Household Food Insecurity across Childhood and Weight Change Attempts in Early Adolescence: Findings from a Nationally Representative Cohort of United States Youth

Abstract

Background

Youth with food insecurity may have a disproportionately high risk of experiencing body image concerns and unhealthy weight control behaviors, but these associations remain poorly understood. Understanding how food insecurity across childhood relates to adolescent weight change attempts may inform important strategies for addressing body image concerns and preventing risky weight control behaviors among youth.

Objectives

This study aims to understand how food insecurity across childhood relates to weight change attempts in early adolescence.

Methods

Children (n = 7115) participated in the Early Childhood Longitudinal Study, Kindergarten Class of 1998–1999, a nationally representative cohort of United States youth. Parents reported household food insecurity (exposure) at 4 timepoints: when children were in kindergarten, grade 3, grade 5, and grade 8. In grade 8, children self-reported trying to lose or gain weight (outcomes). Associations were tested using adjusted Poisson regression models.

Results

Compared with children who remained food secure from grade 5 to grade 8, those who recently became food insecure were more likely to attempt weight loss in grade 8 [prevalence ratio (PR) = 1.25; 95% confidence interval (CI): 1.03, 1.52]. Children with food insecurity during 3–4 (compared with 0) total timepoints were also more likely to attempt weight loss in grade 8 (PR = 1.24; 95% CI: 1.04, 1.47). Associations with weight gain attempts were statistically nonsignificant.

Conclusions

Both transient and persistent experiences of household food insecurity during childhood were associated with a greater likelihood of attempting weight loss in early adolescence. These findings emphasize the need for long-lasting food insecurity prevention strategies and the development of programs to promote healthy body image and eating patterns among children with food insecurity.

Introduction

Food insecurity (FI), or inadequate access to healthy food due to financial constraints, is highly prevalent in the United States. In 2023, 17.9% of United States households with children had FI [1]. Experiencing FI during childhood might lead to poor body image and dangerous weight control behaviors. Several prior studies have found that FI is cross-sectionally associated with more body weight/shape concerns and body image dissatisfaction in children and adolescents [[2], [3], [4]], including more drive for thinness [2]. FI was also associated with more muscle dysmorphia, or problematic preoccupation with muscle mass, in a study of adolescents and young adults [5]. A small but growing literature is also beginning to recognize an association between FI and more eating disorder behaviors in children and adolescents [6]. However, additional research is needed to understand these findings, including how FI relates to intentional weight change attempts among youth.

Weight change attempts, including intentional attempts to lose weight or gain weight, are a public health concern among children and adolescents. In studies of young adolescents aged 10–15 y, between 25% and 31% of boys and 31% and 51% of girls reported trying to lose weight [[7], [8], [9]], and in a nationally representative study of United States high schoolers, 30% of boys and 7% of girls reported trying to gain weight [10]. Both weight loss and weight gain attempts are highly correlated with body image concerns and low self-esteem [9,[11], [12], [13]]. Furthermore, attempting weight loss in adolescence is strongly associated with greater use of risky weight loss behaviors, like skipping meals and self-induced vomiting, as well as the development of eating disorders [8,9,14,15]. Trying to gain weight or muscle has also been associated with potentially harmful consequences, like greater use of anabolic steroids and greater odds of disordered eating in young adulthood [9,11].

Understanding how FI relates to weight change attempts may inform important strategies for addressing body image concerns and preventing risky weight control behaviors in youth. However, only a few prior studies have investigated how FI is related to weight change attempts among children or adolescents: a few cross-sectional analyses, including some large United States-based studies, found that FI was associated with more weight loss attempts, dieting attempts, and/or unhealthy weight control behaviors (e.g., skipping meals, using laxatives) among youth [2,[16], [17], [18]]. On the other hand, a cross-sectional study among adolescents from Trinidad found that FI was associated with a greater likelihood of attempting to gain weight and was not associated with attempting to lose weight [19]. Longitudinal analyses are lacking, but 2 longitudinal studies examining adolescent FI found no association with unhealthy weight control behaviors in adulthood [18,20].

