Socioeconomic disparities in physical health have been found throughout the life course(Adler et al., 1994; Chen, Matthews & Boyce, 2002; Marmot & Wilkinson, 2006). Numerous studies investigating the association between income and adult health have found that income has a “gradient effect.1)” Moreover, recent findings from studies show that adverse socioeconomic experiences during childhood are associated with adult health outcomes(Galobardes, Lynch, & Smith, 2004; Hayward & Gorman, 2004; Power, Manor, & Matthews, 1999). These findings have increased interest in understanding the short-and long-term effects of poverty during childhood on physical health throughout the life course.

Studies that examined the association between income and physical health have found similar gradient effects for both children and adults(Case et al., 2002; Chen et al., 2002; Chen, Martin & Matthew, 2006; Larson & Halfon, 2010) although there is some evidence suggesting that the magnitude of this gradient effect changes as children grow older.2) Other studies have identified the detrimental effect of poverty on children’s physical health outcomes(Aber, Bennett, Conley, & Li, 1999; Evans & Kim, 2007; Weinick, Weigers, & Cohen, 1998). For example, children who livebelow the Federal Poverty Line(FPL) have been found to have higher mortality and morbidity rates than children who live at or above the FPL(Ernst, Demissie, Joseph, Locher, & Becklake, 1995). These findings have also been reported for children living in other countries including Korea(Kwon & Ku, 2010; Lee, Kim, & Kim, 2010).

However, previous studies examining this association have done so mostly using cross-sectional or pooled data(Case et al., 2002; Currie & Stabile, 2003; Newacheck, Jameson, & Halfon, 1994; Weinick et al., 1998). Even studies that investigated age differentials in the effect of income and children’s physical health relied mostly on cross-sectional data and examined the effect by dividing the children into different age groups(e.g., 0-3, 4-8, 9-12, 13-17; Case et al., 2002; Currie & Stabile, 2003). One limitation of this approach is its inability to isolate cohort effects from actual changes in health over time.

In addition, the vast majority of studies that examined socioeconomic disparities and children’s physical health have compared the health of children in poverty to children from higher-income families(Newacheck et al., 1994; Weinick et al., 1998). As a result, little is known about why the health of children from similar socioeconomic backgrounds varies. Researchers who examined the changes and fluctuations of individual health over the life course in adulthood emphasize that there is not a general health trajectory followed by most people, but rather, there are wide variations among individuals depending on current and past life experiences(Aldwin, Spiro, Levenson, & Cupertino, 2001; McDonough & Berglund, 2003).

The present study aims to examine the health trajectories of low-income children and to explore how indicators of socioeconomic status and material hardship are associated with between-individual differences. Particular attention will be paid to examine the individual and cumulative effects of multiple indicators of material hardship on low-income children’s health trajectories. Using four waves of data from Illinois Families Study-Child Well-being Supplement, the present study aims to address the following research questions:

1)This term is used to describe the phenomenon of relatively wealthier people having better health outcome than the less privileged at every level of the income hierarchy (Adler et al., 1994; Case, Lubotsky, and Paxson, 2002), in terms of both mortality and morbidity (see Adler et al., 1994 for a review).  2)Several studies have found that the effect of income becomes stronger as children age(Case et al., 2002; Currie & Stabile, 2003). However, West (1997) suggests that the income effect is strong during the early years of childhood, diminishes during adolescence, and becomes stronger again during young adulthood. Moreover, Chen and her colleagues (2006) found that depending on the outcome, the income effect may stay the same across age groups.

Most studies on longitudinal health trajectories have focused on adult health (Aldwin et al., 2001; McDonough & Berglund, 2003). To the best of our knowledge, only one group of researchers investigated the trajectories of adolescent health employing a latent growth-curve analytic method(Wickrama, Conger, Lorenz, & Elder Jr., 1998; Wickrama, Lorenz, & Conger, 1997). Using a 12-item self-report scale that asked about adolescents’ physical complaints(e.g., headaches, stomach aches, and allergies), Wickrama and colleagues(1997) found that during the four-year study period, adolescents’ physical complaints increased over time. They also reported that trajectories of adolescent physical health are associated with trajectories of parental support, parental education, and parental rejection(Wickrama et al., 1997; Wickrama et al., 1998).

