insecurity more often perceived their neighborhoods as unsafe, reported less positive perceptions of their physical environment, and a lower level of community trust (Table 2).

Regression Analysis of Individual, Family, and Neighborhood Factors with Food Insecurity

Table 3 presents the results of unadjusted logistic regression models run separately for each covariate. Based on these univariate analyses, being in an older age group and perceiving one’s household to be about the same or of lower socioeconomic status relative to one’s peers were associated with higher odds of high food insecurity among adolescents. All household-level variables were significantly associated with food insecurity, with being unstably housed and being in a larger household associated with higher odds of high food insecurity and having high adult male or adult female support associated with lower odds of high food insecurity. All neighborhood-level variables except for community belonging were also significantly associated with food insecurity; perceiving one’s neighborhood as unsafe was associated with higher odds of food insecurity, and higher physical environment scores and community trust were associated with lower odds of food insecurity.

The multivariate analyses presented in Table 4 presents findings from three multivariate logistic regression models. Model 1 contains individual demographic variables, Model 2 contains individual and household-level variables, and Model 3 contains individual, household, and neighborhood-level variables. Based on results from Model 1, no individual demographic factors were statistically significantly associated with higher odds of food insecurity based on p values of less than 0.05, though based on a p value of 0.10, perceiving oneself as being about the same socioeconomic status as other households in one’s neighborhood was associated with higher odds of food insecurity. Collectively, this multivariate model appears to explain the significant association of older age group and worse perceived relative economic status with food insecurity observed in univariate models, as older adolescents were more likely to report having lower relative economic status (Table 3). Compared with younger adolescents, a smaller proportion of older adolescents perceived themselves to be better off (27.6% vs 45.7%), and a higher proportion perceived themselves to be worse off than their neighbors (8.2% vs 6.5%, p = 0.0069, data not shown).

In Model 2, which includes household-level variables, all household-level variables were significantly associated with food insecurity at the p < 0.05 level, as was identifying as female and being in an older age group. Being unstably housed was associated with higher odds of food insecurity, as was increasing household size. Having high adult male support and high adult female support were both associated with lower odds of food insecurity.

After introducing neighborhood-level variables in Model 3, all household-level variables remained significantly associated with food insecurity to a similar extent as in Model 2. In addition, the neighborhood-level variables physical environment score and community belonging were associated with statistically significantly lower odds of food insecurity at the p < 0.05 level, while perceiving one’s neighborhood to be unsafe was associated with significantly higher odds of food insecurity. Neighborhood-level factors appear to explain the significant association (at the p < 0.05 level) of gender and being in an older age group (17–19) with food insecurity that was observed in Model 2. Older adolescents had lower physical environment scores than younger adolescents (11.53 (95% CI 10.60–12.46) vs. 13.13 (95% CI 12.72–13.55), p = 0.014)). They were also more likely to perceive their neighborhoods as unsafe (45.23% vs 30.61%, p < 0.0308) and to have lower community trust (4.36 (95% CI 4.06–4.67) vs. 4.89 (95% CI 4.54–5.24), p = 0.029).

Discussion

The results of this study confirm the importance of examining food insecurity among adolescents: nearly a third of our sample was food insecure. This high rate of food insecurity is even more concerning given that we collected this data while school was in session, when food insecurity is generally lower than during the summer months when school is out of session [24]. Comparing this proportion to the national rate, in 2018, the prevalence of food insecurity among households with children was 13.8%, and among households with incomes below the poverty line, the prevalence was at 25% [25].

The study also highlights the importance of both household and neighborhood contextual factors in relation to adolescent food insecurity. At the household level, male and female adult support was protective against food insecurity. While it may be difficult to determine the temporal order between family support or food insecurity given the cross-sectional nature of this study, previous research on family connectedness has shown that adolescents with strong relationships with parents and caregivers are protected from a variety of health outcomes [14–17]..

One theory that may offer some explanation for our results is the Family Stress Theory [26, 27]. Accordingly, in low-income households, parents struggling to make ends meet because of lack of income, underemployment, and low-wage jobs experience heightened stress that leads to changes and emotions that hurt their parenting and capacity to meet the needs of their children. McCurdy and colleagues suggest that the link between parenting and child food insecurity in such households primarily operates through maternal depression, which reduces the use of active food acquisition strategies required to meet the nutritional needs of their children [28]. Although this theory has only been used to explain the links between parenting and younger children’s outcomes, it does offer some explanation as to why adolescents who lacked either male or female support in the household were more likely to be food insecure compared with those who did [29, 30]. The challenge with using this theory in relation to our findings, however, is that parenting behaviors typically exert greater influence over young children than older children, as it has been assumed that adolescents have more independent access to food both within and outside the home [31]. More research, therefore, is needed to understand how parental support protects adolescents from food insecurity, as well as how the lack of parental support increases the risk of food insecurity among adolescents.