In summary, previous research is limited, and most prior studies on FI and weight change attempts among youth have been cross-sectional. To the best of our knowledge, previous longitudinal studies were not able to look at exposure to FI before adolescence, and no previous studies have investigated how FI measured at multiple timepoints across childhood is related to weight change attempts in adolescence. Finally, although there are gender differences in the prevalence of weight change attempts, as well as evidence that FI is associated with more weight gain among girls but not boys [[21], [22], [23]], it remains unclear whether gender modifies associations between FI and weight change attempts. Therefore, we aimed to examine how FI across childhood is associated with weight loss and weight gain attempts in early adolescence, and whether these associations are modified by gender. We addressed these research gaps using longitudinal data from a nationally representative cohort of United States children followed from kindergarten through grade 8.

Methods

Participants

This study used publicly available data from the Early Childhood Longitudinal Study, Kindergarten Class of 1998–1999 (ECLS-K98). The ECLS-K98, organized by the National Center for Education Statistics, is a nationally representative sample of United States students who started kindergarten in 1998 [24]. Briefly, children were recruited in kindergarten using a multistage probability sample design. Approximately 21,000 children were included in the baseline sample, and over 9000 students participated in the final follow-up during the 2006–2007 school year, when most of the children were in grade 8. This study combines data across 4 timepoints: when children were in kindergarten (baseline, 1998–1999), grade 3 (2001–2002), grade 5 (2003–2004), and grade 8 (2006–2007). The grade 8 data are representative of the cohort of United States children who started kindergarten in 1998 and are estimated to represent 80% of United States eighth graders in 2006–2007 [24].

Measures

Past year FI was assessed at each of the 4 timepoints using the 18-item United States Household Food Security Survey Module [25]. Parents responded to survey questions during telephone interviews. Responses were scored from 0 to 18 and were grouped into 2 categories: household food security (score 0–2) and household FI (score ≥3) [25]. Consistent with previous research [22,23], we created a 4-category variable for the recent food security transition from grade 5 to grade 8. The 4 categories were: remained food secure, became food secure, became food insecure, and remained food insecure from grade 5 to grade 8. Children without complete data for FI in grade 5 and grade 8 were considered missing for this variable. Also consistent with previous work [26], a cumulative FI measure was calculated as the total number of timepoints during which a child was in a food-insecure household (categorized as 0, 1, 2, or 3–4 timepoints). To preserve statistical power, FI at 3 or 4 timepoints was combined into a single category. Children without complete data for FI at all 4 timepoints were considered missing for this cumulative measure. Recent FI change (from grade 5 to 8) and the cumulative number of timepoints with FI served as the primary exposures in this study.

Children self-reported weight change attempts in grade 8. Via paper-and-pencil questionnaire, they answered the closed-ended multiple-choice question, “Are you trying to do any of the following about your weight?” Response options included: “Lose weight,” “Gain weight,” “Stay the same weight,” or “I am not trying to do anything about my weight.” We examined trying to lose weight (compared with not trying to lose weight) and trying to gain weight (compared with not trying to gain weight) as separate, dichotomous outcomes.

Child age at outcome assessment (completion of the grade 8 child questionnaire), child gender, and child race/ethnicity were obtained primarily from parent interviews. Socioeconomic status (SES) at each timepoint was based on parent report. For each timepoint, the ECLS-K98 includes a composite SES score for each child that was created based on the parent/guardian’s education level, parent/guardian’s occupational prestige, and total household income [24]. When children had 2 parents/guardians, both parent/guardian education levels and occupations were used. Higher composite SES scores indicate higher SES.

BMI (in kg/m²) in grade 8 was calculated from height and weight, which were measured in duplicate according to standard procedures [24]. We converted BMI values to BMI-for-age percentiles and Z-scores according to age and sex based on the Centers for Disease Control (CDC) 2000 Growth Charts for children aged 2–20 y old [27]. The ECLS-K98 dataset reports age in 5 categories (as shown in Table 1), so the midpoint of the relevant category was used to approximate the exact age of each child. Children were then classified into weight status categories using the CDC’s proposed cut-points: “underweight” (BMI-for-age percentile <5), “healthy weight” (BMI-for-age percentile 5 to <85), or “overweight/obese” (BMI-for-age percentile ≥85) [28].

TABLE 1.

Sociodemographic characteristics and prevalence of weight change attempts in grade 8 among children in the Early Childhood Longitudinal Study, Kindergarten Class of 1998–1999 (ECLS-K98).