Children in poverty are disproportionately less healthy than children who are not living in poverty(Bor, Najman, Andersen, Morrison, &Williams, 1993; Federico & Liu, 2003; Hughes & Ng, 2003; Newacheck et al., 1994; Newacheck & Starfield, 1988; van der Lucht & Groothoff, 1995; Weinick et al., 1998). Specifically, Larson and Halfon(2010) reported that children in the lowest income group(below 100% FPL) are approximately 3 times more likely to be in worse health(e.g., general health assessment, activity limitations) and approximately 2 times more likely to have health conditions that affect their physical functioning(e.g., diabetes, Asthma, headaches, ear infections) than children in the highest income group(400% FPL or greater). Other studies have found that children from low-income families are more likely to have greater number of days spent sick in bed, have more absences from school, and experience more days of restricted activity due to illness than children from families with higher income(Bor et al., 1993; Federico & Liu, 2003; Flores, Bauchner, Feinstein, & Nguyen, 1999; Newacheck et al., 1994; van der Lucht & Groothoff, 1995).

There is also a growing body of research in Korea that examined the impact of poverty on children’s physical health(Kwon & Ku, 2010; Lee, Kim, & Kim, 2010). For example, Kwon and Ku(2010) investigated the association between poverty and general assessment of children’s health using the data from the Seoul Panel Study of Children. They found that poverty negatively affects children’s health and that this association is mediated by poor parent child relationship and marital conflict(Kwon & Ku, 2010). Similar results were found with more nationally representative samples(Lee, Kim, & Kim, 2010). Studies have also found that family poverty is associated with children’s physical growth, which was measured by height-for-age and body mass index(Pak, Lee, & Kim, 2007; Pak, Lee, Lee, Kim, & An, 2008).

Conventional measures of SES(e.g., income, parental education, employment status) are most commonly used to examine the link between poverty and children’s health. However, scholars emphasize the need to move beyond conventional measures of SES, and incorporate additional economic indicators such as material hardship to better characterize the economic and living conditions of individuals in poverty(Beverly, 2000; Ouellette, Burstein, Long & Beecroft, 2004). Variables that are more proximal in nature may provide additional insight about the relationship between poverty and health, particularly, when examining within-group variations among children in poverty.

Although there is no general consensus about the definition of material hardship, this study focuses on five domains of material hardship(food insecurity/insufficiency,3) housing--quality, overcrowding, and security, basic utilities, unmet medical needs, and durable goods) identified by Ouellette and colleagues(2004). Previous literature examined the associations between children’s physical health and housing related hardships, food insecurity, and unmet medical needs.4) For example, children living inpoor-quality and overcrowded housing are more likely to suffer adverse health consequences, including accidents, and injuries, lead poisoning, and infectious diseases than children in better quality housing(Kim, Sataley, Curtis, & Buchanan, 2002; Pamuk, Makuc, Heck, Reuben, & Lochner, 1998; Sharfstein, Megan, Kahn, & Bauchner, 2001; Wood, 2003). In addition, children from households experiencing food hardship have worse health outcomes than children from low-income families who do not experience any food hardship(Alaimo, Olson, Frongillo, & Briefel, 2001; Olson, 1999; Weinreb et. al., 2002). The nature of the association between unmet medical needs and children’s health is still not clear due to conflicting results from research(Newacheck et al., 2000; Weathers, Minkovitz, O’Campo, & Diener-West, 2004). However, researchers have found that when these medical needs are not met, children are at risk for developing more severe and chronic physical and psychological conditions(Davidoff, Garrett, Makuc, & Schirmer, 2000; Newacheck et al., 2000; Szilagyi & Schor, 1998).

