Social Science & Medicine 170 (2016) 170e179
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
The relationship between maternal education and reported childhood conditions Edward R. Berchick a, b, * a b
Department of Sociology, Duke University, Box 90088, Durham, NC 27708, USA Department of Sociology, Princeton University, Wallace Hall, Princeton, NJ, 08544, USA
a r t i c l e i n f o
a b s t r a c t
Article history: Received 7 July 2016 Received in revised form 16 October 2016 Accepted 18 October 2016 Available online 19 October 2016
Children of more-educated mothers tend to be healthier than children of less-educated mothers. However, in the United States, evidence for this relationship largely focuses on summary measures of health, such as subjective health status, birth weight, and height. Few studies have examined the relationship between mothers' education and children's reported conditions, the health metric that underlies many policy decisions concerning population health. Contrary to stylized facts about socioeconomic gradients in health, higher detection and reporting rates may lead to higher reporting rates among children of more-educated mothers, despite their better underlying health. This reporting pattern that might not mirror gradients for summary health measures. To examine this possibility, I investigate the association between maternal education and nine health conditions in the 1998e2014 National Health Interview Surveys (n ¼ 176,097). I consider variation in the maternal education gradient across the specific reported conditions that children experience, paying particular attention to how patterns differ across children's ages. Results suggest that, unlike for the income gradient in child health, the relationship between maternal education and reported conditions varies in magnitude and direction across conditions. With some exceptions, the probability of reporting a diagnosed condition increases with maternal schooling. For some diagnoses, like asthma, this relationship is curvilinear, with an inverse gradient for children of the most educated mothers. However, the probability of reporting conditions that require neither diagnosis nor substantial parent-child involvement for detection tends to be flat across maternal education. Contrary to expectations, these relationships tend to be more pronounced for children who are 6 years of age or older than for younger children. These results expand understanding of the production and reporting of early-life health inequalities and illustrate limitations of an oft-used health metric. Reported conditions may underestimate socioeconomic inequalities in children's health. © 2016 Elsevier Ltd. All rights reserved.
Keywords: Child health Disparities Education Socioeconomic status Intergenerational processes Self-reported health
Parents' educational attainmentdparticularly mothers' attainmentdinfluences children's health. Children of more-educated mothers have better subjective health, taller stature, lower obesity rates, and higher birth weights than children of mothers who completed just one less year of schooling (e.g., Currie and Moretti, 2005; Gakidou et al., 2010; Nazarov and Rendall, 2011). Yet most research on the education gradient in child health in the United States focuses on these summary measures of health. With the exception of asthma, injuries, and a few other conditions (see Chen et al., 2002), it is unknown whether there is also a maternal
* Duke University, Box 90088, Durham, NC 27708, USA. E-mail address:
[email protected]. http://dx.doi.org/10.1016/j.socscimed.2016.10.018 0277-9536/© 2016 Elsevier Ltd. All rights reserved.
education gradient in children's likelihood of having a specific parent- or physician-identified illness. I focus on children's reported health conditions and diagnoses because they constitute an important population health metric. Descriptions of the health of the American population, policy decisions concerning child health, and assessments of population health each rely on reported symptoms or diagnoses (e.g., U.S. Department of Health and Human Services 2012). There is nonetheless a crucial tension in parent-reported conditions as a measure of children's health. Reported health conditions are paradoxically markers of both health disadvantage and health advantage. The report of a health condition means that a child is displaying symptoms that affect his quality of life and, possibly, his ability to interact with others around him. However,
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
parental reports of symptoms and diagnoses also mean that a child's parents have recognized health symptoms and/or taken the child to receive medical care. To the extent that mothers are picking up on true symptoms and not over-diagnosing benign symptoms, detection and treatment can reduce illness severity and potentially diminish poor health's consequences for the child's life chances. In this article, I use data from the 1998e2014 National Health Interview Survey (NHIS) Sample Child Files and ask: Is there a maternal education gradient in children's reported health diagnoses and conditions? As gradients may not be in the same direction for all reported conditions (due to differences in their detection and diagnosis), I focus on nine major child illnesses. I also pay particular attention to how patterns differ across children's ages.
1. Background 1.1. Evidence for positive gradients Education gradients in summary health measures (such as birth weight and subjective health) and income gradients in diagnosed health conditions (such as asthma, diabetes, and epilepsy) lend support to the maxim that “more education means better health.” One possibility is that the relationship between mothers' education and children's health looks similar to the maternal education gradient in subjective health: the share of children with a condition may decrease with increasing maternal educational attainment. Each year of educational attainment carries additional knowledge, cognitive and non-cognitive skills, and psychological resources that can buffer against stress and negative health events (see Mirowsky and Ross, 2003; Ross and Wu, 1995). These resources and behaviors also influence how mothers engage with their children's health: Mothers can use psychological resources and health knowledge to prevent illness, recognize early symptoms, and navigate health systems to promote their children's health. Some research suggests mothers' educational attainment promotes these skills and behaviors (see Currie, 2009), whereas other research suggests that education and health preferences are jointly determined by other existing factors (Fuchs, 1982). For example, certain individuals' longer time horizons and greater perceived sense of control may prompt them both to attain more schooling and to make greater health investments. Children of more-educated mothers also tend to live in smaller families with both parents present and have greater income than children of mothers who attained less education. Income and these family structures, in turn, promote favorable child health (Currie, 2009; McLanahan and Percheski, 2008; Ziol-Guest and Dunifon, 2014). Moreover, children of educated mothers are more likely to have health insurance coverage than those from humbler origins, partly because of better-educated mothers’ greater income and greater considerations of long-term investments. They are also more likely to have seen a doctor or a dentist within the last year and to have a usual source of medical care compared to children of mothers who attained less education (Case and Paxson, 2001, 2002). Likewise, even when accessing care, more-educated mothers differ from less-educated mothers. They may have the resources to know about doctors who are more willing to give certain diagnoses (Liu et al., 2010), and ethnographic accounts reveal that socioeconomically advantaged mothers and their children tend to be more assertive and proactive in encounters with physicians than are mothers of low socioeconomic status and their children (Lareau, 2003). Thus, maternal education shapes many health-relevant aspects of the child's environment.