	Unweighted n1	Weighted (%) of sample	Weighted (%) attempting to lose weight	Weighted (%) attempting to gain weight
Overall	7115	100	39.7	11.5
Gender
Boys	3588	51.4	31.9	16.9
Girls	3527	48.6	48.0	5.9
P2			<0.001	<0.001
Age (y)
12.3–13.5	99	1.3	48.8	3.0
13.6–14.0	2200	31.1	42.6	11.4
14.1–14.5	3273	46.5	39.2	10.6
14.6–15.0	1424	19.3	36.3	13.8
15.1–<17	116	1.8	32.3	20.4
P-trend3			0.007	0.06
Race/ethnicity4
Non-Hispanic White	4766	58.1	36.8a,b,c	10.0f,g
Non-Hispanic Black	585	16.7	39.2d,e	21.3f,h,i,j
Hispanic, any race	1096	18.0	47.9a,d	8.0h
Asian	318	3.1	39.7	10.6
American Indian or Alaska native	67	0.8	61.5b,e	9.4i
Native Hawaiian or other Pacific Islander	123	1.4	46.1c	5.1g,j
Multiple races, non-Hispanic	158	2.0	42.1	12.5
P5			<0.001	<0.001
SES score6 in Kindergarten
Quintile 1	874	17.0	43.2	12.4
Quintile 2	1166	18.5	45.9	12.7
Quintile 3	1346	18.9	42.6	9.9
Quintile 4	1692	22.4	37.3	11.3
Quintile 5	2037	23.2	32.2	11.4
P-trend			<0.001	0.55
SES score in grade 3
Quintile 1	848	16.9	45.4	11.5
Quintile 2	1130	18.1	43.0	14.6
Quintile 3	1312	18.0	44.9	9.1
Quintile 4	1704	22.9	37.6	10.6
Quintile 5	2121	24.2	31.4	11.9
P-trend			<0.001	0.60
SES score in grade 5
Quintile 1	846	16.7	44.1	14.4
Quintile 2	1109	17.3	47.5	10.5
Quintile 3	1316	18.5	44.4	11.0
Quintile 4	1757	23.0	34.4	11.0
Quintile 5	2087	24.5	32.6	11.2
P-trend			<0.001	0.38
SES score in grade 8
Quintile 1	883	16.6	44.0	12.8
Quintile 2	1289	19.9	47.0	11.8
Quintile 3	1460	19.9	41.7	10.6
Quintile 4	1560	21.4	36.4	11.2
Quintile 5	1923	22.3	31.4	11.6
P-trend			<0.001	0.64
Food security change from grade 5 to grade 8
Remained food secure	6205	85.4	37.7k,l,m	11.4
Became food secure	275	5.2	49.0k	13.0
Became food insecure	276	5.3	50.9l	12.2
Remained food insecure	225	4.2	53.6m	11.2
P			<0.001	0.95
Total number of timepoints with food insecurity
0	5831	80.2	37.5	11.3
1	605	10.5	44.2	12.0
2	249	5.3	51.1	16.1
3 or 4	228	4.0	54.6	10.1
P-trend			<0.001	0.43

¹Due to missing values, the total n for some characteristics may add up to <7115.

²P values for gender came from Wald tests calculated using separate Poisson regression models where attempting weight loss (compared with not attempting weight loss) and attempting weight gain (compared with not attempting weight gain) were each modeled as a dichotomous outcome, and gender was included as an indicator variable. Models accounted for the complex sampling design.

³Trend tests were Wald tests conducted using separate Poisson regression models where attempting weight loss or attempting weight gain was modeled as a dichotomous outcome, and the sociodemographic characteristic was included as an ordinal variable. Models accounted for the complex sampling design.

⁴In the ECLS-K98 public-use dataset, the exact age of each child is not available. In grade 8, each child is classified into one of the following age categories: 148–<163 mo, 163–<169 mo, 169–<175 mo, 175–<181 mo, and 181–203 mo. For interpretability, we have converted these categories into units of years.

⁵For characteristics with > 2 nonordered categories, omnibus P values came from Wald tests calculated using separate Poisson regression models where attempting weight loss or attempting weight gain was modeled as a dichotomous outcome, and the sociodemographic characteristic was included as a categorical variable. For omnibus P values <0.05, pairwise comparisons were conducted using Wald tests. Statistically significant pairwise comparisons are indicated in the Table using superscript letters. Models accounted for the complex sampling design.