Studies that examine the association between material hardship and children’s physical health have generally concentrated on the impact of a single form of material hardship and seldom investigated multiple dimensions of material hardship(Alaimo et al., 2001; Olson, 1999; Rafferty & Shinn, 1991; Weinreb, Goldberg, Bassuk, & Perloff, 1998; Weinreb et. al., 2002). Although the findings from these studies offer insight into how children’s health is affected by each type of material hardship, they do not provide a clear description of how these interrelated domains of material hardship affect children’s physical health simultaneously. Joint consideration of multiple indicators of material hardship in analytic models is expected to help us better understand relative impact of different types material hardship on children’s physical health trajectory.

Moreover, some researchers assert that the number of co-existing risk factors may have a greater impact on children’s outcome than any specific type of single isk(Sameroff, Seifer, Barocas, Zax, & Greenspan, 1987; Sameroff, Seifer, & Bartko, 1997). Thus, a cumulative risk index where the number of risk factors that children experience is summed as a summary score is utilized to investigate the cumulative effect of risk factors on child outcomes. Currently, the association between indicators of materialhardship and children’s health trajectory is not investigated well, which prevents a better understanding of the relative and cumulative associations between material hardship and children’s health. Thus, this study seeks to examine the differential and cumulative effect of interrelated indicators of material hardship on children’s health trajectory.

Other factors including race, family structure, and health insurance status may confound the associations between socioeconomic status, material hardship, and child health. These variables have all been found to be associated with children’s physical health(Angel & Worobey, 1988; Case et al., 2002; Flores et al., 1999; Montgomery, Kiely, & Pappas, 1996; Schmeer, 2011; Thomas, 2001; Williams, 1999). Racial disparities in health outcomes, above and beyond the intersection of socioeconomic status and race, are found in both adult and child populations(Flores et al., 1999; Shi & Stevens, 2005; Thomas, 2001; Williams, 1999). These studies found that children of color are more likely to be in worse health(Flores et al., 1999; Weinick et al., 1998), less likely to be insured, and less likely to have regular access to health care services(See Stevens & Shi, 2003 for a review) when compared to their counterparts. Similar findings have also been reported for children from middle-class families with private health insurance(Weitzman, Byrd, & Auinger, 1999). Family structure has also been found to be associated with children’s physical health(Angel & Worobey, 1988; Cooper, Arber, & Smaje, 1998). In particular, parents in single-parent families are 1.3 to 1.5 times more likely to report their children as being in poor or fair health(Montgomery et al., 1996) and as having chronic conditions(Cooper et al., 1998) than other parents, controlling for their socioeconomic status.

3)Food insecurity is a concept that has been developed by the United States Department of Agriculture using an 18-item scale. Food insecurity is defined as the “Limited or uncertain availability of nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways” (Nord & Bickel, 2002). On the other hand, food insufficiency is a concept that has been developed by the Community Childhood Hunger Identification Project. Food insufficiency is defined as “inadequacy in the amount of food intake because of lack of money or resources to access enough food,” and is measured by asking the respondents “how often [they did] not get enough food to eat in the last 3 months” (Casey, Szeto, Lensing, Bogle, & Weber, 2001).  4)Among the five domains of material hardship identified by Ouellette and colleagues(2004), I was not able to find previous studies that examined impact of material hardship on children’s physical health for two of the five domains (i.e., difficulty paying for utilities and durable goods). However, difficulty paying for utilities domain was considered in the analyses for consistency. The durable goods domain was not considered in this study due to the lack of adequate measures.

The analyses were conducted using data from the Illinois Families Study(IFS) and the Illinois Families Study-Child Well-being Supplement(IFS-CWB). The explanatory variables were mostly obtained from the first wave of the IFS.5) Information for the outcome variables(i.e., child physical health) was extracted from all four waves of the IFS-CWB.

