To capture temporal dynamics, I constructed a five-year panel dataset using annual child food insecurity estimates and school meal participation rates. I employ fixed effects regression to control for time-invariant differences across counties, along with year fixed effects to account for statewide shifts–particularly in the implementation and expiration of universal school meals during the pandemic.

This regression framework allows for both cross-sectional and longitudinal analyses of how structural, demographic, and policy-based factors interact to shape child food insecurity across the state.

RESULTS

Model 1

In my initial model I sought to uncover the relationship between key poverty indicators and the rate of child food insecurity at the county level in Utah. To do so I developed a linear regression model using child food insecurity rate as the dependent variable. The independent variables in this model include the percentage of food insecure individuals living at or below the SNAP income eligibility threshold, the percentage of food-insecure children living in households below 185% of the federal poverty level, and the average cost per meal in each county.

The coefficient estimate for the percentage of food insecure individuals below the SNAP threshold is 0.236 (p < 0.001), indicating a statistically significant and positive relationship. Specifically, a 1 percentage point increase in the share of food insecure individuals who are income eligible for snap with a 0.236 point increase in the child food insecurity rate. This supports the hypothesis that counties with a higher concentration of deeply poor, food insecure households face more acute childhood hunger. The coefficient for the percentage of food insecure children in households under 185% FPL is 0.0166, but this effect was not statistically significant (p = 0.75). Similarly, the cost per meal variable had a negative coefficient of –0.0187, suggesting that higher food prices might correspond with lower food insecurity—an unexpected and likely spurious relationship that was not statistically significant (p = 0.15). One explanation for this could be that counties with higher food costs are often wealthier or urban areas with greater access to safety nets, skewing the direction of the correlation.

The R-squared value of the model is 0.77, indicating that approximately 77% of the variance in child food insecurity across Utah counties is explained by the model’s three predictors. This relatively high R-squared demonstrates that the core economic indicators selected for Model 1 capture a large portion of the county-level differences in food insecurity. Furthermore, the F-statistic was statistically significant, confirming the overall explanatory power of the model.

Model 1b

Model 1b was developed to assess whether food insecurity among children is significantly associated with the share of food insecure households that fall above the income threshold for SNAP eligibility. This regression model excludes those at or below the snap cutoff to explore how child food insecurity behaves when need exists among households not traditionally targeted by federal assistance. The model regresses county-level child food insecurity rates against two independent variables: the percentage of food insecure individuals whose household income is above the snap threshold and cost per meal.

The results indicate that the percentage of food-insecure individuals above the SNAP threshold is significantly and negatively associated with county level child food insecurity (β = –0.2517, p < 0.001). This suggests that counties where a greater share of food insecure individuals are ineligible for snap tend to have lower child food insecurity rates overall. This may reflect differences in access to alternative resources, greater stability among the near-poor, or the ability to bridge food gaps with private means–even in the absence of federal assistance.

The fact that higher shares of food insecure households that don’t qualify for SNAP corresponds with lower child food insecurity may also suggest that child hunger is more concentrated in counties where food insecurity is dominated by those with the lowest incomes who are more deeply disadvantaged despite qualifying for federal programs. The cost per meal variable was again negative but not statistically significant echoing previous models in suggesting that food prices alone do not predict child food insecurity when controlling for income related variables.

Model 1b demonstrated a strong overall fit with an R-squared of 0.765 indicating that over 76% of the variance in child food insecurity is explained by these predictors. While past studies have emphasized the protective role of food assistance programs for low-income families this model points to a gap in coverage for those just outside the eligibility threshold. It reinforces the need for more expansive policies that do not rely on income-based cutoffs to identify needs. Especially given that many food insecure children live in households that exceed those thresholds but still struggle to access sufficient nutrition.

Model 2

The results from Model 2 show that median household income is a statistically significant predictor of child food insecurity (β = –1.20e-06, p < 0.01), even when controlling for other factors. The negative sign confirms that counties with higher median incomes tend to have lower rates of child food insecurity which is consistent with existing literature linking family income with access to basic needs. This finding strengthens the argument that structural economic inequality at the county level is a key determinant of food access for children.

