Bounding the effects of food insecurity on children’s health outcomes

Abstract

Previous research has estimated that food insecure children are more likely to suffer from a wide array of negative health outcomes than food secure children, leading many to claim that alleviating food insecurity would lead to better health outcomes. Identifying the causal impacts is problematic, however, given endogenous selection into food security status and potential mismeasurement of true food security status. Using recently developed nonparametric bounding methods and data from the 2001–2006 National Health and Nutritional Examination Survey (NHANES), we assess what can be identified about the effects of food insecurity on child health outcomes in the presence of nonrandom selection and nonclassical measurement error. Under relatively weak monotonicity assumptions, we can identify that food security has a statistically significant positive impact on favorable general health and being a healthy weight. Our work suggests that previous research has more likely underestimated than overestimated the causal impacts of food insecurity on health.

Introduction

Positive associations between food insecurity and poor health outcomes among children have been widely documented. Previous research, spanning numerous academic studies, has found that children in households suffering from food insecurity are more likely to have poor health Cook et al., 2004, Weinreb et al., 2002, Dunifon and Kowaleski-Jones, 2003, psychosocial problems Kleinman et al., 1998, Weinreb et al., 2002, Murphy et al., 1998, frequent stomachaches and headaches (Alaimo et al., 2001a), increased odds of being hospitalized (Cook et al., 2004), greater propensities to have seen a psychologist (Alaimo et al., 2001a), behavior problems Slack and Yoo, 2005, Whitaker et al., 2006, worse developmental outcomes Jyoti et al., 2005, Rose-Jacobs et al., 2008, more chronic illnesses (Weinreb et al., 2002), impaired functioning (Murphy et al., 1998), impaired mental proficiency (Zaslow et al., 2009), and higher levels of iron deficiency with anemia (Skalicky et al., 2006). Perhaps paradoxically, food insecurity has also been associated with higher propensities to be obese Casey et al., 2001, Casey et al., 2006, Dubois et al., 2006, Jyoti et al., 2005. These consistently negative health findings emerge from a variety of data sources, employ a variety of statistical techniques, and appear to be robust to different measures of food insecurity.

Based on this evidence, most authors conclude that efforts to reduce food insecurity would lead to improvements in these health outcomes. That is, if children in food insecure households were to become food secure, they would be expected to achieve health outcomes like those in observationally similar food secure households. The central vehicle for helping alleviate food insecurity among children is the Food Stamp Program, now called the Supplemental Nutrition Assistance Program (SNAP) (U.S. Department of Agriculture, Food and Nutrition Service, p. 7). This program directly augments a household’s resources available for purchasing food. Prior research has suggested that SNAP leads to reductions in food insecurity (e.g., Gundersen and Oliveira, 2001). Other policies, such as the Earned Income Tax Credit (EITC) and educational programs to assist families with food budgeting, may indirectly alleviate food insecurity. More generally, any program that expands a low-income household’s budget opportunities may lead to less food insecurity.

Irrespective of how policymakers pursue these improvements, the general conclusion that efforts to reduce food insecurity would lead to improvements in health outcomes is tenuous for two main reasons. First, food insecurity is not randomly distributed among the population. Even after controlling for characteristics that can be observed in the data, there may remain important unobserved factors that lead some children to be simultaneously at higher risk of being food insecure and of being in poor health. Due to these unobserved influences, a policy prescription that would alleviate food insecurity among these children, even if worthwhile on other grounds, might not lead to the predicted improvements in health status.

Second, this literature presumes that food insecurity is accurately measured in household surveys. However, food insecurity status may be mismeasured for a variety of reasons. Food insecurity is partially subjective, and there may be divergences between how experts and various households interpret the survey questions (Gundersen and Ribar, 2005). Even if food insecurity were objectively defined, some parents might misreport being food secure if they feel ashamed about heading a household in which their children are not getting enough food to eat (Hamelin et al., 2002). Alternatively, some households might misreport being food insecure if they believe that reporting otherwise could jeopardize their eligibility for an assistance program – especially one that provides food assistance, like SNAP. More generally, validation studies consistently reveal large degrees of response error in popular surveys, even for variables thought to be relatively objective. In an important survey of the causes and consequences of measurement error, Bound et al. (2001) conclude that response error constitutes a serious problem for applied econometric work across a wide range of topics. Moreover, they find little reason to believe that such errors tend to occur randomly. Instead, they provide evidence that response errors tend to be correlated with the underlying variable of interest and common socioeconomic attributes. Consistent with this concern, Black et al. (2003) find that more than a third of respondents to the U.S. Census reporting a professional degree have no such degree, with widely varying patterns of false positive and false negative reports across demographic groups. A priori, there seems to be good reason to consider the consequences of at least some small degree of misclassification in food insecurity responses. Even small degrees of classification error can lead to large degrees of uncertainty for inferences.¹

In this paper, we reconsider what can be learned about the effects of food security on child health outcomes when formally accounting for the uncertainty created by unobserved counterfactuals and questions about the reliability of self-reported food insecurity status. In the absence of strong (and untestable) assumptions on counterfactual outcomes and the reporting error process, we cannot fully identify the impact of food security on health. Nevertheless, we can provide informative bounds on these impacts using relatively weak nonparametric assumptions. Our analysis applies and extends recent partial identification bounding methods that allow researchers to consider relatively weak nonparametric assumptions (see, e.g., Manski, 1995, Pepper, 2000, Molinari, 2008, Molinari, forthcoming, Kreider and Pepper, 2007, Gundersen and Kreider, 2008, Kreider and Hill, 2009, Kreider et al., 2009). Using these methods, coupled with data from the 2001–2006 National Health and Nutrition Examination Survey (NHANES), we assess the impact of food insecurity on the health of children. We focus on two key measures studied in previous work: a child’s general health status and obesity status.

In the next section, we describe the central variables of interest in this paper – food insecurity, general health outcomes, and obesity – followed by a description of the 2001–2006 NHANES. In Section 3, we highlight the statistical identification problems created by selection issues and the potential unreliability of self-reported food insecurity. We then show how the average treatment effects of interest can be partially identified under various assumptions about the classification error and selection processes. Section 4 presents our empirical results, and Section 5 concludes.