171
1.2. Evidence for negative gradients However, reports of illnesses do not perfectly capture a child's true health status, as reported health conditions must be observed and/or diagnosed. Reporting differences rather than differences in the underlying “objective” risk of a health event may shape the maternal education gradient in health conditions. Although children of less educated mothers may be more sick than their more advantaged counterparts (due to differences in income, neighborhoods, and so forth), estimates of children's health conditions come from parental reports of either observed health symptoms (e.g., diarrhea) or physician diagnoses (e.g., asthma). As a result, the maternal education gradient in reported conditions may not reflect the maternal education gradient in children's true epidemiological risk. Specifically, a number of differences across levels of educational attainment mean that children of mothers who completed more education could be more likely to have a reported health condition than children of mothers with lower educational attainments, regardless of the incidence of actual illness. First, more-educated mothers’ greater knowledge and health conscientiousness (Mirowsky and Ross, 2003) mean that more-educated mothers may be better able to identify low-severity or subtle symptoms than are less-educated mothers. More-educated mothers also may be better positioned than less-educated mothers to recognize symptoms in the first place. College-educated mothers spend more time in child-engaged activities compared to less-educated women (Guryan et al., 2008; Leibowitz, 1974; Sayer et al., 2004), allowing them to monitor their children's wellbeing more closely. Thus, higher-quality parent-child interactions may offer advantaged mothers more opportunities than are available to less-educated mothers to recognize minor symptoms, identify a condition, find the appropriate source of care, and take their children to it (King and Bearman, 2011). They are also more likely to have health insurance, to have health care access, to obtain information via social networks, to live in better-off neighborhoods, and to live in areas where awareness of certain conditions is more prevalent (Case and Paxson, 2001; King and Bearman, 2011). 1.3. Variation across conditions Health conditions vary in their causes and consequences and in their detection and reporting patterns in ways that suggest that the magnitude and direction of the maternal education gradient in health conditions may differ across conditions. Compared to their less educated counterparts, more-educated mothers could have children with better objective health but report more child health conditions. Grouping conditions on the basis of the degree to which they require either health knowledge and parent-child interaction for identification or medical access for diagnosis, therefore, can help tease apart the relative contributions of parental resources. In this paper, I consider three categories of reported health conditions. The first grouping includes conditions that are diagnosed by a physician (e.g., autism, asthma). In reporting these conditions, the NHIS questionnaire asks parents “Has your child ever been diagnosed with the following condition(s) …” or some variant of that question (Minnesota Population Center and State Health Access Data Assistance Center, 2015). A diagnosis, by definition, means that mothers must recognize symptoms, access medical care, and engage with a physician before they can possibly report a diagnosed condition. As a result, children with moreeducated mothers are hypothesized to be more likely to be reported as having a physician-diagnosed health condition than children of less-educated mothers (Hypothesis 1).
172
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
Existing research offers mixed support for a positive gradient in physician-diagnosed conditions, variously reporting a positive, negative, or flat maternal education gradient in asthma diagnosis (Chen et al., 2002). For example, consistent with hypotheses, Cunningham et al., (1996) find that maternal education is positively associated with asthma. However, Weitzman et al., (1990) find that children of less-educated mothers had 70 percent increased odds of asthma than did children of more-educated mothers. However, the maternal education gradient for these conditions might appear quadratic due to countervalence of the risk of actual disease incidence and the risk of identification/diagnosis (Hypothesis 1a). For example, knowledge about conditions may not monotonically increase with each year of maternal schooling; after a certain level (e.g., college), education may not offer any additional resources that affect perceptions of child health. If this plateauing occurs, then we would observe a non-linear gradient in which increasing levels of maternal schooling above a certain point decrease children's likelihood of having a reported condition among children of the most educated women. Such a decrease could be attributable to more-educated mothers' greater income and propensity to live in family configurations conducive to child health (Currie, 2009; Ziol-Guest and Dunifon, 2014), which would reduce the objective probability of actually experiencing a health event. In other words, there may be an inverse-U-shaped gradient in reported child health diagnoses. A second category of reported health conditions includes conditions with two properties: (i) diagnosis is not a precondition for parental report (see Table 1) and (ii) detection requires parents to observe children over extended periods. Recognizing that a child complains of stomach pains only after a particular subset of foods (i.e., has a food allergy) may require repeated parental involvement to discern the allergic pattern (James, 2003). Although diagnosis by a physician is not necessary based on question wording, parents may also take their children to an allergist to identify the condition. Social awareness and surveillance also affect a parent's behaviors (Waggoner, 2013). Thus, the advantages associated with increased education (such as knowledge and time spent with children) are likely to matter for detecting these conditions. Children of moreeducated mothers may be more likely to be reported as having these conditions than their peers born to less-educated mothers (Hypothesis 2). The third group of health conditions I consider are those based on identified symptoms and are readily observable and identifiable to mothers (or teachers) (e.g., diarrhea, ear infections, and stutter/stammer). The increased knowledge and skills associated with greater educational attainment are unlikely to matter for detecting these conditions; for example, given its symptoms,
whether a child has diarrhea is self-evident. Yet other healthprotective advantages conferred by education likely reduce the risk of experiencing these illnesses (see Case et al., 2002). Therefore, children of mothers who completed more education may be less likely to have a readily observable health condition than are children of mothers who completed less schooling (Hypothesis 3). Alternatively, mothers' education may be unrelated to children's chances of having one of these conditions (i.e., the gradient may be flat across levels of educational attainment), as they are difficult to prevent (e.g., ear infections, stomach bugs picked up from classmates) (Hypothesis 3a). Education would influence neither the incidence of these health condition nor the recognition of their symptoms.