⁶The composite socioeconomic status (SES) score was created based on each parent/guardian’s education, each parent/guardian’s occupational prestige, and household income. Higher scores indicate higher SES.

Statistical analysis

Study weights in the ECLS-K98 account for the stratified and clustered sampling design, intentional subsampling of children who moved school districts between grades, child-level and school-level nonresponse at baseline, and child-level nonresponse over follow-up [24]. The current study was restricted to children with nonmissing data for relevant longitudinal study weights, who also had data on grade 8 weight change attempts and household FI for ≥1 timepoint. Of the 7558 children with longitudinal survey weight data, 7115 (94%) were included in the analytic sample after exclusions for missing exposure or outcome data. Within the analytic sample, the number of children who were nonmissing for each FI exposure variable ranged from 6913 (cumulative number of timepoints with FI) to 7099 (FI in kindergarten).

Descriptive statistics and analyses were weighted to account for the complex sampling design of the ECLS-K98. Poisson regression was used to model prevalence ratios for the outcomes of attempting weight loss (compared with not attempting weight loss) and attempting weight gain (compared with not attempting weight gain) in grade 8. Recent change in FI and cumulative number of timepoints with FI were modeled separately as categorical exposures. Analyses were adjusted for child age at completion of the grade 8 student questionnaire, child gender, child race/ethnicity, and SES score at all 4 timepoints. Analyses for FI change from grade 5 to grade 8 were additionally adjusted for FI in kindergarten and grade 3. When appropriate, trend tests were conducted using Wald tests with the exposure modeled ordinally. To examine gender as a potential effect modifier on the multiplicative scale, we tested an interaction term between gender and the respective FI variable in each fully adjusted Poisson model. For the outcome of weight loss attempts, we additionally tested for effect modification on the additive scale by replacing the natural log link function with an identity link function in each interaction model. For the outcome of weight gain attempts, there was insufficient statistical power to run additive interaction models, so effect modification by gender was assessed on the multiplicative scale only. Statistical significance was assessed using an alpha level of 0.05 for all analyses.

Finally, we also conducted supplementary analyses examining FI at each of the individual timepoints as separate, dichotomous exposures. We used the same modeling approach as described for the primary exposures. These supplementary analyses were adjusted for child age, child gender, child race/ethnicity, continuous SES score corresponding to the FI timepoint of interest, plus continuous SES score and FI status at all prior timepoints (when applicable).

Results

Overall, 48.6% of participants were girls; over 95% of the sample was between 13.6 and 15.0 y of age at grade 8 follow-up; and 58.1% of participants were non-Hispanic White. Among participants who had nonmissing data for BMI measured in grade 8 (unweighted n = 6735), the BMI-for-age distribution was as follows: 2.9% were “underweight” (BMI-for-age percentile <5), 61.5% were “healthy weight” (BMI-for-age percentile 5 to <85), and 35.6% were “overweight” or “obese” (BMI-for-age percentile ≥85). From kindergarten through grade 8, 80.2% of children were food secure at all timepoints (i.e., had FI at 0 timepoints), 10.5% had FI at 1 timepoint only, 5.3% had FI at 2 timepoints, and 4.0% had FI at 3–4 timepoints. From grade 5 to grade 8, 85.4% of children remained food secure, 5.2% became food secure, 5.3% became food insecure, and 4.2% remained food insecure. Over half (51.2%) of participants reported weight change attempts in grade 8. Specifically, 39.7% were trying to lose weight and 11.5% were trying to gain weight.

Complete sociodemographic characteristics and their bivariate associations with weight change attempts are shown in Table 1. In bivariate analyses, girls were more likely to attempt weight loss, whereas boys were more likely to attempt weight gain. Younger age in grade 8 was associated with a greater likelihood of trying to lose weight. Among children in the youngest age category of ≤13.5 y, the prevalence of attempting weight loss was 48.8%. In contrast, among children in the oldest age category of ≥15.1 y, the prevalence of attempting weight loss was 32.3%. Qualitatively, non-Hispanic White children had the lowest prevalence of trying to lose weight, whereas American Indian or Alaskan Native children had the highest prevalence of trying to lose weight. The prevalence of trying to gain weight was lowest among those who were Native Hawaiian or Other Pacific Islander and highest among those who were non-Hispanic Black. Statistically significant differences by race/ethnicity are indicated in Table 1. For each of the 4 timepoints, a higher SES score was associated with a lower prevalence of attempting weight loss. In contrast, trend tests showed no statistically significant associations between SES and attempting weight gain.