However other variables in the model such as unemployment rate and renter share were not statistically significant in this specification. Notably the coefficient for % FI < SNAP threshold was positive( β = 0.068) but not significant (p = 0.33), and the cost per meal remained insignificant as in previous models. Despite this the overall model fit was strong with an R-squared of 0.87 and a statistically significant F-test (p < 0.001), indicating that the full set of predictors accounts for a large share of the variation in food insecurity across counties.

Model 2b

To isolate the effects of other structural and program-relevant variables I estimated a second version of the model that excludes median household income. This adjustment allowed us to see whether unemployment, renter share, and poverty related food insecurity indicators would emerge as significant predictors on their own.

The results reveal that the unemployment rate becomes marginally significant (β = 0.47, p = 0.05), indicating that counties in Utah with higher unemployment tend to experience higher rates of child food insecurity. This aligns with federal research from the USDA which found that children in households with unemployed adults were three and a half times more likely to face food insecurity (2022). Additionally, the % FI < SNAP threshold variable becomes more pronounced and approaches significance (β = 0.138, p = 0.086), suggesting a potential link between concentrated economic hardship and child hunger when income effects are not directly included. Although no other individual variables reached conventional levels of statistical significance the overall model retained strong explanatory power (R-squared = 0.81).

These findings contribute to literature by confirming that income remains one of the most powerful predictors of food security even when other economic stressors like unemployment and rental housing share are considered. Removing income from the model revealed latent relationships between unemployment and food insecurity that are often masked in full economic models. This suggests that in lower income counties joblessness plays a critical role in shaping food outcomes for children.

From a policy perspective the shift in significance between model 2 and 2b reinforces the need to design multifaceted interventions which include both income supports (tax credits, job programs, ect) and direct food access policies like universal school meals. These results validate concerns that the school meals eligibility threshold at 185% of poverty of the poverty line excludes many children at genuine risk of food insecurity.

Model 3

The results of model 3 reveal that the percentage of food-insecure individuals below the SNAP threshold remains a key predictor (β = 0.173, p = 0.0569) with marginal significance at the 10% level. This finding is consistent with models 1 and 2b and reinforces the idea that extreme economic hardship is closely tied to food insecurity among children.

Additionally, the unemployment rate again approached significance (β = 0.503, p = 0.0788), continuing to suggest that local labor market conditions play an important role in shaping food insecurity risk. These findings echo national research that links parental joblessness with food hardship and economic precarity, especially among children (USDA, 2022).

However other variables–particularly those introduced in this model–were not statistically significant. The share of adults with less than a high school diploma (β = -0.197, p = 0.3181) and the English learner share (β = -0.0082, p = 0.9367) did not exhibit meaningful associations with county food insecurity in this specification. While these factors may still be important at the household level, their county-level effects were likely absorbed by stronger predictors like unemployment and SNAP eligible food insecurity.

The model’s performance remained strong with an R-squared of 0.82, indicating that the included predictors collectively explain over 80% of the variation in county level child food insecurity. This suggests that despite some individual coefficients lacking statistical significance the broader combination of economic hardship indicators and food program thresholds provide a powerful explanation of the spatial disparities in food insecurity across utah counties.

These results add nuance to existing literature by testing the relevance of educational attainment and language access barriers at a localized level. While prior studies have shown households headed by adults without high school degrees are far more likely to experience food insecurity this county level analysis found no such relationship when controlling for other factors. Similarly, while language barriers are often cited as barriers to benefit participation (Bartfeld, 2013), the inclusion of the English learner share variable did not improve model fit or predictive strength in this setting. The findings suggest that county level variation in food insecurity is better explained by direct measures of income and employment. Interventions aimed at boosting participation in food assistance programs like school meals should continue to focus on removing structural barriers like income thresholds and documentation requirements especially in communities with elevated unemployment or poverty rates.