1.4. Age differences We may expect differences in the strength and magnitude of the association between mothers' education and children's health across childhood. Both the education gradient in subjective health and the income gradient in diagnosed conditions increase in magnitude across childhood (Case et al., 2002; Currie et al., 2008). Age differences could extend to the education gradient in diagnosed conditions. First, parental investments vary substantially by age. Education disparities in parental economic and time investments in their children are most pronounced before age five and almost disappear by adolescence (Aguiar and Hurst, 2009; Hurst, 2010). For children 0e5 years old, college educated mothers spend 9.4 more hours a week on child care than mothers who completed less education, but that difference narrows to only 1 h for children 13e17 years old (Hurst, 2010). Second, older children are less dependent on their parents to identify health issues than their younger peers. Older children spend more time outside the home, interact with adults who may detect symptoms (e.g., teachers and school nurses), and can better articulate and identify their own symptoms. For these reasons, there may be weaker gradients during primary school (ages 6e12) and adolescence (ages 13e17) than during the parent-intensive pre-schooling period (ages 0e5) (Hypothesis 4). The foregoing considerations lead me to ask: To what extent is maternal education associated with children's reported health conditions? Specifically, is there variation in the maternal education gradient across the specific reported conditions that children experience? For each of these questions, I pay particular attention to how patterns differ across children's ages.
Table 1 List of selected reported health conditions and prevalence, 1998e2014 National Health Interview Surveys. Condition
Description
Prevalence, by child age 0e5 yrs
6e12 yrs
13e17 yrs
All
Asthma Allergies Respiratory Skin Food Frequent diarrhea Ear infection Stutter/stammer ADHD
Ever diagnosed with asthma
8.2
14.3
16.2
12.8
Respiratory allergy or hay fever in past 12 months Skin allergy in past 12 months Food allergy in past 12 months Frequent or repeated diarrhea in past 12 months 3 ear infections in past 12 months Stutter/stammer in past 12 months Ever diagnosed as having attention deficit disorder or attention deficit hyperactivity disorder Ever diagnosed with autism
8.1 11.5 4.6 2.1 10.5 2.3a 1.4b
12.8 10.3 4.4 1.2 4.5 1.8 7.8
13.3 8.7 4.3 1.3 2.2 1.1 9.9
11.4 10.2 4.4 1.5 5.8 1.7 6.8
0.5b
0.9
0.7
0.7
Autism a b
Prevalence for 3e5 year olds; condition not asked for 0e2 year olds. Prevalence for 2e5 year olds; condition not asked for 0e1 year olds.
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
2. Data and methods 2.1. Data Analyses used data from the Child Sample files of the 1998e2014 waves of the National Health Interview Survey (NHIS). The NHIS is an annual snapshot of the health of the noninstitutionalized American population. In each year, the NHIS selected households (via a multi-stage probability design) and collected information about the health and demographic characteristics of all household members. In the Sample Child files, a knowledgeable adult was asked to report detailed information about health conditions for a randomly selected child from the household roster. NHIS data used in analyses were obtained from the University of Minnesota's Integrated Health Interview Series (Minnesota Population Center and State Health Access Data Assistance Center, 2015). Of the 226,653 selected “sample children” in the 1998e2014 NHIS, 206,097 completed the survey. Following the precedent of prior research (e.g., Case et al., 2002; Currie et al., 2008), analyses were limited to children who could be linked to their family characteristics. Therefore, all children who were not the sons or daughters of the household reference person and/or spouse, children who were not living with either parent, and children who were not members of the primary family within the household were excluded. I also excluded children of mothers who were less than 18 years of age at the time of the NHIS survey and children whose parents did not complete the sample child questionnaire. Missing data were handled by multiple imputation, but results were robust to listwise deletion. The exclusions above reduce the sample size to 176,097 children, but, as described below, the sample size varies across conditions. 2.2. Measures 2.2.1. Health conditions Table 1 describes the nine health conditions included in analyses. These conditions cover a wide range of health phenomena from which children suffer and are all family reported. There was substantial variation in prevalence across conditions: less than 1% of children had an autism diagnosis at any age (the least common condition) and about 16.2% of adolescents had asthma (the most common condition). In general, the prevalence of conditions increased with children's age. 2.2.2. Maternal education As my focus is on education gradients in child health, I measured mothers' education by years of completed schooling. The NHIS asked respondents to their years of educational attainment with the highest grade completed if they completed 12 or fewer years (0e12) and a categorical measure of their highest credential for schooling if they completed more than 12 years (associate's degree, bachelor's degree, etc.). Following the convention of prior research (e.g., Lynch, 2006; Hayward et al., 2015), I converted the categorical measure to a continuous one by assigning the midpoint of the interval or assigning some college and associate degrees as equal to 14 years of schooling, college as equal to 16 years, and greater than college as equal to 18 years. For ease of interpretation, I centered maternal educational attainment on 12 years of schooling. All models include education and education-squared due to the aforementioned non-linearity that could emerge for the children of the most educated mothers. A non-linearity could emerge as the material advantages earned by the most educated mothers (e.g., increased earnings) begin to outweigh detection differences. They might also result from a
173
reduction in the amount of health-relevant knowledge and behaviors associated with each additional year of schooling. Alternatively, an education-squared term might reveal an even steeper gradient for children of the most educated mothers. The small number of children with certain conditions did not allow additional education terms (e.g., interactions between years of schooling and educational credentials or higher-order education terms). 2.2.3. Other family resources I measured family income in logged adjusted dollars. I converted the categorical NHIS measure to a continuous one using the same method as prior literature (e.g., Currie et al., 2008). Specifically, for each NHIS income bracket in each survey year, I calculated the mean family income for CPS household heads whose education and race match the NHIS family's household and whose income falls with the income bracket. I then adjusted this dollar amount to 1999 dollars using the Consumer Price Index (Bureau of Labor Statistics, 2016) and logged the adjusted amount to account for the skewed distribution. I measured whether a child's father was present with a binary indicator (1 ¼ present). I also included a measure of mothers' age at children's birth. As there is evidence that children of mothers who were aged 25e34 at the time of birth are healthier than children € and Fenelon, born to younger or older mothers (e.g., Myrskyla 2012), I included separate indicators for whether a mother was aged < 25 years or > 34 years at the time of birth (reference: mothers aged 25e34 years). I also included a child's number of siblings (squared-rooted to capture economies of scale) and maternal employment status (employed part time (<40 h), employed full time, not employed) (see Craig and Mullan, 2010; Downey, 1995). 2.2.4. Covariates All models controlled for child's age (in years), sex (male, female), race (white, black, or other), ethnicity (Hispanic, nonHispanic), and region (South, other region). 2.3. Analytic strategy I first regressed each of the health conditions on maternal schooling and the basic covariates described above. In these initial models, mothers' education captures human capital as well as other family socioeconomic resources. To disentangle the association between mothers' education and children's health from the contribution of the other resources, I then introduced measures of family income, maternal age, father's presence, father's education, and health insurance. All models were estimated with linear probability models, as there are concerns about the comparability of coefficients across logistic regression estimate (Allison, 1999; Mood, 2010). All patterns were nonetheless robust to logistic regression. In line with other research that has used the NHIS Sample Child files (e.g., Mehta et al., 2013), I included respondents insofar as they had information for a given health condition, even if respondents were missing data on other conditions. For example, a child whose asthma status was unknown was included in autism models if his parent provided diagnosis information for that condition. Therefore, sample sizes vary across health outcomes and are listed in each table. After estimating the model for children of all ages, I then estimated models separately for three age groups: 0e5 year olds, 6e12 year olds, and 13e17 year olds. These age divisions roughly capture children's health before formal schooling, during primary school, and during adolescence, respectively. Table 2 describes the sample's characteristics.