Our study also found that unstably housed youth had significantly higher odds of food insecurity. While it still is not clear whether these young people were moving to different locations with their parents or on their own, previous research has found similar linkages with food insecurity among families and unaccompanied youth. Cutts et al., for example, found that crowded housing and multiple moves among families with young children were associated with increased risk of food insecurity among all the children in the family [32]. Likewise, researchers who studied food insecurity among unaccompanied youth who had experienced homelessness and/or unstable housing in San Francisco found that food was one of their greatest needs [33]. In fact, the 2019 San Francisco Point in Time Youth Homeless Count Survey revealed that 67% of homeless youth reported needing food, while the proportion benefitting from government food provision programs decreased significantly from 2015, from 49 to 32% [34]. Other studies have found food insecurity rates among homeless and unstably housed youth ranging from 28 to 85% of study samples [35–37].

Turning to the association between neighborhood factors and adolescent food insecurity, an emerging body of research has documented the protective effects of neighborhood social attributes on food insecurity. Results of these studies are fairly consistent, regardless of the measure of social control and support. In general, higher community-level social support is linked to lower odds of experiencing food insecurity [38]. In our study, we found that perceived safety, the perception of the physical environment, and sense of community belonging all protected adolescents from food insecurity. Especially in regard to community belonging, evidence suggests that neighborhood support and belonging may help to shield individuals and families who are facing economic hardships, and therefore help to curb food insecurity [39]. While the bulk of this research has focused on mothers of small children, our study shows that even for adolescents, neighborhood belonging is important. Other studies suggest that the reason for this may be through neighbors sharing resources such as food or strategies to mitigate food insecurity [40].

In terms of perceived safety, a number of existing studies suggest that our observed association between perceived safety and food insecurity may primarily operate through psychological distress and mental health [10, 40]. Knowles and colleagues, for example, found that anxiety and depression frequently co-occur with food insecurity and that these same measures are also more likely among residents who live in violent neighborhoods [41]. However, given that many of these studies are cross-sectional in design, it is not yet clear whether food insecurity causes psychological distress, which, in turn, leads to violence, or whether the reverse is true. In our study in Baltimore, violence—including violence among youth—is highly prevalent [42]. While the association between perceived safety and food insecurity may operate through mental health outcomes, it may also be possible that accessing food in a dangerous neighborhood may, in and of itself, contribute to food insecurity. Research on obesity has pointed towards residents in neighborhoods with high violence leaving their homes less due to fear, which may also extend to our findings [43]. Indeed, in our qualitative study, adolescents reported feeling often afraid in trying to seek food in their neighborhoods [1].

Neighborhood-level factors appeared to be prominent in the relationships observed between older youth and food insecurity in our study. Older adolescents reported less positive physical environment scores, and a higher proportion reporting feeling unsafe in their neighborhoods, both of which are associated with food insecurity. As young people enter later adolescence and adulthood, they are more likely to experience disconnection, formally defined as being out of school and unemployed, which is associated with compromised health outcomes including food insecurity. Disconnection is strongly related to neighborhood environment, with up to 23% of young people experiencing disconnection in the city’s most deprived and impoverished neighborhoods that experience higher crime rates [44]. These findings point towards the need to understand how adolescents’ experiences of food insecurity evolve as they age and gain increased independence.

This study has a number of important limitations. First, although respondent-driven sampling was a specific technique used to recruit adolescents from diverse social backgrounds, the sample in this study is not representative of the general adolescent population in Baltimore and RDS samples often have confidence intervals that are underestimates of population variance [45]. Secondly, all of the measures of neighborhood-level factors are obtained from the same survey respondents used to examine food insecurity. As a result, same-source bias could have occurred, which can explain why adolescents who were food insecure were more likely report lower perceptions of community belonging, safety, and of the physical environment. Given this constraint, and also that this is a cross-sectional survey, causal pathways cannot be determined. Finally, federal nutrition assistance program participation was not measured, despite research that has shown SNAP (Supplemental Nutrition Assistance Program and WIC (Women’s Infants, and Children) Program) to be associated with reduced food insecurity among adults and young children [46, 47].

Implications for Research and Practice

Despite these limitations, there are a number of implications that can be drawn from the results of this study. In particular, this study demonstrates that a significant proportion of low-income adolescents experience food insecurity in disadvantaged neighborhoods, but because adolescents have largely been ignored, it is difficult to know how widespread this issue might be among adolescents in other poor settings in the USA. It also highlights the importance of social ties, whether they are with parents or neighbors, as protective factors against food insecurity, as well as the risks associated with perceptions about safety and the physical environment of a neighborhood. This is particularly noteworthy because these latter factors may hold direct implications for where certain food access programs or services should be located in a given urban area.

Current types of food assistance programming for adolescents primarily focus on food provision, access to food, and healthy meal preparation. While each of these strategies is important, our study suggests that they may not be sufficient. Instead, our findings show the importance of addressing contextual factors, such as housing stability, family support, as well as neighborhood safety and organization. Indeed, qualitative findings from our previous study further support these conclusions [1], as many adolescents felt that programs which only provided food or access to healthy food was too stigmatizing. Combining food provision with other activities, such as those which strengthen both family and neighborhood relations, may offer the most success. Several young people, in fact, recommended implementing more community gardens, because they not only provided food but also enabled community members to come together. Given that adolescents often have very different neighborhood experiences compared with adults, engaging them in these efforts will be especially important. Young people know best where and how programs for youth can be implemented and having them help to design such programs can also help to ensure their success and sustainability.