In adjusted analyses of recent FI change from grade 5 to grade 8 (Table 2), becoming food insecure (compared with remaining food secure), was associated with a greater likelihood of attempting weight loss [prevalence ratio (PR) = 1.25; 95% confidence interval (CI): 1.03, 1.52]. Remaining food insecure (compared with remaining food secure) was marginally associated with a greater likelihood of trying to lose weight (PR = 1.23; 95% CI: 1.00, 1.51; P = 0.053). Becoming food secure (compared with remaining food secure) was not statistically significantly associated with trying to lose weight. In the adjusted analysis of cumulative FI and weight loss attempts, there were no statistically significant associations for experiencing FI at 1 or 2 timepoints (compared with 0 timepoints). However, experiencing FI at 3–4 timepoints was associated with a greater prevalence of weight loss attempts in grade 8 (PR = 1.24; 95% CI: 1.04, 1.47). There was also evidence for a dose–response relationship between a greater number of timepoints with FI and higher prevalence of weight loss attempts (P-trend = 0.01).

TABLE 2.

Associations between longitudinal household food security measures and prevalence of attempting weight loss in grade 8 among children in the Early Childhood Longitudinal Study, Kindergarten Class of 1998–99 (ECLS-K98).

	Unweighted proportion attempting to lose weight	Weighted (%) attempting to lose weight	Unadjusted prevalence ratio (95% CI)1	Adjusted prevalence ratio (95% CI)2
Food security change from grade 5 to grade 83
Remained food secure	2363/6205	37.7	Ref	Ref
Became food secure	137/275	49.0	1.30 (1.08, 1.57)	1.11 (0.89, 1.39)4
Became food insecure	142/276	50.9	1.35 (1.12, 1.63)	1.25 (1.03, 1.52)
Remained food insecure	113/225	53.6	1.42 (1.20, 1.69)	1.23 (1.00, 1.51)
Total number of timepoints with food insecurity5
0	2199/5831	37.5	Ref	Ref
1	281/605	44.2	1.18 (1.05, 1.32)	1.04 (0.92, 1.19)
2	118/249	51.1	1.36 (1.12, 1.66)	1.20 (0.97, 1.47)
3 or 4	123/228	54.6	1.46 (1.24, 1.71)	1.24 (1.04, 1.47)
P-trend6			<0.001	0.01

Abbreviation: CI, confidence interval.

¹From separate Poisson regression models where trying to lose weight (compared with not trying to lose weight) was modeled as a dichotomous outcome, and each food security exposure of interest was included as a categorical variable. All models accounted for the complex sampling design.

²Adjusted for gender, age category, race/ethnicity, and continuous socioeconomic status (SES) score at all 4 timepoints (kindergarten, grade 3, grade 5, and grade 8). Models for food security change from grade 5 to grade 8 were also adjusted for food insecurity in kindergarten and grade 3.

³n, unadjusted analysis = 6981. n, adjusted analysis = 6908.

⁴Using the remained food insecure group as the comparison for those who became food secure, the estimate was 0.91 (0.69, 1.19).

⁵n, unadjusted analysis = 6913. n, adjusted analysis = 6908.

⁶Wald tests conducted by modeling the total number of timepoints with food insecurity as an ordinal variable.

FI change from grade 5 to grade 8 was not statistically significantly associated with trying to gain weight in the adjusted analyses (Table 3). The association between cumulative FI with weight gain attempts was also not statistically significant.

TABLE 3.

Associations between longitudinal household food security measures and prevalence of attempting weight gain in grade 8 among children in the Early Childhood Longitudinal Study, Kindergarten Class of 1998–99 (ECLS-K98).

	Unweighted proportion attempting to gain weight	Weighted (%) attempting to gain weight	Unadjusted prevalence ratio (95% CI)1	Adjusted prevalence ratio (95% CI)2
Food security change from grade 5 to grade 83
Remained food secure	658/6205	11.4	Ref	Ref
Became food secure	34/275	13.0	1.14 (0.64, 2.02)	1.02 (0.54, 1.93)4
Became food insecure	26/276	12.2	1.07 (0.61, 1.88)	0.90 (0.46, 1.75)
Remained food insecure	28/225	11.2	0.98 (0.57, 1.69)	0.79 (0.41, 1.52)
Total number of timepoints with food insecurity5
0	617/5831	11.3	Ref	Ref
1	64/605	12.0	1.06 (0.73, 1.54)	1.03 (0.70, 1.52)
2	36/249	16.1	1.42 (0.93, 2.17)	1.34 (0.85, 2.13)
3 or 4	23/228	10.1	0.89 (0.48, 1.67)	0.88 (0.50, 1.55)
P-trend6			0.43	0.63

Abbreviation: CI, confidence interval.