174
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
Table 2 Percents or means (and standard errors) for key social and demographic characteristics and child health conditions, by maternal education. Maternal education
Female (%) Age (yrs) South (%) Income (1999 USD) Mother employed (%) >40 ha Age at birth <25 years (%) >34 years (%) Father present (%) Siblings (#) Health (%)b Asthma Allergies Respiratory Skin Food Frequent diarrhea Ear infection Stutter/stammer ADHD Autism
<12 years
12e15 years
16þ years
49.8 8.4 (0.04) 38.1 16,729.5 (0.01) 36.7 21.0
49.0 8.7 (0.02) 36.4 35,683.7 (0.01) 63.0 37.9
48.8 8.2 (0.03) 32.9 76,797.7 (0.01) 66.7 42.0
16.4 11.0 69.5 1.2
8.4 12.2 75.6 1.0
1.2 24.7 89.5 0.9
12.1
13.9
10.8
7.5 7.2 3.0 1.7 6.6 2.8 5.9 0.4
11.8 10.5 4.3 1.6 5.8 1.7 7.6 0.7
12.5 11.2 5.2 1.2 5.5 1.1 5.7 0.8
a
Conditional on employment. See Table 1 for additional detail. Source: 1998e2014 National Health Interview Surveys. b
3. Results 3.1. Results by type of condition As noted earlier, there are differences in the extent to which parental knowledge and conscientiousness may be required to recognize symptoms or seek a diagnosis and differences in terms of whether the question asked if a child was diagnosed with a condition. Therefore, I examined the maternal education gradient in the nine conditions described in Table 1, roughly grouped by these factors (Table 3). For each condition, I first regressed the health condition on maternal education, maternal education squared (to capture non-linearities that could result from differences in the risk of illness versus the risk of identifying/diagnosing symptoms), and basic covariates (age, sex, race/ethnicity, and region). Model 2 introduced measures of family socioeconomic resources (income, family size, maternal age, father presence, and father's education) and health insurance status. The first row of Table 3 includes three physician-diagnosed conditions: asthma, autism, and attention-deficit/hyperactivity disorder (ADHD). The report of one of these conditions means that a parent has been able to identify symptoms, take her child to a doctor, describe the child's symptoms to an appropriate source of medical care, and receive a diagnosis. Results for this set of conditions offer mixed support for hypotheses. Consistent with Hypothesis 1a, mother's association is weakly and curvilinearly associated with asthma, the most prevalent condition in the analyses, as indicated by the statistically significant education terms. A positive main effect and negative quadratic effect produce an inverse-U-shaped pattern. For children of the least educated mothers, each year of schooling was associated with an increase in the probability of having a reported health condition; additional schooling was associated with a decrease in the probability of having a reported health condition for children of mothers who completed more than high school. Introducing other socioeconomic resources to the model (the second column for each condition) tended to suggest a stronger
gradient than did the unadjusted model, particularly at higher levels of maternal schooling. Although the socioeconomic measures are correlated, accounting for other advantages helps to disentangle maternal education, skill, and health preferences from families' ability to afford better nutrition, neighborhoods, and medical care (Currie, 2009; Evans et al., 2012). These results suggest that the importance of mother's education for the surveillance and diagnosis of asthma is stronger taking into account the other health advantages that children of educated mothers enjoy. That is, differences in behaviors and skills are larger net of differences in the ability to afford health-promoting resources and health care access. The general inverse-U-shaped pattern is consistent with the aforementioned argument (Hypothesis 1a) that, among more educated mothers, health-conscientious characteristics may plateau in their importance for obtaining asthma diagnoses but may still reduce children's underlying health risks. The increased knowledge and ability to recognize symptoms associated with each year of schooling increased the likelihood that a parent identified and reported a condition, but only to a certain point. Above that point (around 14 years of educational attainment), knowledge did not appreciably increase asthma diagnosis, whereas other parental resources (such as income and family structure) could potentially reduce the risk of a health event. Yet this pattern did not extend to other aspects of socioeconomic resources, such as family income and paternal education. Family income has been the focus of most research on socioeconomic gradients in American children's health (e.g., Case et al., 2002; Currie et al., 2008). As noted in Table 3, family income was negatively associated with asthma. Each change in the probability of having a diagnosis was two-and-a-half-times larger with a percent increase in family income than it was with each year of maternal schooling. Additionally, fathers' education, net of mothers' education and family income, was not associated with asthma diagnosis. Results were less consistent with hypotheses (Hypotheses 1 and 1a) for the other two diagnoses examined: autism and ADHD. A curvilinear maternal education gradient in ADHD suggested that the probability of diagnosis was highest for children of the least and most educated mothers. However, accounting for socioeconomic and health insurance differences across levels of education, the relationship reversed and became inverted-U shaped. Though this curvilinear relationship between more maternal schooling and higher probability of ADHD diagnosis was statistically significant, it was relatively weak. Additionally, as