Model 4

Model 4 centers on the relationship between child food insecurity and the demographic composition of student populations across Utah counties. Drawing on research that links racial and ethnic disparities to differences in food access, as well as evidence highlighting the vulnerability of children with disabilities, experiencing homelessness, or classified as economically disadvantaged, this model evaluates how these factors vary in predictive strength when measured at the county level. The full multivariate specification of model 4 includes shares of Hispanic, Black, and American Indian students, along with shares of students experiencing homelessness, students with disabilities, and those classified as economically disadvantaged. Among these the economically disadvantaged student share was statistically significant and positively associated with food insecurity (β = 0.277, p < 0.001). This suggests that counties with a higher proportion of economically disadvantaged students also experience substantially higher rates of child food insecurity, reinforcing well-documented links between low household income and reduced access to stable food sources.

None of the racial or ethnic demographic shares reached statistical significance in the full model. Several other variables such as homelessness (β = –0.164, p = 0.778) and disability (β = – 0.109, p = 0.683), exhibited unexpected negative or null associations. These results may reflect multicollinearity or the small number of counties used in the regression which can suppress individual significance when multiple vulnerable groups are included simultaneously.

To further examine the individual contribution of each variable a trimmed model excluding the economically disadvantaged share (Model 4b) was estimated. In this alternate model the disability share emerged as a significant predictor (β = 0.566, p = 0.047), which indicates that when not overshadowed by economic disadvantage the presence of students with disabilities correlates with higher levels of child food insecurity. This finding aligns with national research documenting higher food insecurity among households with children who have disabilities due to higher costs, reduced employment options for caregivers, and eligibility gaps in public benefits (Brucker et al., 2021).

A series of bivariate regressions provided additional insight. The American Indian student share, while not significant in the multivariate model, had a significant positive association with child food insecurity when tested alone (β = 0.178, p = 0.022). This result underscores the disproportionate food insecurity burden in Utah’s tribal communities–particularly important given the geographic concentration of tribal lands in more remote southeastern counties. This finding contributes to the broader literature of food insecurity disparities among indigenous populations (Jernigan et al. 2017).

Other racial and ethnic bivariate regressions, including Hispanic, Black, and White shares were not statistically significant. Notably the Pacific Islander student share showed a marginally significant negative association (β = -2.07, p = 0.083) suggesting potential differences in food security outcomes or community level factors not fully captured in this dataset. Similarly the Asian share had a negative association that approached significance (β = -2.23, p = 0.070). While these results should be interpreted with caution due to small population sizes they may reflect localized support systems or cultural factors that merit further investigation.

The bivariate model for the homeless student share revealed a significant and positive relationship with food insecurity (β = 0.466, p = 0.028), suggesting that when measured in isolation homelessness is indeed a relevant risk factor, though its predictive power was diminished when controlling for overlapping vulnerabilities in the multivariate model. The English Learner (ELL) share model did not show a significant association with food insecurity (p = 0.222), suggesting that language status alone may not be a strong determinant of food insecurity at the county level in Utah.

Taken together the results of model 4 and its bivariate extensions highlight the complex and overlapping nature of demographic vulnerability. While not all individual population groups reached significance in the full model, several including students with disabilities, and American Indian students, and students experiencing homelessness displayed clear positive associations with food insecurity when examined independently. These findings contribute to the literature by demonstrating that county level child food insecurity in Utah is shaped not only by economic conditions but also by the composition of student populations who face intersecting barriers to access, inclusion, and opportunity.

These results underscore the value of expanding universal school meal access as a strategy for reducing hunger across multiple vulnerable groups- not just those who qualify through conventional income thresholds. Additionally targeted outreach or supplemental support for districts with high concentrations of students with disabilities, tribal youth, and those experiencing homelessness could improve outcomes for populations at highest risk.

Model 5

Model 5 brings together the most inclusive set of explanatory variables synthesizing economic need, educational attainment, and demographic vulnerability into a single regression model. This comprehensive model was designed to capture the broader social determinants of child food insecurity by combining key economic indicators with county-level population characteristics. Specifically, the model includes the percentage of food-insecure individuals below the SNAP threshold, unemployment rate, cost per meal, renter share, educational attainment, and shares of American Indian, homeless, and disabled student populations.