¹From separate Poisson regression models where trying to gain weight (compared with not trying to gain weight) was modeled as a dichotomous outcome, and each food security exposure of interest was included as a categorical variable. All models accounted for the complex sampling design.

²Adjusted for gender, age category, race/ethnicity, and continuous socioeconomic status (SES) score at all 4 timepoints (kindergarten, grade 3, grade 5, and grade 8). Models for food security change from grade 5 to grade 8 were also adjusted for food insecurity in kindergarten and grade 3.

³n, unadjusted analysis = 6981. n, adjusted analysis = 6908.

⁴Using the remained food-insecure group as the comparison for those who became food secure, the estimate was 1.29 (0.46, 3.60).

⁵n, unadjusted analysis = 6913. n, adjusted analysis = 6908.

⁶Wald tests conducted by modeling the total number of timepoints with food insecurity as an ordinal variable.

Overall, there was no evidence that associations between primary FI measures and weight loss attempts were modified by gender on the multiplicative or additive scales (P interactions >0.05, data not shown). There was also no evidence for gender differences in the associations between FI measures and weight gain attempts on the multiplicative scale (P interactions >0.05, data not shown), although additive interactions could not be assessed for weight gain attempts.

Supplemental Table 1 shows the distribution of FI at each of the individual timepoints, and findings from supplementary analyses examining FI at individual timepoints in relation to weight loss attempts in grade 8 are shown in Supplemental Table 2. After adjustment for covariates, FI in grade 3 (compared with no FI in grade 3) was associated with a greater likelihood of attempting weight loss, as was FI in grade 8 (PR = 1.22; 95% CI: 1.02, 1.47 and PR = 1.19; 95% CI: 1.02, 1.41, respectively). However, FI in kindergarten and FI in grade 5 were not statistically significantly associated with weight loss attempts. All associations between FI at individual timepoints and weight gain attempts in grade 8 were statistically nonsignificant (Supplemental Table 3). None of the interaction terms between gender and FI were statistically significant in supplementary analyses (P interactions >0.05, data not shown).

Discussion

In this nationally representative longitudinal study of United States children, we found that children who became food insecure from grade 5 to grade 8 were more likely to attempt weight loss compared with those who remained food secure. In addition, greater cumulative exposure to FI from kindergarten through grade 8 was associated with a greater likelihood of attempting weight loss in grade 8. There was no evidence that gender modified associations between FI and weight change attempts. Our findings suggest that childhood FI may be a risk factor for weight loss attempts among young adolescents.

Overall, these results corroborate and build on findings from Masler et al. [16], who showed that household FI was cross-sectionally associated with more weight loss attempts in a nationally representative sample of United States children aged 8–15 y. Our finding that recently becoming food insecure was associated with a greater likelihood of attempting weight loss is novel. To the best of our knowledge, the association between recent FI change and weight loss attempts has not been previously explored. In addition, we found that experiencing FI during 3–4 (compared with 0) timepoints was associated with a greater likelihood of attempting weight loss, with evidence for a significant linear trend between a greater number of timepoints with FI and higher prevalence of attempting weight loss. This last finding reinforces that both transient and persistent experiences of FI across childhood may contribute to a greater likelihood of attempting weight loss among young adolescents. These findings are also consistent with broader research showing that transitions into FI and persistent FI may have negative consequences for other aspects of child health and development [22,23,26,29].

Although the mechanisms relating FI to weight loss attempts warrant further research, we hypothesize that adolescent body image and food- and weight-related parenting practices might be involved. There are a few studies linking FI with poor body image [[2], [3], [4]], which is a key risk factor for weight loss attempts [9,12,13]. As suggested by Altman et al. [3], it is possible that experiencing FI increases children’s focus on their eating habits and bodies, leading to more body dissatisfaction. FI has also been associated with worse mental health in some studies of children [30], which may contribute to poor body image and weight loss attempts.