with asthma, the direction of the gradient contrasts with evidence of the family income's protective influence. Maternal education, however, was linearly associated with autism diagnosis (consistent with Hypothesis 1), but the small magnitude of the associationda 0.03 percentage increase in probability of diagnosis for each year of maternal educationdsuggested that the maternal education gradient in autism was relatively flat for the average child. Other work focused on select geographies, such as California, suggests that there may be strong socioeconomic patterns that are related to social influence (e.g., King and Bearman, 2011), but the relative rarity of autism in the United States (and, therefore, in the NHIS sample) may have obscured these underlying processes. Moreover, these generally weak patterns could have been due to offsetting influences of lower probabilities of having the condition but higher probabilities of detection and diagnosis (Hypothesis 1a). The second row of Table 3 includes three allergies, conditions that are not necessarily physician diagnosed but that require repeated parental involvement and/or conscientiousness to discern the allergic pattern. Findings for the three types of allergiesdrespiratory, skin, and food allergiesdwere also hypothesized
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
175
Table 3 Relationship between mothers' education and children's reported/diagnosed health conditions. Condition
Education Education2 Income Paternal ed. Intercept
Education Education2 Income Paternal ed. Intercept
Education Education2 Income Paternal ed. Intercept
Asthma
ADHD
Autism
(1)
(1)
(1)
(1)
(1)
(2)
0.0007 (0.0004)* 0.0006 (0.0001)*** e e 0.1533 (0.0019)***
0.0026 (0.0004)*** 0.0005 (0.00003)*** 0.0091 (0.001)*** 0.0007 (0.0004) 0.1517 (0.0021)***
0.0010 (0.0003)*** 0.0004 (0.00003)*** e e 0.0784 (0.0015)***
0.0016 (0.0001)*** 0.0003 (0.00003)*** 0.0094 (0.0012)*** 0.0020 (0.0003)*** 0.0722 (0.0017)***
0.0003 (0.0001)*** 0.00001 (0.00001) e e 0.0093 (0.0004)***
0.0003 (0.0001)** 0.000004 (0.00001) 0.0014 (0.0004)*** 0.0017 (0.0001) 0.0105 (0.0006)***
Respiratory allergies (1)
(2)
Skin allergies (1)
(2)
Food allergies (1)
(2)
0.0051 (0.0003)*** 0.0001 (0.00004) e e 0.1202 (0.0017)*
0.0047 (0.0004)*** 0.0001 (0.00004) 0.0029 (0.0015)*** 0.0003 (0.0004) 0.1302 (0.0021)***
0.0045 (0.0003)*** 0.0001 (0.00004)** e e 0.1030 (0.0015)***
0.0045 (0.0004)*** 0.0001 (0.00004) 0.0019 (0.0014) 0.0002 (0.0004) 0.1050 (0.0019)***
0.0023 (0.0002)*** 0.00002 (0.00003) e e 0.0385 (0.0010)***
0.0022 (0.0003)*** 0.00003 (0.00003) 0.0015 (0.0009) 0.0007 (0.0003)* 0.0426 (0.0012)***
Frequent diarrhea (1)
(2)
Stutter/stammer (1)
(2)
Ear infections (1)
(2)
0.0005 (0.0002)** 0.0001 (0.00002)** e e 0.0121 (0.0006)***
0.0001 (0.0002) 0.00003 (0.00002) 0.0027 (0.0006)*** 0.0001 (0.0002) 0.0111 (0.0008)***
0.0013 (0.0002)*** 0.0001 (0.00002)** e e 0.0211 (0.0007)***
0.0004 (0.0002)* 0.00002 (0.00002) 0.0045 (0.0007)*** 0.00003 (0.0002) 0.0190 (0.0009)***
0.0011 (0.0003)*** 0.0001 (0.00003)** e e 0.0540 (0.0012)***
0.0001 (0.0003) 0.00004 (0.00003) 0.0056 (0.0003)*** 0.0011 (0.0003)*** 0.0523 (0.0014)***
*p < 0.05, **p < 0.01, ***p < 0.001. Model 1 adjusts for sex, race, ethnicity, region, age, and survey year. Model 2 introduces maternal employment, father presence, maternal age at birth, number of siblings, family income, paternal education, and health insurance status. Source: 1998e2014 National Health Interview Surveys.
(Hypothesis 2) to have positive gradients because prior research suggests that more-educated mothers spend more child-focused time with their children than less-educated mothers and are likely more health conscientious (Guryan et al., 2008; Sayer et al., 2004; Ziol-Guest and Dunifon, 2014). Consistent with this perspective, children's likelihood of having respiratory, skin, and food allergies increased linearly with maternal schooling. The coefficients on maternal education were positive and the coefficients on education-squared were insignificant. Although modest, the gradients were larger than for diagnoses. For example, each year of maternal schooling was associated with a half of a percentage point increase in the probability of having a respiratory or skin allergy. Accounting for moreeducated mothers' other advantages did not appreciably change the shape or magnitude of the gradient for any of the three allergies. Surprisingly, net of education, family income was not associated with the probability of reporting skin or food allergies. The final row considers the last three reported health conditionsdfrequent diarrhea, stutter/stammer, and ear infections. These illnesses involve symptoms that are immediately recognizable and are of a long duration. Educated mothers’ knowledge and increased access to health care were not expected to substantially influence the identification of these conditions, and thus we would expect to find negative (Hypothesis 3) or flat (Hypothesis 3a) gradients in the relationship between maternal education and these conditions. Consistent with the latter prediction, there were few substantively meaningful differences in a child's likelihood of having frequent diarrhea, ear infections, or stutter/stammer across levels of maternal schooling. Most of the statistically significant differencesdwhich were generally negative maternal education gradientsdwere likely due to other socioeconomic differences across levels of maternal schooling, as they were no longer significant after introducing other resources into the models. This finding may be attributable to the non-preventable nature of many of these conditions. For example, classmates and playmates expose even the most advantaged children to various diarrhea-causing infections.