The results of model 5 reveal that only one predictor–% food insecure below the SNAP threshold– is statistically significant at the 0.001 level (β= 0.2459, p < 0.001). This variable consistently emerges as the most robust predictor across all models reaffirming that counties with a larger proportion of extremely poor food-insecure residents experience substantially higher food insecurity rates. This finding underscores the central role that deep poverty plays in driving hunger among children, and aligns with prior research that emphasizes the heightened vulnerability of households near or below federal assistance thresholds.

Although the other variables in the model– such as unemployment rate, cost per meal, renter share, and educational attainment–are theoretically important and supported by the literature, they do not reach statistical significance in this specification. It is possible that some of their explanatory power is absorbed by the strength of the SNAP related poverty variable or that collinearity between predictors slightly suppresses their individual effects. For example, both educational attainment and unemployment are generally correlated with income levels and poverty status in most literature.

Interestingly, while the coefficient for unemployment remains positive suggesting that higher unemployment may increase food insecurity it is not statistically distinguishable from zero (p = 0.29). Similarly, the coefficient for cost per meal remains negative but not significant indicating that food prices may not directly drive food insecurity.

The inclusion of demographic vulnerability indicators– American Indian share, homelessness share, and disability share– adds depth to the model particularly given the importance of these groups in food insecurity literature. While these variables are not individually significant their inclusion helps assess whether disparities persist after controlling for structural economic conditions. Notably the direction of the coefficients aligns with expectations: higher disability share and American Indian share are positively associated with child food insecurity, though not significantly so in this model. The homelessness share shows a negative coefficient in this specification which is inconsistent with prior findings and may reflect collinearity in this fully saturated model.

The model fit is strong with an R-squared of 0.83 and adjusted R-squared of 0.76, indicating that over 76% of the variation in child food insecurity across counties is explained by the included variables. The RESET test returned a p value of 0.64, suggesting no specification errors and validating the mode’s functional form. Likewise, the Breush-Pagan test for heteroskedasticity yielded a p-value of 0.1495 providing no evidence of heteroskedasticity and supporting the reliability of OLS estimates.

Taken together the findings from Model 5 highlight the dominant role of poverty intensity particularly among those below the SNAP threshold in shaping child food insecurity rates in Utah counties. While other socioeconomic and demographic factors are important for contextualizing disparities, the model reinforces the literature’s emphasis on income based eligibility thresholds and adds further support for policies that extend food assistance beyond narrowly targeted income brackets. This finding bolsters arguments for universal school meal programs which can help mitigate gaps for those outside of eligibility cutoffs.

Panel Data Models

To complement the cross sectional models and examine changes over time I constructed a series of panel data models using county level observations from 2018 to 2022. These models leverage the balanced panel structure of 29 Utah counties over five years to assess whether increased participation in free and reduced price lunch programs is associated with changes in child food insecurity rates, while accounting for unobserved, time-invariant characteristics of each county.

The first model estimated was a fixed effects model focusing on county level variation over time. Surprisingly Free and Reduced Lunch Participation was not statistically significant in this model ( β= –0.038, p = 0.42). This result suggests that after controlling for all time invariant characteristics of each county changes in free and reduced lunch participation within counties were not strongly associated with short run changes in food insecurity during this period.

I then did a cross-sectional OLS model using data from the 2022 year alone. In this specification free and reduced lunch participation exhibited a highly significant positive association with child food insecurity(coefficient = 0.186, p < 0.001), explaining approximately 60 percent of variation across counties. While initially counterintuitive this result likely reflects the endogeneity of school lunch program participation, where higher levels of food insecurity motivate districts to intensify enrollment efforts. This finding also highlights the difficulty in separating causal effects from targeted program delivery in static cross-sectional data.

A pooled OLS model that combined all county-year observations without fixed effects similarly produced a statistically significant and positive coefficient for free and reduced lunch participation (0.189, p < 0.001). With an R-squared of 0.47 this model captured meaningful variation but failed to distinguish persistent differences between counties from within-county temporal changes.

The most informative specification included both county and year fixed effects. In this model the free and reduced lunch participation coefficient was statistically insignificant (0.055, p = 0.182) but the inclusion of year indicators revealed critical shifts across the five-year period. Notably the coefficient for 2021 was large, negative, and highly significant, indicating a substantial drop in food insecurity coinciding with the peak implementation of federal universal school meal waivers during the pandemic. In contrast, the 2022 coefficient was large and positive suggesting a significant rebound in food insecurity following the expiration of those waivers and the reversion to means-tested meal eligibility in many districts. These year specific trends suggest that universal access to school meals had a substantial short-term protective effect of child food insecurity, while its removal contributed to renewed hardship.

Together, the panel data models reveal important temporal dynamics that are not visible in cross-sectional analysis alone. While the fixed effects specification did not detect a robust within- county causal effect of participation on food insecurity, the sharp contrasts observed during and after the universal free meals period reinforce the broader literature emphasizing the value of universal programs. These results support recent national findings that temporary policy changes during the pandemic—including expanded access to school meals—produced tangible reductions in child hunger. Moreover, they suggest that narrowly targeted, means-tested models may leave substantial gaps in coverage, especially for children in households just above eligibility thresholds.

This longitudinal analysis contributes to the existing scholarship by grounding national trends in localized Utah data, offering rare insight into how state and county policy environments mediate program effectiveness. In doing so, the findings provide strong empirical support for policy frameworks that expand access universally, rather than relying solely on income-based eligibility criteria to address child food insecurity.

CONCLUSIONS

This thesis has examined the socioeconomic and demographic predictors of child food insecurity across Utah’s 29 counties, with particular attention to the role of school meal programs as both a stabilizing intervention and a potential long-term policy solution. Through a combination of cross-sectional OLS models and longitudinal fixed effects panel regressions, the analysis has generated several key findings that advance our understanding of local-level food insecurity and its relationship to school-based nutrition access.

Across multiple models, the most consistent and robust predictor of child food insecurity was the percentage of food-insecure individuals living in households below the Supplemental Nutrition Assistance Program (SNAP) income threshold. This finding underscores the severity of need among Utah’s poorest families and reaffirms that economic hardship remains the most immediate driver of food insecurity among children. However, the results also reveal a more nuanced picture. Model 1b showed that counties with a larger share of food-insecure households above the SNAP threshold tended to experience lower child food insecurity, suggesting that families just outside eligibility for federal programs may still benefit from other buffers—such as higher employment rates, stronger local supports, or expanded school meal access.

Models incorporating broader economic indicators revealed that median household income remained a significant predictor of food insecurity, while variables like unemployment rate, housing tenure, and cost per meal were less consistently predictive once income was accounted for. Demographic-based models further highlighted the vulnerabilities of specific populations.

While most racial and ethnic student share variables were not individually significant in multivariate models, the proportion of economically disadvantaged students was a strong and consistent predictor of higher food insecurity. Notably, American Indian student share was positively and significantlyassociated with child food insecurity in bivariate analysis, echoing national studies documenting elevated hunger rates among Indigenous communities. Shares of homeless and disabled students also showed positive and statistically significant associations with food insecurity, suggesting that school district-level data on student subpopulations can provide early indicators of food access challenges.

Panel models allowed for a dynamic analysis of school meal participation over time. Although fixed effects models did not detect a statistically significant relationship between increased free and reduced-price lunch participation and reductions in child food insecurity within counties, cross- sectional models showed a strong positive association—likely driven by targeted program expansion in high-need areas. Most importantly, year fixed effects revealed a dramatic drop in food insecurity during the peak of universal school meal waivers in 2021, followed by a notable rebound in 2022 once those federal waivers expired. This temporal pattern provides compelling evidence for the effectiveness of universal meal access, particularly during times of economic distress.

Taken together, these findings contribute to the existing literature by localizing national debates within the context of Utah’s unique demographic and policy landscape. They confirm many established associations—such as the links between poverty, unemployment, demographic background, and food insecurity—while also introducing new evidence on the limitations of narrowly targeted nutrition programs. The observed rebound in food insecurity after universal meals were removed raises critical questions about the sustainability and equity of current access models.

The policy implications of this research are clear. Programs like the National School Lunch Program are most effective when access is universal rather than contingent on narrow eligibility screens that may miss children in need. This thesis offers empirical support for recent legislative efforts in Utah to restore universal free school meals and highlights the importance of maintaining these supports beyond emergency contexts.

Future research should build upon these findings by incorporating household-level survey data to better assess the causal impact of school meal participation on child outcomes. I utilized county level estimations and census figures which likely weakened associations and suppressed significance. Using data that are direct survey responses from households would allow for more precise measurement of food insecurity and a richer analysis of its correlates. Additionally, exploring the intersection of food insecurity with transportation access, school district funding, and local food environments could further deepen our understanding of the structural determinants of hunger. As food insecurity continues to affect thousands of children across the state, data-driven, equity-centered policy solutions remain essential to ensuring that no child goes hungry.

Bibliography

Arteaga, I., & Heflin, C. (2014). Participation in the National School Lunch Program and food security: An analysis of transitions into kindergarten. Children and Youth Services Review, 47(Part 3), 224–230. https://doi.org/10.1016/j.childyouth.2014.09.014

Judi Bartfeld & Rachel Dunifon, 2006. “State-level predictors of food insecurity among households with children,” Journal of Policy Analysis and Management, John Wiley & Sons, Ltd., vol. 25(4), pages 921-942.

Chaparro, M. P., Cruthirds, S., Bell, C. N., & Wallace, M. E. (2022). State-level socioeconomic racial inequity and food insecurity in the U.S. American Journal of Preventive Medicine, 63(6), 971–978. https://doi.org/10.1016/j.amepre.2022.06.019

Coleman-Jensen, A., McFall, W., & Nord, M. (2013). Food insecurity in households with children: Prevalence, severity, and household characteristics, 2010–11 (EIB-113). U.S. Department of Agriculture, Economic Research Service.

Gundersen, C., & Kreider, B. (2009). Bounding the effects of food insecurity on children’s health outcomes. Journal of Health Economics, 28, 971–983.

Gundersen, C., Kreider, B., & Pepper, J. (2012). The impact of the National School Lunch Program on child health: A nonparametric bounds analysis. Journal of Econometrics, 166, 79–91.

Huang, J., Kim, Y., & Barnidge, E. (2016). Seasonal difference in National School Lunch Program participation and its impacts on household food security. Health & Social Work, 41(4), 235–243. https://doi.org/10.1093/hsw/hlw043

Huang, J., Matta Oshima, K., & Kim, Y. (2010). Does food insecurity affect parental characteristics and child behavior? Testing mediation effects. Social Service Review, 84(3), 381–401.

Jernigan, V. B. B., Huyser, K. R., Valdes, J., & Simonds, V. W. (2017). Food insecurity among American Indians and Alaska Natives: A national profile using the Current Population Survey–Food Security Supplement. Journal of Hunger & Environmental Nutrition, 12(1), 1–10. https://doi.org/10.1080/19320248.2016.1227750

Kabbani, N. S., & Kmeid, M. Y. (2005). The role of food assistance in helping food insecure households escape hunger. Review of Agricultural Economics, 27(3), 439–445.

Parkinson, R. (2024, November 8). The Cold Truth of Hunger in Utah . Utah Food Bank. https://www.utahfoodbank.org/2024/11/08/the-cold-truth-of-hunger-in-utah/

Sonik, R., Parish, S. L., Ghosh, S., & Igdalsky, L. (2016). Food insecurity in U.S. households that include children with disabilities. Exceptional Children, 83(1), 42–57. https://doi.org/10.1177/0014402916651847

Fafard St-Germain, A. A., & Tarasuk, V. (2020). Homeownership status and risk of food insecurity: examining the role of housing debt, housing expenditure and housing asset using a cross-sectional population-based survey of Canadian households. International journal for equity in health, 19(1), 5. https://doi.org/10.1186/s12939-019-1114-z

USDA. (2025, January 8). Food security in the U.S. – key statistics & graphics. Food Security in the U.S. – Key Statistics & Graphics | Economic Research Service. https://www.ers.usda.gov/topics/food- nutrition-assistance/food-security-in-the-us/key-statistics- graphics#:~:text=7.2%20million%20children%20lived%20in,experienced%20very%20low%20food% 20security.

APPENDIX