FI may also influence parenting practices that could affect a child's body image. First, concern about the availability of quality food could increase parental focus on children's eating behaviors and child weight. For example, FI has been associated with more parental concern about child weight, and among boys, this association remained statistically significant even after adjustment for child weight status [31]. In addition, parents may use controlling feeding practices as a means to cope with limited and unpredictable access to food [32]. For example, FI has also been associated with problematic parental feeding practices, like restrictive feeding and pressure to eat [[31], [32], [33]]. In turn, there is evidence that parental feedback on child weight/weight-related behaviors and use of controlling parental feeding practices are associated with more body dissatisfaction and weight change behaviors in adolescents [34,35].

We also suspect that weight status is a partial mediator of the associations between FI and weight loss attempts in this study. Although the exact nature of the association between FI and weight status in children remains equivocal, there is some longitudinal evidence that FI during childhood is associated with more weight gain in girls [[21], [22], [23]], including previous analyses conducted in the ECLS-K98 cohort [21,23]. Higher weight status, in turn, has been associated with poor body image, more internalized weight bias, and more weight loss attempts in children [12,36,37]. Because weight status is a likely mediator, we chose not to adjust for or stratify by weight status in this study, as adjusting for a mediator would induce overadjustment bias in our analyses focused on understanding overall associations between FI and weight change attempts [38]. Of note, FI and weight status could also interact with regard to their effects on variables like weight-related parenting behaviors and body image. Furthermore, weight status likely has bidirectional associations with other relevant variables, like parent feeding practices. Future research should formally examine the roles of variables such as weight status, internalized weight bias, body image, mental health, parent feeding practices, and weight-related parenting behaviors regarding associations between FI and weight change attempts.

Supplementary analyses showed that FI in kindergarten was not statistically significantly associated with attempting weight loss in grade 8. This finding may be due to the long timespan between kindergarten and grade 8. It is possible that early childhood FI may be less relevant for weight loss attempts in adolescence compared with more recent experiences of FI. The prevalence ratio estimates for FI at all other timepoints were >1, suggesting an association with more weight loss attempts, although only grade 3 and grade 8 FI had statistically significant associations. These findings suggest that grade 3 (when children are ∼8–9 y old) and grade 8 (when children are ∼13–14 y old) might be sensitive periods regarding the association between FI and attempting weight loss. Grade 3 FI may be important because it may immediately precede or coincide with the development of poor body image, as research suggests that poor body image is already stabilized by, and possibly before, 11 y of age [39]. Grade 8 FI may also be especially salient in this study because it was assessed concurrently with weight loss attempts. In addition, by grade 8, most children are likely in the midst of puberty, a critical period for body image and eating disorder development [40]. Finally, if weight status is a mediating variable, our different findings at specific timepoints might reflect differences in the association between FI and weight status over time from kindergarten to grade 8 [21].

In the present study, we only found statistically significant evidence for associations between FI and weight loss attempts, whereas all associations between FI and weight gain attempts were statistically nonsignificant. This finding is in contrast to the study by Gulliford et al. [19], who found that among 16-y-old adolescents in Trinidad, those with FI were more likely to attempt weight gain compared with those with food security. In addition, Gulliford et al. [19] found no association between FI and weight loss attempts. These different findings might reflect cultural and environmental differences in Trinidad compared with the United States. Gulliford et al. [19] noted that the prevalence of trying to gain weight was higher among girls in Trinidad compared with girls in England, where the latter cultural context may be more similar to the United States. Differences in weight status, as well as differences in how FI relates to and/or interacts with weight status, could also help explain different findings between the samples. Compared with the ECLS-K98 sample, the study sample in Gulliford et al. [19] had a very different weight distribution: In Gulliford et al. [19], ∼16% of participants had a BMI-for-age Z-score (BMIZ) that was ≥1, and 29% had a BMIZ of <−1. In contrast, 36.4% of the adolescents in our sample had a BMIZ that was ≥1 and only 6.7% had a BMIZ of <−1. Intriguingly, Gulliford et al. [19] also did not find evidence for an association between FI and weight status in their study sample. In contrast, there is evidence for an association between FI and weight status among girls in the ECLS-K98 [21,23].