What, then, do these differences mean for the average child? Fig. 1 plots the predicted maternal education gradient for each specific condition based on the estimates for both models 1 and 2 in Table 3. These plots demonstrate that, unlike with the income gradient in specific childhood health conditions and diagnoses (Case et al., 2002), the maternal education gradient was weak (i.e. basically flat) or curvilinear for most reported conditions. The inverse-U-shaped patterns for asthma and ADHD each peaked for mothers with some college. Among children of mothers who completed less schooling, educational attainment increased the chances of having either diagnosis. However, for children of mothers who completed some college, each year of schooling was associated with a lower probability of having any reported condition. Even for these conditions and those with linear gradients, the probability of having a health condition does not change substantially across years of educational attainmentdslopes in Fig. 1 are relatively modest. 3.2. Age-stratified results Do these patterns differ across childhood? Children aged five years and younger enjoy more parent-child interaction than do older children (Aguiar and Hurst, 2009). School-aged children spend fewer time with their mothers and are likely at greater risk of symptom detection due to teachers, school nurses, coaches, and other adults. These other individuals may reduce the importance of maternal education for symptom surveillance and recognition, flattening or evening reversing gradients that were present in earlier in life (Hypothesis 4). Table 4 presents the results for the full-control model (Table 3, Model 2) for each condition across three age ranges: 0e5 years, 6e12 years, and 13e17 years. These three intervals roughly correspond to the pre-schooling, early schooling, and adolescence periods, respectively. In general, the maternal education gradient in children's reported conditions exhibited clear age patterns. For nearly all conditions, children's probability of having reported conditions (the
176
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
Fig. 1. Predicted maternal education gradient in reported health conditions. Source: 1998e2014 National Health Interview Surveys.
intercept of the lines) increased as they entered primary school, i.e. after age 5. However, adolescents generally experienced the same association between mothers' education and health conditions as their primary-school-aged peers. The gradient predicted from estimates (Fig. 2) suggests that part of this weaker gradient may stem from a ceiling effect, as the youngest children are relatively healthy compared to older children. Notably, for all age ranges, both family income and father's education were associated with lower probabilities of reported conditions (not shown), suggesting that maternal education had a different relationship with children's reported health conditions than did other family socioeconomic resources. This pattern was inconsistent with Hypothesis 4; moving from the most parent-child-time intensive ages to primary school settings to adolescence strengthened, not weakened, the importance of parental resources. Exceptions include conditions for which there was no maternal education gradient at any age (e.g., frequent diarrhea, stutter/tamer), and skin allergies. Part of the finding for the latter may be due to older children's non-use of diapers and their increasing independence in dressing (where mothers might have discovered skin irritations).
4. Discussion Reported conditions serve as a crucial metric for assessing the health status of American children. Decisions about the allocation of medical resources and assessments of progress toward ameliorating health disparities, for example, each depend on estimates of diagnosed conditions (e.g., U.S. Department of Health and Human Services 2012). Although prior research has investigated the relationship between family income and reported health conditions and between maternal education and subjective measures of health, there is a lack of research examining children's health conditions across levels of maternal education. In contrast to the education gradient in summary measures of health (e.g., subjective health, height) and the income gradient in diagnosed conditions (e.g., asthma, diabetes, digestive disorders), the shape and magnitude of the maternal education gradient in specific childhood health conditions varied across health conditions and was not always positive. Only for a handful of conditions did the dictum that more education implies better health hold true. For other conditions, either inverse-U-shaped or flat gradients were observed, suggesting that most mothers' education influences
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
177
Table 4 The relationship between mothers' education and children's reported/diagnosed health conditions, by child's age. Child's age
Asthma Education Education2 ADHD Education Education2 Autism Education Education2 Resp. allergies Education Education2 Skin allergies Education Education2 Food allergies Education Education2 Freq. diarrhea Education Education2 Stutter/stammer Education Education2 Ear infection Education Education2
0e5 years
6e12 years
13e17 years
0.0006 (0.0006) 0.0003 (0.0001)***
0.0043 (0.0007)*** 0.0005 (0.0001)***
0.0035 (0.0010)*** 0.0006 (0.0001)***
e e
0.0009 (0.0006) 0.0005 (0.0001)***
0.0037 (0.0007)*** 0.0003 (0.0001)**
e e
0.0005 (0.0002)* 0.00002 (0.00002)
0.0006 (0.0002)** 0.00003 (0.00002)
0.0027 (0.0006)*** 0.0002 (0.0001)*
0.0065 (0.0007)*** 0.0001 (0.0001)
0.0051 (0.0008)*** 0.00003 (0.0001)
0.0059 (0.0008)*** 0.0001 (0.0001)
0.0039 (0.0006)*** 0.0001 (0.0001)
0.0037 (0.0007)*** 0.0001 (0.0001)
0.0023 (0.0005)*** 0.00003 (0.0001)
0.0020 (0.0004)*** 0.00002 (0.0001)
0.0025 (0.0005)*** 0.00002 (0.0001)
0.0003 (0.0003) 0.0001 (0.00003)
0.00001 (0.0003) 0.00001 (0.00003)
0.0006 (0.0003) 0.00001 (0.00004)
0.0002 (0.0004) 0.00004 (0.00004)
0.0006 (0.0003) 0.0001 (0.00003)
0.0004 (0.0003) 0.00003 (0.00003)
0.0015 (0.0006)** 0.0002 (0.0001)
0.0010 (0.0005)* 0.00004 (0.0001)
0.0003 (0.0004) 0.00003 (0.0001)
*p < 0.05, **p < 0.01, ***p < 0.001. All models adjust for sex, race, ethnicity, region, age, survey year, maternal employment, father presence, maternal age at birth, number of siblings, family income, paternal education, and health insurance status. Source: 1998e2014 National Health Interview Surveys.
children's health in a way that is distinct from other aspects of family socioeconomic circumstance. Specifically, estimates of education gradients in reported child health conditions and diagnosis, unlike other socioeconomic gradients, appeared to be influenced by both health and detection differences across levels of maternal education. The observed positive associations between maternal education and child health were likely attributable to reporting differences, not poorer health among advantaged children. Rather, findings likely revealed the ways in which mothers who attained more education tend to be better positioned to notice the early onset of symptoms or lowseverity symptoms than those who attained less education (King and Bearman, 2011; Ziol-Guest and Dunifon, 2014). For example, educated mothers are more involved in their children's activities and more closely monitor their children's behaviors than those who attain less education (Suizzo and Stapleton, 2007). They are also more likely than less-educated mothers to be health conscientious and to possess greater health knowledge (Currie, 2009). These behaviors and preferences may mean that more-educated mothers are better positioned than less-educated mothers either to over-diagnose minor conditions or to detect emerging illnesses before they become severe. Consistent with this interpretation, children of more-educated mothers were more likely than children of less-educated mothers to have reported conditions that require health care access or recognition of symptoms that are generally not immediately identifiable (e.g., asthma, respiratory allergies). In contrast, children of more-educated mothers were no more or less likely to have conditions that have obvious, recognizable symptoms, such as stuttering and diarrhea, than their less advantaged peers. The recognizability of symptoms and their consequences means that identification of these conditions is not sensitive to a mother's
medical knowledge or the time she spends with her children. Because mothers must recognize sometimes low-severity and difficult-to-identify symptoms (e.g., with autism) over repeated observations (e.g., with allergies) and possibly take their child to a physician for these conditions (for diagnosis and/or allergy tests), the identification of autism, ADHD, and allergies appeared to be more sensitive to mothers' knowledge and preferences than for conditions that are fairly recognizable. Age-related trends, however, were not fully consistent with this interpretation. One would expect that the positive association between maternal education and child health would weaken across childhood. As children enter formal schooling, they spend less time with their mothers (Aguiar and Hurst, 2009) and encounter other adults (e.g., teachers, school nurses, coaches) who could be able to detect (and report to mothers) various health-related symptoms. Yet the relationship actually was larger starting for six year olds and remained relatively unchanged (for most conditions) as children entered adolescence. This unexpected finding could be due to younger children's relatively good state of health, which introduces floor- and ceiling-effects for maternal education. Another possibility is that there are differences in which the ways mothers receive this information and/or process it that differ across levels of education or differences in spatial clustering by levels of education (Liu et al., 2010). Unfortunately, the NHIS data did not allow me to examine these (or other) possibilities. Importantly, consistent with prior research, family income and paternal education were each associated with lower probabilities of having most conditions/diagnoses, and, in general, income had a stronger influence on the probability of having a reported health condition than maternal education did. This dissimilarity in the direction and magnitude suggests that the portfolio of healthpromoting behaviors, attitudes, and orientations that education
178
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
Fig. 2. Predicted maternal education gradient in reported health conditions, by child's age.
implies (Mirowsky and Ross, 2003) does not fully extend to other aspects of socioeconomic position. Education itself or the characteristics that drive individuals to attain more schooling are intimately tied to the ways in which mothers interact with their children and consider their well-being. Education may not offer the same health protection as family income resources (given the relative magnitude of coefficients) or may offer increased ability to identify low-severity or subtle symptoms while also mitigating the underlying risk of the conditions themselves. Moreover, results regarding the relative strength of maternal and paternal education are consistent with some prior work (e.g., Cochrane et al., 1982), but stand in contrast to work outside the United States that finds that maternal characteristics do not have consistently stronger associations with children's health outcomes than paternal characteristics (e.g., Corsi et al., 2015). The differences in the relative importance of mothers' and fathers' socioeconomic resources is likely determined by the extent to which mothers or
fathers are responsible for caregiving and health care access in a given location at a given time. Future work may be able to investigate this possibility through considerations of period trends and geographic variation. Other research has examined self-reported measures of “objective” health in adult populations. Baker et al., (2004), for example, compared self-reports of diagnosed conditions in Canadian adults to their health insurance records. The researchers found substantial variation in self-report that was related to the intensity of the illness and socioeconomic resources. My results were consistent with these findings and extended them to an American population of parental reports of health. Taken together, results suggested that maternal educational advantage protects child health, but unobserved characteristics of better-educated mothers mean that they may be more likely to identify and report many types of health conditions than are lesseducated mothers. If this interpretation is indeed correct,
E.R. Berchick / Social Science & Medicine 170 (2016) 170e179
research that uses reported conditions may underestimate socioeconomic disparities in child health. Caution must be taken in considering which aspects of parental socioeconomic advantage should be used to measure socioeconomic inequalities in child health. Policy-decisions and research that draw on reported conditions need to consider that differences in reporting across levels of maternal schooling may lead to an underestimate of the true health differences between advantaged and disadvantaged children. Acknowledgements I would like to thank Sara S. McLanahan, Scott M. Lynch, Marta Tienda, Elizabeth M. Armstrong, Tod Hamilton, Patrick Ishizuka, Zitsi Mirakhur, and Taylor Holubar for their helpful comments on earlier drafts of this paper. Partial support was provided by NIHCD grant P32CHD047879 while the author was at the Office of Population Research, Princeton University. References Aguiar, Mark, Hurst, Erik, 2009. A summary of trends in U.S. Time use: 1965-2005. Soc. Indic. Res. 93 (1), 57e64. Allison, Paul D., 1999. Comparing logit and probit coefficients across groups. Sociol. Methods Res. 28, 186e208. Baker, Michael, Stabile, Mark, Deri, Catherine, 2004. What do self-reported, objective measures of health measure? J. Hum. Resour. 39 (4), 1067e1093. Bureau of Labor Statistics, 2016. CPI Detailed Report. http://www.bls.gov/cpi/ cpid1606.pdf. Case, Anne, Lubotsky, Darren, Paxson, Christina, 2002. Socioeconomic status and health in childhood: the origins of the gradient. Am. Econ. Rev. 92 (5), 1308e1334. Case, Anne, Paxson, Christina, 2001. Mothers and others: who invests in children's health? J. Health Econ. 20, 301e328. Case, Anne, Paxson, Christina, 2002. Parent behavior and child health. Health Aff. 21 (2), 164e178. Chen, Edith, Matthews, Karen A., Thomas Boyce, W., 2002. Socioeconomic differences in children's health: how and why do these relationships change with age? Psychol. Bull. 128 (2), 295e329. Cochrane, Susan H., Leslie, Joanne, O'Hara, Donald J., 1982. Parental Education and child health: intracountry evidence. Health Policy Educ. 2 (3e4), 213e250. Corsi, Daniel J., Subramanian, S.V., Ackerson, Leland K., Smith, George Davey, 2015. Is there a greater maternal than paternal influence on offspring adiposity in India. Arch. Dis. Child. 100, 973e979. Craig, Lyn, Mullan, Killian, 2010. Parenthood, gender and work-family time in the United States, Australia, Italy, France, and Denmark. J. Marriage Fam. 72, 1344e1361. Cunningham, J., Dockery, D.W., Speizer, F.E., 1996. Race, asthma, and persistent wheeze in philadelphia schoolchildren. Am. J. Public Health 86 (10), 1406e1409. Currie, Janet, 2009. Healthy, wealthy, and wise: socioeconomic status, poor health, and human capital development. J. Econ. Lit. 47 (1), 87e122. Currie, Janet, Decker, Sandra, Lin, Wanchuan, 2008. Has public health insurance for older children reduced disparities in access to care and health outcomes? J. Health Econ. 27, 1567e1581. Currie, Janet, Moretti, Enrico, 2005. Biology as Destiny? Short and long-run determinants of intergenerational transmission of birth weight. J. Labor Econ. 25 (2), 231e264. Downey, Douglas B., 1995. When bigger is not better: number of siblings, parental
179
resources, and educational performance. Am. Sociol. Rev. 60, 746e761. Evans, William, Wolfe, Barbara, Adler, Nancy, 2012. The SES and health gradient: a brief review of the literature. In: Wolfe, B., Evans, W.N., Seeman, T.E. (Eds.), The Biological Consequences of Socioeconomic Inequalities. Russell Sage, New York, pp. 1e37. Fuchs, Victor R., 1982. Time preference and health: an exploratory study. In: Fuchs, Victor R. (Ed.), Economic Aspects of Health. University of Chicago Press, pp. 93e120. Gakidou, Emmanuela, Cowling, Krycia, Lozano, Rafael, Murray, Christopher JL., 2010. Increased educational attainment and its effect on child mortality in 175 countries between 1970 and 2009: a systematic analysis. Lancet 376, 959e974. Guryan, Jonathan, Hurst, Erik, Kearney, Melissa, 2008. Parental education and parental time with children. J. Econ. Perspect. 22 (3), 23e46. Hayward, Mark D., Hummer, Robert A., Sasson, Isaac, 2015. Trends and group differences in the association between educational attainment and U.S. Adult mortality: implications for understanding Education's causal influence. Soc. Sci. Med. 127, 8e18. Hurst, Erik, 2010. Comment on ‘Rug Rat Race.’” Brookings Papers on Economic Activity Spring 2010, pp. 177e184. James, John M., 2003. Respiratory manifestations of food allergy. Pediatrics 111 (3), 1625e1630. King, Marissa D., Bearman, Peter S., 2011. Socioeconomic status and the increased prevalence of autism in California. Am. Sociol. Rev. 76 (2), 320e346. Lareau, Annette, 2003. Unequal Childhoods: Class, Race, and Family Life. University of California Press, Berkeley, CA. Leibowitz, Arleen, 1974. Education and home production. Am. Econ. Rev. 64, 243e250. Liu, Ka-Yuet, King, Marissa, Bearman, Peter S., 2010. Social influence and the autism epidemic. Am. J. Sociol. 115 (5), 1387e1434. Lynch, Scott, M., 2006. Explaining life course and cohort variation in the relationship between education and health: the role of income. J. Health Soc. Behav. 47, 324e338. McLanahan, Sara, Percheski, Christine, 2008. Family structure and the reproduction of inequalities. Annu. Rev. Sociol. 34, 257e276. Mehta, Neil K., Lee, Hedwig, Ylitalo, Kelly R., 2013. Child health in the United States: recent trends in racial/ethnic disparities. Soc. Sci. Med. 95, 6e15. Minnesota Population Center and State Health Access Data Assistance Center, 2015. Integrated Health Interview Series: Version 5.0. University of Minnesota, Minneapolis. Mirowsky, John, Ross, Catherine E., 2003. Education, Social Status, and Health. Aldine de Gruyter, New York. Mood, Carina, 2010. Logistic regression: why we cannot do what we think we can do, and what we can do about it. Eur. Sociol. Rev. 26 (1), 67e82. Myrskyl€ a, Mikko, Fenelon, Andrew, 2012. Maternal age and offspring adult health: evidence from the health and retirement study. Demography 49, 1231e1257. Nazarov, Zafar, Rendall, Michael S., 2011. Education and the Effect of Childcare on Obesity. RAND Labor and Population Working Paper. Ross, Catherine E., Wu, Chia-ling, 1995. The links between education and health. Am. Sociol. Rev. 60 (5), 719e745. Sayer, Liana C., Bianchi, Suzanne M., Robinson, John P., 2004. Are parents investing less in Children? Trends in mothers' and fathers' time with children. Am. J. Sociol. 110, 1e43. Suizzo, Marie-Anne, Stapleton, Laura M., 2007. Home-based parental involvement in young children's education: examining the effects of maternal education across U.S. Ethnic gGroups. Educ. Psychol. 27 (4), 533e556. United States Department of Health and Human Services, 2012. Healthy People 2020. http://www.healthypeople.gov. Waggoner, Miranda, 2013. Parsing the peanut panic: the social life of a contested food allergy epidemic. Soc. Sci. Med. 90, 49e55. Weitzman, M., Gortmaker, S., Sobol, A., 1990. Racial, social, and environmental risks for childhood asthma. Am. J. Dis. Child. 144 (11), 1189e1194. Ziol-Guest, Kathleen M., Dunifon, Rachel E., 2014. Complex living arrangements and child health: examining family structure linkages with children's health outcomes. Fam. Relat. 63, 424e437.