IFS is a five-year panel study that followed a stratified random sample of 1,899 families who were residing in Illinois and receiving welfare cash benefits in September, 1998(Lewis et al., 2000). The study followed welfare recipients and their families for five years after the implementation of Temporary Assistance to Needy Families(TANF) in Illinois. The aim of the study was to examine the well-being and adaptation of families who were affected by welfare reform. The annual survey asked for information from current and former welfare recipients about their welfare and work experiences, financial hardships, health and well-being, and family characteristics. Individual-level administrative data which included information on welfare receipt and employment were obtained from the Children’s Integrated Database on Child and Family Programs in Illinois(IDB) created by the Chapin Hall Center for Children at the University of Chicago.

Following the first wave of the IFS survey, all families with a child under the age of three were selected for a supplementary study-IFS-CWB. The youngest child in each family was considered to be the “target child,” or the child of interest for the study. Survey questions for IFS-CWB gathered in-depth information about each target child including detailed information about target child’s health conditions, physical needs, medical providers, and access to health care services.

After the initial response rate of 72%(N=1363) during the first wave of IFS in 1999and 2000, the response rates the four waves of IFS-CWB survey were 95%(n=553), 88%(n=484), 92%(n=445), and 91%(N=405), respectively. Since one target child was selected from each household, the total number of families, as well as the total number of target child included in the first wave of IFS-CWB was 553. Sampling weights were created and applied to all the analyses to adjust for the over- representation of sample members from smaller counties and for non-responses. For Hierarchical Generalized Linear Modeling(HGLM) analyses, sampling weights were applied at the individual level(i.e., level-2).

Attrition analyses indicate that those who did not remain in the study were more likely to be African American and have a shorter history of welfare receipt compared to those who remained in the study. To minimize biases associated with differential attrition, all individuals who provided any information regarding the outcomes and full information for level-2 predictors were included in the study. This also reflects the level of flexibility that HGLM allows in dealing with missing observations and making the maximum use of available data to estimate parameters of individual change(Raudenbush & Bryk, 2002; Singer and Willett, 2003).

1) Health outcomes

Two indicators of physical health were examined. For both dependent variables, issues related to unequal periods between interviews were adjusted for by adding an exposure variable(i.e., number of months between each interview) in HGLM models.

Number of sick visits : Caregivers were asked “since your last interview, how many times has the target child gone to see a doctor or other health professional because he or she was sick?” The number of visits counted by the primary caregiver was used as a count variable in the analysis.

Number of sick days in bed : Caregivers were also asked “since your last interview, how many days has the target child stayed home in bed because he or she was sick?” The number of sick days counted by the primary caregiver was included as a count variable.

2) Basic child and caregiver characteristics

Basic demographic characteristics, such as child’s gender, birth weight, and chronic physical conditions were included in the statistical models as dummy variables. Male children, children with chronic physical conditions that limit their activities, children whose birth weight was less than 2,500 gram or children whose birth weight was missing were assigned “1.” Child’s age at W2 IFS interview was included as a continuous variable.

To examine whether health care access affects the associations between socioeconomic status, material hardship, and children’s physical health, target child’s health insurance status, assessed at the W2 IFS interview, was considered as a control variable. The type of children’s health insurance was assessed by asking caregivers whether the “child is currently covered by a health insurance plan” at the time of the interview and “the type of health insurance.” Two dichotomous variables were created: one variable identified children who did not have any health insurance at the time of the interview, and the other variable identified children who had private health insurance. The reference group comprised of children who were on public health insurance, such as Medicaid or the State of Illinois’ Children’s Health Insurance Program(KidCare).

Caregiver characteristics such as age, race and ethnicity, marital status, and household characteristics including region of residence and number of children living in the household were also considered as control variables. Caregiver’s age at the W1 IFS interview was reported in number of years. For caregiver’s race and ethnicity, two dichotomous variables which indicate that the caregiver either 1) identified herself as Hispanic or 2) identified herself as Non-Hispanic white or other race or ethnicity were constructed. The African American caregivers were the reference group. Two dichotomous variables that indicate caregivers who were currently married or who were not married but cohabiting with a significant other were also constructed, leaving the remaining caregivers as the reference group. A region of residence(metropolitan area or downstate Illinois) and number of children living in the household were also included.

3) Socioeconomic characteristics

Variables that measure caregivers’ socioeconomic status including income, education, welfare history, and work experience were included in the analytic model to examine how conventional measures of socioeconomic status explain between-individual differences in low-income children’s physical health trajectories. Household income was measured as an ordinal variable from “1= less than $2,500” to “15=$50,000 or more.” A dichotomous variable that captured caregiver’s level of education was also created (high school graduate or greater or otherwise). For welfare history, the number of months on welfare during a 6-year period prior to the W2 IFS interview year (1995-2000) was computed from the administrative data. Similarly, past work experience was computed as the number of quarters working during the 6-year period prior to the W2 IFS interview year.

4) Material hardship indictors

In order to examine how material hardship affects children’s health, the study utilized two approaches to examine the association. First, each material hardship variable was dichotomized so that “1” indicated high risk and “0” indicated low risk (Appleyard, Egeland, van Dulmen, & Sroufe, 2005; Gutman, Sameroff, & Eccles, 2002).6) To understand the unique effect of each type of material hardship, these indicators were simultaneously included in the analytic model. Second, a cumulative material hardship index was formulated by adding the values for each material hardship item to estimate the number of material hardship experienced by each child.

Housing : Three aspects of housing(housing insecurity, housing quality, and overcrowding) were assessed. Housing insecurity was assessed using a five-item scale on caregivers’ experiences of housing insecurity. Caregivers were asked whether they had experienced more than a two-day period of having to “stay at a homeless shelter or domestic violence shelter,” “live in a car or other vehicle,” “live in an abandoned building,” “live on the streets” or “stay with a friend or relative for less than two weeks because you had nowhere else to go.” Children whose caregivers answered “yes” to at least one of these items were considered to have “housing insecurity.”

Caregivers’ housing conditions were assessed by asking whether any of the seven items(e.g., leaky roof, ceiling, or walls; rats, mice, roaches, or other insects; broken windows) applied to their current living situation. Children whose primary caregiver responded “yes” to two or more problems were considered as “having poor quality housing”(Mayer & Jencks, 1989).

The final aspect of housing quality included a measure of crowdedness computed as a ratio of the number of people per bedroom. Children in households with more than 2 members per bedroom were considered to live in crowded households.

Food hardship : In the present study, we define food hardship as often “having reduced quality and variety of children’s diet”(Nord & Bickel, 2002). The measure of food hardship consists of four child-specific food insecurity items that were derived from the USDA Core Food Security Module(Price, Hamilton, & Cook, 1997). Each item assesses whether a child is experiencing a specific type of food hardship regardless of an indication of hunger. Children whose espondents answered “often” to any of the four items in the scale were considered to be “experiencing food hardship.”

Utilities : Respondents’ financial capacity to afford essential utilities was assessed by asking respondents whether utilities(gas, electricity, oil) had been turned off since the last interview because payments were not made. If the respondent answered “yes” to any of the items, the children were considered to be living in households experiencing “utility hardships.”

Unmet medical needs : In order to assess whether the family was able to afford adequate health care, respondents were asked whether there had been a time when a child in the family, not necessarily the target child in this study, needed medical services from a doctor, prescription drugs, or dental care but did not receive the service because the respondent was not able to afford it. Children were assigned “1” if the respondent answered, “yes” to any of these questions regarding unmet medical needs.

Growth curve analysis using the Hierarchical Generalized Linear Modeling(HGLM) technique was employed in this study. The multilevel modeling technique accounts for within-individual changes over time as well as between-individual differences by dividing the model into two levels. This approach treats the longitudinal data as repeated measures(level-1) nested within individuals(level-2). Level-1 models allow us to estimate and understand how children’s physical health outcomes change over time. Level-2 models examine whether there are significant variations in each person’s health trajectory(Singer & Willett, 2003). Treating repeated measures as nested within individuals allows for greater flexibility in dealing with unequal spacing between observations and missing observations and allows us to make the maximum use of any and/or all available data to estimate parameters of individual change(Raudenbush & Bryk, 2002; Singer and Willett, 2003). In addition, a random effects model was fitted to account for the between-individual differences in low-income children’s health trajectories.

Since the variance for level-1 random effects for count outcome variables is not homogeneous and is dependent on the predicted value, Hierarchical Generalized Linear Models(HGLMs) with varying exposures were estimated. In HGLM, a level-1 link function is used to transform the outcome variable so that it could have a linear function with the explanatory variables. For count data, HGLM uses a Poisson model and a log link function(Raudenbush & Bryk, 2002). If Y_ti is the number of events occurring during a time length of mti, the level-1 unconditional model for the count data can be expressed as:

The expected number of events occurring during a time period of mti would be the product of the length of exposure and the event rate(λ_ti). The link function(η_ti) for a level-1 Poisson sampling model is the log event rate(η_ti) that can have any value ranging from negative to positive. In addition, the time variable, t_ti, is defined as the wave of the IFS-CWB interview. As previously mentioned, the first wave of IFS-CWB interview coincided with the second wave of IFS. As a result, π_0i represents the log event rate of the dependent variable for child i at t_ti=0, which is at W1 IFS-CWB interview. On the other hand, the slope(π_1i) represents the rate of change in the log event rate(Snijders and Bosker, 2002).

The procedures used to fit level-2 models that examine the association between the explanatory variables and these growth parameters(i.e., π_0i and π_1i) are as follows. First, a level-2 unconditional model can be expressed as:

Since π_0i represents the average log event rate at W1 IFS-CWB, β₀₀ can be interpreted as the average log event rate at W1 IFS-CWB. In addition, π_1i represents the expected amount of change in log event rate, and β₁₀ represents the average amount of change in log event rate. The error terms, u_0i and u_1i, reflect the amount of variation that exist between individuals in relation to their growth parameters. These random effects are expected to have variances of τ₀₀ and τ₁₁ with a covariance of τ₀₁.

The level-2 models with explanatory variables(X_qi) can be expressed as:

β₀₀ is the expected log event rate at t_ti=0(W1 IFS-CWB interview); β₀₁ is the amount of change in the expected log event rate at t_ti=0 when there is a unit increase in X_1i; β₁₀ is the expected rate of change in the log event rate; β11 is amount of change in average rate of change in the log event rate when there is a unit increase in X_1i. The random components, u_0i and u_1i, can be interpreted as the random effects of child i on the log event rate at W1 IFS-CWB interview and on the rate of change in the log event rate after accounting for all the variables.

Using these strategies, indicators of material hardship and socioeconomic status, controlling for demographic variables of the child and the caregiver, were employed to estimate the intercept(π_0i) and the slope(π_1i).

5)Because detailed information about the target child’s health care utilization and insurance status was included in the IFS-CWB, access to health care services variables (i.e., regular source of health care and type of health insurance) were extracted from wave 1 IFS-CWB, which coincides with wave 2 of IFS.  6)Although the scales used in this analysis were dichotomized, the statistical properties of the original scales (e.g., housing problems and food hardship) were evaluated. Each scale yielded values of Cronbach’s alpha greater than .70.

Means and standard deviations of the key outcome variables for each wave are presented in <Table 1>. Children had, on average, 2.07 sick visits to the doctor between Wave 1(W1) and Wave 2(W2) IFS interview and spent, on average, a reported 1.41 days in bed due to illness. The mean number of sick visits and number of sick days in bed decreased as each wave progressed.