Finally, our results showed no evidence that any associations between FI and weight change attempts were modified by gender. This finding is consistent with the study by Altman et al. [3], who found that the association between child-reported FI and child body dissatisfaction was not modified by sex in a cross-sectional study of children in grades 4–8. In addition, it could be that gender does not modify the association between FI and weight loss attempts but does modify associations between FI and specific weight loss methods, which are known to pattern by gender [8,9]. Children were not asked how they tried to lose weight in the ECLS-K98, so we were unable to study specific weight loss behaviors in the present study.

Our results suggest that FI interventions might help prevent body image concerns and unhealthy weight control behaviors, in addition to the myriad other negative health consequences that have been linked to childhood FI [22,23,26,29]. We found that both transient and persistent experiences of FI were associated with a higher prevalence of weight loss attempts, suggesting that long-term FI prevention programs that are consistently maintained throughout childhood and adolescence would likely have the largest benefit. In addition, this research reinforces the need for schools, healthcare providers, public health professionals, and community leaders to proactively connect youth and families to resources for food assistance, because preventing FI may have long-lasting benefits for the health and well-being of children.

Our findings also suggest that combining FI and eating disorder prevention strategies might help minimize the risk of dangerous weight control behaviors. For example, experiences with FI should be addressed in eating disorder prevention and treatment settings. In addition, programs aimed at promoting positive body image and preventing eating disorder behaviors could be implemented in schools that have a high prevalence of FI among students. There are existing body image intervention programs that have been well-validated among elementary-age girls, which might be feasible for schools to implement [41]. However, there remains an urgent need for programs that are validated for use in boys and low-income populations [41].

This study has several strengths. Data came from a large, nationally representative cohort study of United States children. We were able to characterize exposure to FI across childhood, including a measure of cumulative exposure from kindergarten through grade 8. At all 4 timepoints, household FI was assessed using a well-validated and widely accepted measure [25].

This study also has some limitations. First, FI was not measured every year between kindergarten and grade 8, meaning we could not fully capture total FI exposure during this timeframe. However, we did include 4 measures of FI across 8 y of data collection, which is more comprehensive than most studies focused on FI in children [29]. We also did not have data on child-reported experience of FI, which could differ from parent reports. Although we had a large sample, statistical power could have been limited due to the relatively low prevalence of some primary variables, including FI (<10% at each timepoint) and weight gain attempts (11.5%). Limited statistical power is especially possible for the models used to test for effect modification by gender. For weight gain attempts, we could only assess effect modification by gender on the multiplicative scale because the additive interaction models did not converge. Finally, weight change attempts were not assessed prior to grade 8, so none of the outcomes could be considered prospective. However, we hypothesize that weight change attempts would have been uncommon when the cohort was recruited in kindergarten, and reverse causation bias is unlikely, because child weight change attempts would be unlikely to influence parent report of household FI.

In conclusion, we found that recently becoming food insecure and greater cumulative exposure to FI throughout childhood were each associated with a greater likelihood of trying to lose weight in early adolescence. These findings are especially important in light of growing evidence that FI is related to more eating disorder behaviors in children and adolescents [6]. Results from this study emphasize the need for more research to understand the complex relationships between FI, body image, intentional weight change attempts, and eating disorder behaviors. Finally, these findings underscore the need for long-lasting FI prevention and intervention strategies, as well as the development of programs to promote healthy body image and eating patterns among children with FI.

Author contributions

The authors’ responsibilities were as follows – MRB: analyzed the data, wrote the article, and had primary responsibility for final content; and all authors: designed the research, read, and approved the final manuscript.

Data availability

Data used in this study are publicly available without restriction through the National Center for Education Statistics website at https://nces.ed.gov/ecls/dataproducts.asp.

Funding

Article preparation was supported by funding from the Horace H. Rackham School of Graduate Studies at the University of Michigan (Ann Arbor, MI), as well as funding from the National Heart, Lung, and Blood Institute (Grant Number T32HL150452, PI: Dianne Neumark-Sztainer) and the National Institute of Mental Health (Grant Number T32MH082761, PI: Carol B. Peterson). Content in this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Heart, Lung, and Blood Institute, the National Institute of Mental Health, or the NIH.

Conflict of interest

MRB reports financial support was provided by the NIH, including the National Heart, Lung, and Blood Institute and the National Institute of Mental Health. The other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Appendix A. Supplementary data

The following is the Supplementary data to this article: