The relationship between family income and Child Health : revised evidence from England

(1)

BACHELOR THESIS

The Relationship between Family Income and

Child Health: Revised Evidence from

England

Author:

Evelien DE WILT

Supervisor:

Stephanie CHAN

A thesis submitted in fulfilment of the requirements for the degree of

Bachelor of Science

in

(2)

Statement of Originality

This document is written by Evelien de Wilt who declares to take full responsibility for the contents of this document.

I, Evelien, declare that the text and work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

The positive relationship between child health and family income is a well-documented and important policy-related topic in child health literature. Nevertheless, the mechanisms by which family income affects child health remain poorly understood and national studies are divided on whether the income gradient increases in child age. This paper contributes to the on-going research on the income-child health gradient and provides updated results on how family income directly and indirectly affects child health in England. It will also make comparisons with

previous influential research, in particular to that of Currie, Shields and Price (2007) on the income-child health gradient in England for the period 1997-2002. Using data from the 2005-2010 Health Surveys for England, this study finds a positive, significant but small income gradient in a child’s subjective health status: a gradient that does not increase in child age. Furthermore, estimates suggest that the mother’s education and employment status, parental health and nutrition play important roles in the importance of the child health-income gradient. Nevertheless, if objective measures of child health are used, the income gradient disappears. These results, together with findings of a very small income gradient in subjective child’s health status, propose the similar conclusion to Currie et al. (2007) that family income remains a non-determinant of child health in England.

(4)

Introduction

The relationship between health and income is a well-documented and important policy-related topic in the fields of economics and other social sciences (Currie, Shields, & Price, 2007). In fact, various different studies have observed the phenomenon that relatively wealthier people live longer and healthier, therefore proving a relationship between income and health (Case, Lubotsky, & Paxson, 2002). However, even though higher income is associated with better health, the relationship between the two factors is the steepest at the bottom of the income distribution. This means that the health gain after an income increase is much higher for lower-income individuals than for those with higher lower-incomes. It is for this reason that the lower-income-health relationship is often referred to as the “income gradient” in a person’s health status (Evans, Wolfe, & Adler, 2012).

Nevertheless, this income-health gradient remains difficult to study because the real causal pattern is unclear. Does low income lead to poor health or vice versa? Are there other factors that simultaneously contribute to low income and poor health? What really lies behind the income gradient (Fletcher & Wolfe, 2010)? For example, is poor health of those living in poverty the result of inadequate nutrition, lack of access to proper health care or simply the higher stress levels that are associated with living in severe poverty? Or is it that higher income will lead to better health as those individuals have access to better quality medical care, nutritious food and generally live in a safer environment (Fletcher & Wolfe, 2010)?

As a result of this continuous struggle to find the mechanisms behind the income gradient in a person’s health status, another strand of literature has recently started to focus on the effects of family income on the health of children (Burgess, Propper, & Rigg, 2004). Children are an important group to study because for most children, family income does not depend on their own health. This means that that income gradient in child health is less likely to be effected by reverse causation (Fletcher & Wolfe, 2010). Furthermore, the income-child health relationship is

important to examine as it has been shown that poor health in childhood may result in negative outcomes in the future. For example, health inequalities in childhood may result in lower educational achievement, bad health and poor labour market outcomes in adulthood (Currie, Shields, & Price, 2007) (Siponen, Ahonen, Savolainen, & Hameen-Anttila, 2011). Research into

(6)

development and provides useful insights when creating governmental welfare and financial policies (Currie & Stabile, 2003).

This paper will contribute to the on-going research on the income-child health gradient by investigating how family income affects child health in England and how certain child health related factors – e.g. parental characteristics, child nutrition and family lifestyle – affect this income gradient. The main research question is therefore “Is there evidence of a significant income-child health gradient in England?” If this question proves to be true, this paper will focus on the following sub research question: “If there is evidence of a significant relationship between family income and child health, are there any mechanisms that affect this income gradient and therefore prove an indirect route by which family income affects child health?”

Previous studies on this topic in England by Currie et al. (2007), Burgess et al. (2004) and Propper et al. (2007) find significant yet small income-child health gradients but do conclude that certain child health related factors such as parental behaviour, chronic health conditions and family nutrition influence child health through family income. However, as all these studies use relatively older data from before 2005, this paper will provide an updated research on the English income-child health gradient by using the 2005-2010 data from the Health Surveys of England (HSEs) and compare results to those of previous English studies.

The structure of this paper is as follows: Section 1 will give a literature review of the main studies on the income gradient in child health and its contributing mechanisms. Section 2 covers an overview and description of the HSEs data and the sample characteristics used in this study, followed by an explanation of the child health- and income- measures and the issue of missing values in Section 3. In Section 4, the results of the statistical models will be presented and compared with previous studies. Finally, Sections 5 and 6 will conclude and discuss the findings and limitations of this paper.

(7)

1. Literature Review

1.1 Main studies on the income gradient in child health

As mentioned in the introduction, the relationship between a child’s health and his/her parental socioeconomic status has only been a topic of research in recent years. In an influential

contribution to child health literature, Case et. al (2002) investigate the correlation between family income and child health combining data from four different US sources, including the annual National Health Interview Survey (NHIS) from 1986 to 1995, the 1988 child health supplement to the NHIS (NHIS-CH), the Panel Study of Income Dynamics with its associated 1997 Child Development Supplement (PSID-CDS) and the Third National Health and Nutrition Examination Survey (NHANES) from 1988 to 1994. Their research uses an ordered probit model with different subjective child health proxies (self-rated health status, parental-assessed health status and physician-assessed health status) as dependent variables and the log of permanent family income as the independent variable. Not only do their findings show a significant and positive relationship between household income and children’s health, the income gradient in children’s health also increases with age. Specifically, Case et al. (2002) find that a doubling of family income is associated with a 4% increase in the probability of children in the 0-3 age bracket being in ‘excellent’ or ‘very good’ health, followed by a 4.9% increase for children aged 4-8, 5.9% for children aged 9-12 and 7.2% for 13-15 year olds. Their research therefore

concludes that a family’s long-run average permanent income is a strong determinant of children’s health and that the positive effect of increased income on child health accumulates over children’s lives.

In a similar research to Case et al. (2002), Currie and Stabile (2003) explore the

relationship between family income and child health in Canada using panel data from the 1994, 1996 and 1998 National Longitudinal Survey of Children and Youth (NLSCY), a data set that collects detailed information on the health and demographics of Canadian children. In their OLS regression, Currie and Stabile (2003) use total household income as the independent variable and four different child health proxies as dependent variables, including the subjective reported health

(8)

chronic conditions and the number of days of hospitalization in the past year. Their results confirm the findings of Case et al. (2002) and also show that the relationship between family income and child health becomes more pronounced as children age.

Finally, Condliffe and Link (2008) reach the same conclusion when investigating the US panel data sets of the PSID Child Development Supplements 1 and 2 and the Medical

Expenditure Panel Survey (MEPS) for the period of 1996 to 2002. Using the parental reported health status of the child as the child health proxy, the results of their ordered probit model show that children from relatively lower-income families are less able to respond to particular health shocks1 and that the negative effects of each shock accumulates as children age.

In contrast to the findings of a positive and age-enhancing income gradient in child health by the foregoing literature, Currie et al. (2007) find different results when exploiting the 1997-2002 Health Surveys of England (HSEs). Using an ordered probit model similar to Case et al. (2002) and a subjective general health status measure as the child health proxy, they find that although the income gradient in child health is significant, it is very small and does not increase with children’s age. Furthermore, their research shows no evidence of an income gradient in child health when using more objective child health proxies, including nurse examinations of the child and blood test results. Their results therefore suggest that family income is not a major

determinant of child health in England.

Another research that provides counterevidence to the positive income gradient in child health is from Burgess et al. (2004). Using data from the Avon Longitudinal Study of Parents and Children (ALSPAC) in the United Kingdom from 1991 and 1992 and an OLS regression model, Burgess et al. (2004) find that although children from poorer households have relatively poorer health, the direct impact of family income on child health is small and again does not increase with age.

As a result of these apparent differences between American and English income-child health gradients, originally presented in the results of Case et al. (2002) and Currie et al. (2007), Case, Lee and Paxson (2008) recently re-examined the Health Surveys of England for the period of

1_{In medical terms, a health shock is defined as a situation where an individual’s health status deteriorates and the}

individual gets sick, injured or suffers from a chronic or acute disease. The impact of a health shock is proven to be more severe for individuals from low-income households as they cannot bear the costs for treatment or have no access to quality medical care (Liu, 2014).

(9)

1997 to 2005 and compared the results with the American NHIS data for the same period respectively. Their study indicated that differences between British and American data are

reduced when the same time period is compared. Furthermore, they show that Currie et al. (2007) presented biased conclusions on the English income-child health gradient as a result of incorrect coding of children’s chronic conditions. In fact, Case et al. (2008) find a significant, positive relationship between family income and child health and an age-enhancing income gradient in both the United States and the United Kingdom. Moreover, their results show that chronic health conditions in the United Kingdom have larger adverse effects on children’s health status and that income plays a bigger role in providing support in case of these chronic conditions.

The study of Case et al. (2008) therefore presents contradicting results, as one would expect that equal access to high quality medical care in the United Kingdom would reduce the effects of chronic conditions on children’s health status and make the effect of these conditions less dependent on income. The authors themselves acknowledge this contradiction and explain the possibility that the differences in results from both countries do not explain anything about the different health care systems or social environments. They conclude that precise comparisons on the income-child health gradient between countries might therefore be impossible.

It is thus clear that child health literature presents mixed and contradicting results on the hypothesis that family income is a major determinant of child health. According to Case et al. (2008), these mixed results are not by cause of different health care systems, but primarily as a result of different surveys, data collection protocols, questions and sample sizes. Moreover, the foregoing literature shows that a gradient even exists in countries with publicly funded health services, such as the United Kingdom and Canada. The question therefore arises whether there are possible mechanisms contributing to the income gradient in child health even if health care is partially or fully publicly funded in the investigated countries (Khanam, Nghiem, & Connelly, 2009). The following subsection will cover several different conclusions from researches on these possible mechanisms behind the income-child health gradient.

(10)

1.2 Contributing mechanisms behind the child health-income gradient

As with the contradicting evidence on the increasing and positive income gradient in children’s health, the empirical evidence on the mechanisms via which family income may affect a child’s health is also far from established. Case et al. (2002) analyse whether family income affects child health though one of the following factors: chronic health conditions, parental health, health insurance, genetic ties, health at birth and several parental and child health related behaviours. In order to assess the significance and correlation with family income of these possible explanatory factors, and ultimately with child health, Case et al. (2002) use an OLS regression with family income as the independent variable and the different possible factors as dependent variables. They find that chronic health conditions occur more frequently with children from lower income families and that these families are less able to handle these conditions. This might provide a possible explanation for the positive income gradient in children’s health. Furthermore, their results show that all other factors do not significantly correlate with family income and therefore do not contribute to the relationship between family income and child health, with the exception of chronic conditions.

Currie et al. (2007) find a similar result in England when regressing subjective child health status on chronic health conditions and family income in an ordered probit model.

Moreover, they examine the role of child nutrition and family lifestyle and find that nutrition and lifestyle choices do not decrease the importance of the income-health gradient, therefore

suggesting that those two factors are important determinants of a child’s health status. Propper, Rigg and Burgess (2007) also find evidence that parental and maternal behaviour (e.g. diet, exercise, housing conditions and breast-feeding) influences child health and, more importantly, that the income gradient in child health dissappears completely when controls for parental health are added in the regression analysis.

Finally, Khanam et al. (2009) present similar results for Australian children and document that parental, and in particular maternal health, plays a significant role in the positive relationship between family income and child health. Their results indicate that when parental health is added as a control variable to the regression model, the income coefficient is reduced to almost zero and no evidence is found for an income-child health gradient at all. A possible explanation for this result is that children from less healthy parents are more prone to disease or illness as their

(11)

parents cannot provide them with a clean and healthy home environment to grow up in. Also, children from sick parents may receive less parental care and support and are exposed to high levels of stress during their childhood (Khanam, Nghiem, & Connelly, 2009).

In contrast to the studies mentioned above, Dowd (2007) finds no evidence of certain explanatory variables in the income-child health gradient for the United States. Using data from the 1988 and 1991 US National Maternal and Infant Health Survey (NMIHS), he assesses whether the health status of the mother during pregnancy and early infancy can possibly explain the relationship between family income and child health at the age of 3. However, even though his research shows evidence that certain maternal health risk factors, e.g. maternal smoking during pregnancy or breastfeeding after birth, are significantly related to family income and maternal education, they do not play a significant role in the relationship between family income and child health. Furthermore, his research suggests that there must be other pathways by which family income affects child health and that subjective and objective child health status measures do not relate to family income in the same way.

In conclusion, this section provided an overview of national studies on the family income

gradient in children’s health. Even though several studies show a significant and positive income gradient, no consensus is reached regarding the hypothesis that the gradient increases in

children’s age. Furthermore, the mechanisms by which family income leads to better health remain poorly understood. This paper therefore aims to provide another insight into the family income- child health gradient in England for the years 2005-2010 and attempts to unravel the correlation between income and child health by examining routes by which child health related factors might indirectly affect a child’s health. The following section will give a description of the Health Surveys of England data used in this paper.

(12)

2. Data and Sample Characteristics

2.1 The annual Health Surveys of England (HSEs)

This paper utilizes the pooled data from the 2005-2010 Health Surveys of England (HSEs) in order to investigate the relationship between family income and child health in England. The HSE is a series of annual surveys since 1991, commissioned by the UK’s Department of Health and conducted by the the Joint Health Surveys Unit of the National Centre for Social Research and the Department of Epidemiology and Public Health at University College London (Currie et al., 2007). It is a nationally representative survey of private households in England and is used to (1) monitor trends in the nation’s health, (2) find ways of improving people’s health, (3) identify inequalities in health, (4) estimate the prevalence of certain risk factors and (5) help plan new national health policies and regulations (Office for National Statistics, 2014).

The HSE generally includes adults aged 16 and over, but since 1995 has also surveyed a maximum of two children (aged 2 to 15) from the sampled households. The information on the health of younger children (age 12 or younger) is obtained from a parent, while older children between the ages of 13 and 15 are asked questions directly, with parental consent. From 2001 onwards, the survey also obtains information about toddlers and infants under the age of 2 and includes an additional ‘boost’ sample of children aged 2 to 15 in order to monitor trends in children’s health more accurately (University College London, 2015).

The survey uses the Postcode Address File as a sampling frame and typically has a average response rate of 75%, obtaining a sample of around 16,000 individuals per year. In order to obtain a truly representative picture of everyone living in England, addresses are chosen at random and change each year, ensuring that each yearly HSE investigates different families than the year before. There are several ways by which data is collected, including self-completion questionnaires, face-to-face interviews and medical examinations by registered nurses. If a respondent gives his/her consent, blood and saliva samples are taken and tested for haemoglobin and ferritin counts and active hormone levels (Currie et al., 2007). Other core questions of the HSE include general health and psycho-social indicators, alcohol consumption, smoking, questions on health services and use of medicines, measurements of blood pressure, height and weight and important individual and household demographic characteristics such as family

(13)

income, ethnicity and education (University College London, 2015). Information on family income has only been included in the survey since 1997 onwards.

In addition to general health, demographic and psycho-social indicators, each annual health survey has a particular theme and focuses on a specific disease, condition or population group (Office for National Statistics, 2014). Several examples to date include (1) physical activity and fitness in 2008 and 2012, (2) the health of minority ethnic groups in 1999 and 2004 and (3) general wellbeing in 2010 and 2011. All topics are repeated at suitable intervals in order to monitor specific changes over time (University College London, 2015).

2.2 Sample characteristics

The six years of pooled data used for this research gives a sample of 94415 individuals, out of which 33902 are children under the age of 16. Unfortunately the boost samples of children aged 2 to 15 in each yearly dataset could not be used because the parents of the children are not sampled and thus parental information for each child is missing. This results in 14863 children being omitted from the dataset. In addition to missing parental information, another 5723 children are excluded from this research as their total family income is unknown.

Out of the remaining 13316 children, it was possible to match parental characteristics on up to two parents for 94.9 % of cases. This process of identifying the legal guardian(s) of each child and matching their information was executed in Excel with use of the household serial numbers and each individual’s person number. For each child, the dataset also allows to exactly identify the child’s genetic or non-genetic (foster, adopted, and step-) parent(s) and possibly any other blood relatives with legal parental responsibility (e.g. grandparents or older siblings). Furthermore, it is possible to determine whether the absense of one genetic or non-genetic parent is the result of a permanent absense, a temporary absence at the time of the interview or a

partially incomplete survey of the specific household (not all household members are included in the survey). For the purpose of this research, children that are under legal custody of a

grandparent or older sibling are dropped from the dataset, resulting in a final sample size of 12050 children in 7789 families.

(14)

2.3.1 Subjective measure of child health

The subjective measure of child health used in this research results from the survey’s health question asked to parents about the general health of their child or to children themselves if aged between 13 to 15 years old. Each ordinal response category and the respective percentage of children in each category are given below:

1. Very good (62.5%). 2. Good (32.4%). 3. Fair (4.3%). 4. Bad (0.7%). 5. Very bad (0.1%).

In order to obtain similar results to Currie et al. (2007) and because only 0.8% of children are in bad or very bad health, the ‘bad’ and ‘very bad’ health categories are collapsed into one category, resulting in a four-point ordinal indicator for self-assessed general health.

2.3.2 Objective measures of child health

In addition to self-assessed general health status, this research will investigate the relationship between family income and several objective measures of child health, including birth weight, height, obesity and high blood pressure. These objective child health measures are derived from blood samples and measurements taken by a qualified nurse during the time of the survey. A child is considered ‘obese’ if his/her BMI is higher than 29.92, resulting in a binary variable for obesity with 1 being ‘obese’ and 0 being ‘non-obese’ or ‘healthy weight’. Out of the 10457 children for whom the BMI is known, 149 of them are obese. Furthermore, a child has a high

2_{BMI is calculated by dividing a person’s weight in kilograms by the square of height in meters. However, as}

children’s body composition changes and varies as they age, the BMI for children is calculated in age- and sex-specific percentiles and no sex-specific measure for ‘obesity’ exists. For this reason, and to estimate similar results to those of Currie et al. (2007), this research uses the BMI cut-off for general obesity in adults (29.9) (Centres for Disease Control and Prevention (2015).

(15)

blood pressure if his/her diastolic pressure if higher than 89.93, resulting in a binary variable for High Blood Pressure (HBP) with 1 being ‘high’ and 0 being ‘normal’. Out of the 5240 children for whom diastolic pressure was measured, 101 of them have high blood pressure. Interestingly, no significant correlation is found between obesity and high blood pressure for these sampled children.

2.4 Measure of family income

The main measure of family income reported by the HSE is the current total pre-tax annual family income, presented in 31 different bands ranging from less than 500 pounds to more than 150,000 pounds. In order to stay consistent with Case et al. (2002) and Currie et al. (2007), this pre-tax family income is adjusted to reflect inflationary pressures over time with use of the UK Average Earnings Index (AEI)4. This is done by taking the midpoint of each income band and deflating this calculated income to 2005 prices. As the AEI was replaced by the Average Weekly Earnings Index (AWE) in July 2010, this research will drop data from August-December 2010 because the AEI for these months is unknown. As a result, the average deflated total family income over five years is £33454.9, with a minimum income of only £226.98 and a standard deviation of £27718.89. For statistical purposes, the deflated total family income is converted into natural logarithms.

2.5 Missing values

One of the most common problems that occur in statistical analyses is dealing with missing information on some variables. In this specific research, the main problem arises when parental information of the genetic or non-genetic father or mother of the child is missing. For example, a child that is raised in a single parent family only has parental information on either the mother or the father. Also, some household members may be left out of the dataset as a result of a

3_{According to the National Health Service of England (NHS), high blood pressure occurs when a person (both adult}

and child) has a systolic pressure of over 139.9 and a diastolic pressure of over 89.9.

(16)

temporary absence at the time of the interview or an incomplete survey. This causes a significant issue for regression analysis of the sample dataset as the statistical software Stata drops all children for whom one value of their parental variables is missing, thus only including children (5702 in total) for whom information on both parents is known.

A solution to this problem of missing control variables and/or predictors is presented by Paul D. Allison in his book Missing Values (2001). According to Allison, the “Dummy Variable Adjustment” or “Missing-Indicator” method presents an easy strategy for missing predictors that includes creating a dummy variable D that is equal to 1 if data is missing on a specific variable X, and 0 if otherwise. Furthermore, all missing values of variable X are replaced by a constant arbitrary value c (either 0 or the mean). The regression will then include the dependent variable Y, the newly adjusted variable X, the dummy variable D and possibly any other variables in the model. This technique can also be extended for multiple independent variables with missing values. Even though this method generally results in biased estimates of the coefficients and is unacceptable when data are truly missing, it is considered appropriate in cases where the missing value simply does not exist. For example, if a child is raised by only his/her mother, information on the father in the survey is missing because the father was never actually part of the family.

In this research, two dummy variables are generated for the absence of either the father or the mother and incorporated in the regression analysis. Furthermore, all missing values of each parental continuous variable (e.g. age) are replaced with the mean value whereas all missing values of each parental binary variable (e.g. employment or limiting illness) are replaced with 0. This results in a complete set of parental information for all 12050 children in the dataset. An example of the original data and the final adjusted data according to the Dummy Variable Adjustment method is given below:

Table 1: Original data on parental information

Father’s age Mother employed

48 .

52 1

. 0

. 1

(17)

Table 2: Adjusted data on parental information, including two ‘absence’ dummy variables

Father’s age Mother employed Father absent household Mother absent household

48 0 0 1 52 1 0 0 535 ₀ ₁ ₀ 535 ₁ ₁ ₀ 58 0 0 1

3. Empirical Results

This section will cover a number of econometric models to examine the size and significance of the child health-income gradient in England for the years 2005-2010, replicating analyses by Case et al. (2002) for the United States and Currie et al. (2007) for England over the period 1997-2002. First, the raw relationship between family income and self-assessed child health will be examined, in absence of any control variables. This will give a first description of the income gradient in child health and will give evidence for whether this gradient increases with child age.

Second, child general health status will be regressed on the natural logarithm of family income with two sets of control variables, using ordered probit models. These models will be estimated for the whole sample of children, the whole sample of children excluding children with a limiting illness and for the four different age groups (0-3, 4-8, 9-12 and 13-15)6. Furthermore, in order to investigate any possible mechanisms by which family income affects child health, the ordered probit models will be extended to include additional parental and demographic variables that might be important in determining a child’s health status. This research will also continue on the analysis of Currie et al. (2007) on the influence of nutrition on child health and will give new evidence on this topic with more recent data.

Finally, the income gradient in objective child health measures, including birth weight, height, obesity and high blood pressure will be estimated with use of OLS and binary probit models.

5_{The missing values of the variable ‘Father’s age’ in Table 1 are replaced with the mean value of all existing values.}

The mean is calculated by adding all the existing values together and dividing by the number of values in the set: (48 + 52 + 58) 3 = 52.67 = 53.

(18)

3.1 Ordered Probit Models

As this research requires the use of ordered probit models to investigate the income gradient in child health in England, this section will provide a short overview on the model’s general theory and pitfalls.

An ordered probit model is a type of regression analysis that is used when the variable to be modeled has a natural ordinal interpretation. Examples of ordinal variables include rating systems (poor, fair, good, excellent), employment status (unemployed, part-time, full-time), grades (A, B, C, D, E, F) or in the case of this research: a person’s health status (very bad, bad, fair, good, very good). It should be noted however that the categories for the dependent ordinal variable are rankings, indicating that the numbers used to code these rankings do not make any sense. For example, if a person’s health status is coded 1 to 5, with 1 being in very good health and 5 being in very bad health, the difference between the first and the second outcome (1 and 2) may not be the same as the difference between outcomes 3 and 4. This means a person’s health status could also be coded 1, 5, 7, 19 and 22, as the numbers to code the rankings do not matter.

When put in formulas, the ordered probit model is an index model for a single, latent and unobservable variable y* that is a linear combination of some predictors 𝑥_!!_{, plus a disturbance}

term 𝑢_! that has a standard normal distribution (Williams, 2006):

𝑦_!∗ _{= 𝑥}

!!𝛽 + 𝑢!

The unobservable variable y* can take on any value from 0 to m according to the following formula (with j = 0, …, m):

𝑦_! = 𝑗 if 𝛼_!!! < 𝑦_!∗_{≤ 𝛼} !

The probability that observation i will select any alternative j is (Williams, 2006): 𝑝!" = 𝑝 𝑦! = 𝑗 = 𝑝 𝛼!!! < 𝑦!∗ ≤ 𝛼! = 𝐹 𝛼!− 𝑥!!𝛽 − 𝐹(𝛼!!!− 𝑥!!𝛽)

(19)

This means that an ordered probit model with j alternatives will have one set of coefficients with (𝑦 − 1) intercepts. One of the most important assumptions underlying the ordered probit (and ordered logit) regression model is that the relationship between all pairs of outcome groups is the same, therefore resulting in only one model and one set of coefficients. This means that the coefficients that describe the relationship between one outcome category and another are the same as any other pair of outcome group. This assumption is often referred to as the proportional odds assumption (Williams, 2006).

Furthermore, an ordered probit model with j alternatives will have j sets of marginal effects, whereby the marginal effects of each variable on the different alternatives sum up to zero. For example, if the marginal effect on health status 5 (very bad health) of the natural logarithm of family income is -0.0205, a person is approximately 2% less likely to be in very bad health if family income increases with one log-point. The theoretical function for the marginal effect of an increase in a regressor xr on the probability of selecting alternative j is given below (Williams,

2006):

𝛿𝑝_!" / 𝛿𝑥_!" = {𝐹′(𝛼_!!!− 𝑥_!!_{𝛽) – F’ 𝛼}

!− 𝑥!!𝛽 }𝛽!

In addition to the theory on ordered probit models, two of its main associated problems occur when (1) the error variances may not be homoskedastic and (2) when the proportional odds assumption does not hold (Williams, 2006). When an ordinal model incorrectly assumes

homoskedastic error variances for all cases of y*, the standard errors of the model are wrong and the estimated parameters will be biased. A common solution to this problem is the use of

heteregeneous choice models; models that explicitly specify the determinants of

heteroskedasticity in order to correct for it (Williams, 2008). The model accomplishes this by simultaneously estimating two equations: one for the different outcome determinants and one for the determinants of the residual variance. The equation for the different outcome determinants is given below (Williams, 2010):

𝑦_!∗ ₌ _𝑥

!" !

(20)

In the above, x is a vector of k values for the ith observation. The equation for the determinants of the residual variance is as follows:

𝜎! _{= exp (} _𝑧

!" !

𝛾_!)

In the above, 𝜎!_{is a parameter that allows the variance to be adjusted in any direction, z is a}

vector of j values for the ith observation and 𝛾! shows how the different z vectors affect the

variance. If 𝜎!_{= 1, the error variances of the ordered probit or logit model are homoskedastic and}

the model can be estimated with use of general ologit and oprobit commands. If 𝜎!_{≠ 1, the error}

variances are heteroskedastic and the ordered models have to be estimated with the hetprob command instead (Williams, 2010).

A second problem that might occur with the ordered probit/logit model is a violation of the proportional odds assumption. If this is the case, and the relationship between all pairs of outcome groups is not the same, different models are needed to describe the relationship between each pair and different sets of coefficients will be generated. The proportional odds assumption can be tested with use of the Brant test, a user-written command that tests whether there is any difference between the different sets of coefficients. If this is not the case, the proportional odds assumption is not violated and the ordered logit and probit model can be estimated with simple ologit or oprobit commands (Williams, 2006).

3.2 A raw sketch of the child health-income gradient

The figure below shows the relationship between total deflated family income and self-assessed child health status by age group, in absence of any control variables. It is clear that the health of children from poorer households is lower than the health of children from relatively wealthier families, thus proving an income gradient in child health for both younger and older children in England. Furthermore, the figure provides no direct evidence that the income gradient increases with age, as the raw correlation coefficients for each age group do not increase systematically.

(21)

Figure 1: The raw child health-income gradient in England by age group7

3.3 The income gradient in child health

The estimates of the ordered probit models with self-assessed child health status, the natural logarithm of family income and two sets of control variables are presented in Table 3. The first set of control variables includes the full set of age and year dummies, male, black, Asian, other ethnic minority, the logarithm of household size, father’s age, mother’s age, absence of the father and absence of the mother. The second set of control variables “controls 2” includes all variables of “controls 1” and the additional variables on the employment and educational status of the father and the mother.

When “controls 1” are included, there is a significant and positive relationship between family income and child health, estimated using the whole sample of children8. The income coefficient of -0.222 implies that if income increases, a child is less likely to be in the worse

7_{The results of this figure were generated in Stata, using the data of this research.} 8

Age 0 – 3 Age 4 – 8

Age 9 – 12 Age 13 – 15

Raw correlation age 0 – 3: -0.1639 Raw correlation age 4 – 8: -0.1426 Raw correlation age 9 – 12: -0.1382 Raw correlation age 13 – 15: -0.1600

(22)

categories of health (fair, bad and very bad health status). In terms of probability, if logincome increases with one unit, a child is approximately 2.1% less likely to be in poor health (very bad, bad and fair health relative to good and very good health status). The first ordered probit model also shows an increase in the size of the income gradient for age groups 9-12 and 13-15 in comparison with age groups 0-3 and 4-8. However, as there is a drop in the size of the income gradient from age group 0-3 to age group 4-8, there is no direct evidence that the child health-income gradient increases systematically in child age in England9.

As mentioned by Currie et al. (2007), one potential concern with these estimates is the possible causality between the employment status of the child’s father and/or mother and the child’s health. For example, if a child has a limiting chronic health condition, the parent might not be able to work in order to take care of the child. This would result in a upward bias in the child health/family income gradient. In order to check for this possibility, the second ordered probit model excludes all children with a limiting chronic health condition (1734 children) and reestimates the first model with the same control variables. The results show no apparent change in the income gradient for the whole sample of children.

The third and final ordered probit model includes additional explanatory variables for parental employment and education status. These additional control variables “controls 2” explain approximately 14.3%10_{of the income gradient for the whole sample of children, 11.1%}

for age group 4-8, 20% for age group 9-12 and 25.8% for age group 13-15. Interestingly, and similar to results found by Currie et al. (2007), the magnitude of the income gradient for age group 0-3 remains unchanged. This result might imply that parental employment and education status does not affect the income-child health gradient when children are still infants, but becomes more important as children grow older. Furthermore, the estimates of this model show no evidence of an increasing income gradient in child age, with the gradient dropping

systematically from -0.019 for age group 0-3 to -0.016 for age group 9-12 but rising to -0.023 for age group 13-15.

9_{An additional ‘equality of coefficients’ test does however show significant differences in coefficients per age}

group, therefore suggesting that the income gradient in child health changes for each age group. Results of this test are given in Appendix 4.

10_{The percentage change in the magnitude of the income gradient is defined as the percentage change between the}

marginal effect of the model with “controls 1” and the marginal effect of the model with “controls 2”. For example, the percentage change of 14.3% for the whole sample of children was calculated as follows:

(23)

Table 3: The income gradient in the general self-assessed child health status (ordered probit models) General Child Health Status

Age group All 0-3 4-8 9-12 13-15

Controls 1 Log family income -0.222 (15.01) [-0.021] -0.222 (5.44) [-0.020] -0.208 (7.92) [-0.018] -0.238 (8.91) [-0.020] -0.243 (7.78) [-0.031] Sample 12050 1691 4239 3580 2540

Controls 1 (additional experiment)

Log family income (excluding children with chronic limiting condition) -0.211 (12.88) -0.206 (4.70) -0.209 (7.27) -0.222 (7.44) -0.221 (6.37) Sample 10316 1470 3683 3028 2135 Controls 2 Log family income -0.186 (11.26) [-0.018] -0.220 (4.86) [-0.019] -0.189 (6.46) [-0.016] -0.186 (6.34) [-0.016] 0.179 (5.10) [-0.023] Sample 12050 1691 4239 3580 2540

Notes 1: Coefficients on the natural logarithm of family income are reported. Absolute t-statistics are given in parentheses (all income coefficients are statistically significant at 1% level). The probability of a child being in very bad, bad or poor health relative to being in good or very good health after an one log point increase in family income is given in square brackets. Notes 2: “Controls 1” includes the full set of age and year dummies, male, black, Asian, other ethnic minority, the logarithm of household size, father’s age, mother’s age, absence of the father and absence of the mother. Notes 3: “Controls 2” includes all variables of “controls 1” and the additional variables on the employment and educational status of the father and the mother. Notes 4: additional tables with the coefficients of the ‘control’ variables can be found in Appendix 3. Notes 5: the “proportional odds assumption” is tested for with use of a Brant test. No violation of the assumption was found.

(24)

3.4 Extended models on the income gradient

Table 4 reports the results of the two extended ordered probit models on the child health-income gradient, expanding on the second set of control variables “controls 2” and including additional parental and child explanatory variables11

. First of all, both models show no evidence that parental education status affects child health, with the exception of the mother having a university/college degree as her highest certificate. This might be explained by the fact that higher educated women have generally better jobs, earn a higher income, and are therefore more able to provide their children with a safer environment and quality medical care in comparison with families where the mother did not earn a college degree. This assumption is further

supported by a significant and positive relationship between the mother’s employment status and child health.

Another interesting result is that children of Asian descent have significantly worse health than other English children. According to a recent report by the National Obesity Observatory (2011), this finding might be explained by the increasing prevalence of obesity-related conditions – including cardiovascular diseases and type 2 diabetes – amongst children and adults of Asian descent. Whilst many ethnic minorities generally have healthier eating patterns than the British white population, unhealthy and fatty diets and insufficient physical exercise are common amongst some ethnic minorities, in particular those of South Asian origin. This is a new

conclusion in comparison with the results of Currie et al. (2007), who concluded that both Asian and Black ethnic minority children have significantly worse health than white children. In this research, no significant relationship is found between being black and children’s health. Furthermore, both models show that children from larger households have relatively worse health, which might indicate that children from larger families miss out on parental support, attention and resources.

In addition to the second set of controls previously included, model 3 includes the

additional explanatory variable of a child’s birth weight in order to analyse biases on the income gradient caused by the “foetal origins hypothesis”. This hypothesis states that many chronic and degenerative health conditions in adulthood are triggered by negative conditions of the “in-utero”

11_{In contrast to Table 3, Table 4 will not provide an overview of the income gradient in child health and its}

associated control variables by age group. As section 3.2 provided no direct evidence for an increasing income-child health gradient in child age, there is no need for further proof in Table 4.

(25)

Notes 1: *** indicates statistical significance at 1% level, ** at 5% level and * at 10% level. Notes 2: “Controls 3” includes the same variables General Child Health Status

Controls 3 Controls 4

Log family income -0.184*** -0.177***

Age Included Included

Male 0.004 0.001

Black 0.067 0.073

Asian 0.117** 0.125**

Other ethnic minority 0.194 0.231*

Log household size 0.082* 0.093**

Father’s age -0.004 -0.005**

Mother’s age -0.009*** -0.010***

Father has no qualification 0.078 0.080

Mother has no qualification 0.064 0.058

Father has A-level highest qualification or

equivalent -0.068 -0.070

Mother has A-level highest qualification or

equivalent -0.032 -0.028

Father has degree highest qualification or

equivalent -0.058 -0.052

Mother has degree highest qualification or

equivalent -0.074** -0.070*

Father employed 0.027 0.099*

Mother employed -0.111*** -0.086***

Birth weight (kg) -0.049** -0.048**

Genetic father has a limiting chronic health

condition − 0.205***

Genetic mother has a limiting chronic

health condition − 0.212***

Non-genetic father has a limiting chronic

health condition − -0.115

Non-genetic mother has a limiting chronic

health condition − 0.272

Sample 12050 12050

(26)

environment, in particular when caused by bad nutrition and smoking behaviour of the mother. For example, famous British physician and epidemiologist David J. Parker has shown that individuals who had inadequate nutrition as a foetus “in utero” have higher chances of becoming obese and are more likely to suffer from chronic diseases such as cardiovascular problems and diabetes (Almond and Currie, 2011). As birth weight proxies for healthy in-utero conditions12

, Model 3 therefore controls to what extent birth weight captures variations of a child’s health status and diminishes the income-child health gradient. The results in Table 4 show a significant and positive relationship between child health status and the child’s birth weight, supporting the hypothesis that children born with higher weight are healthier in the future. However, as the income-child health gradient remains virtually unchanged there is no direct evidence that birth weight contributes to the positive relationship between family income and child health in England.

Finally, the second extended ordered probit model includes additional explanatory binary variables on whether the genetic or non-genetic parent of the child has a limiting chronic health condition. These variables are added to analyse the hypothesis that healthier parents are more likely to have healthier children. This assumption is based on the following three aspects: (1) if parents are in bad health, they might be less able to provide their children with proper care and medical resources, (2) certain parental health conditions might be genetic and can therefore be passed on to the child and (3) there might be certain familial characteristics that affect both the health of the child and parents and that are unobservable in the sampled data set. It should also be noted that information on the health of the genetic or non-genetic parent(s) of the child is only included when the parent(s) are part of the interviewed household. This means that if a child lives with his/her mother and stepfather and this specific household is interviewed during the survey, only these two parents are included in the dataset.

For the purposes of this research, binary information on limiting chronic health conditions for both genetic and non-genetic parents are included, with the expectation that a limiting chronic health condition of a non-genetic parent has no or very little effect on the health status of the

12_{A healthy birth weight for a newborn baby is considered any weight between 5.5 and 8.8 pounds. A low birth}

weight baby can be the result of several reasons, including the health of the mother, genetic factors, problems with the placenta or substance abuse by the mother. High birth weight babies are often heavy because their parents are heavy or the mother has suffered diabetes during pregnancy. Birth weight is therefore considered as an appropriate proxy for the “in-utero” conditions of an unborn child (Li, Ley, Tobias et al., 2015).

(27)

child. The results presented in Table 4 support this assumption and show significant estimates for only the genetic parents of the child. The coefficients of 0.205 and 0.212 imply that children whose genetic parents have a limiting chronic health condition are more likely to be in the worse categories of health (fair, bad and very bad health status), with the effect being similar for both the mother and the father.

3.5 Contributing mechanisms behind the income gradient: nutrition

This section will continue on the research by Currie et al. (2007) on the importance of nutrition in determining a child’s health. In contrast to their relatively small sample of 1914 children13, this research has a significantly larger sample as a result of the HSE including core questions on child fruit and vegetable consumption in each yearly dataset since 2004. Table 5 presents the estimates of the ordered probit model on self-assessed child health status and “controls 2”, with the

additional variables for the number of days a child eats fruit each week (0 to 7) and the number of days a child eats vegetables each week (0 to 7). Out of the 9480 children for whom information on their nutrition is known, 2818 children (29.7%) eat fruit on a regular basis (more than three times per week) but only 162 (1.71%) eat fruit every day of the week and only 532 (5.6%) a portion of vegetables. Furthermore, approximately 15 and 6 per cent of children never or rarely eat fruit or vegetables respectively.

The results in Table 5 show that children who eat fruit on a regular basis are significantly in better health. Interestingly, the relationship between eating vegetables and health for children is negative, implying that eating many vegetables leads to worse health. These conclusions are contradicting to those of Currie et al. (2007), who found no significant health benefit for fruit consumption and only a small but significant relationship between the number of days a child eats vegetables and the child’s health. Nevertheless, as concluded by Currie et al. (2007), the inclusion of variables on nutrition does not lead to a considerable change in the magnitude of the income gradient and leaves the gradient significant at a 1% level.

(28)

Table 5: The importance of diet in explaining a child’s health status (ordered probit model) General Child Health Status

Log family income -0.177***

Number of days the child eats fruit each week 0.051***

Number of days the child eats vegetables each week -0.072***

Sample 9480

Notes: “Controls 2” includes the full set of age and year dummies, male, black, Asian, other ethnic minority, the logarithm of household size, father’s age, mother’s age, absence of the father, absence of the mother and the additional variables on the employment and educational status of the father and the mother.

3.6 The relationship between income and objective measures of child health

To complement the analysis on general self-assessed child health and to provide a better

understanding of the child health/family income gradient, this section will cover the relationship between family income and objective measures of child health. As mentioned in Section 2.3.2, the measures for birth weight, height, obesity and high blood pressure are derived from blood samples and measurements taken by a qualified nurse during the time of the survey. As birth weight and height are continuous variables, they will be linearly regressed on the natural

logarithm of family income, whereas the coefficients on obesity and high blood pressure will be estimated with binary probit models (all four models using “controls 2”). As with previous research, the results show that children from richer families have significanly higher birth weight and are significantly taller in comparison with children from poorer households. Furthermore, children are less likely to be obese if they come from a wealthier family background, but more likely to have high blood pressure. Nevertheless, the coefficients on obesity and high blood pressure are so small that they can be considered irrelevant for the purposes of this research. Also, Currie et al. (2007) show that the income gradient in a child’s height disappears completely when control variables for parental height are included in the regression. This might be explained by the fact that tall parents generally have tall children, therefore indicating that income does not play a role in determining a child’s height. In conclusion, the results of Table 6 find evidence that

(29)

family income is important in determining birth weight, but not in any other objective child health measure.

Table 6: The income gradient and objective measures of child health Birth weight

(OLS) Height (OLS) Obese (probit)

High blood pressure (probit) Log household

income 0.043*** 0.465*** -0.003** 0.005*

Sample 12050 10624 10457 5240

Notes: “Controls 2” includes the full set of age and year dummies, male, black, Asian, other ethnic minority, the logarithm of household size, father’s age, mother’s age, absence of the father, absence of the mother and the additional variables on the employment and educational status of the father and the mother.

4. Conclusions

This paper contributes to the growing literature on the income gradient in child health by investigating the relationship between subjective and objective measures of child health and family income in England for the period 2005 to 2010. It also presents econometric evidence on the mechanisms by which income possibly leads to better health, including variables on parental employment, education and health status and family nutrition. Furthermore, in order to make a relevant comparison with previous research on the income-child health gradient in England by Currie et al. (2007) for the period 1997-2002, this paper uses similar specifications and

econometric methods to derive its empirical estimates.

There are four results of this paper’s analyses that should be highlighted. First, using similar econometric methods as Case et al. (2002), Currie and Stabile (2003) and Currie et al. (2007), this research finds a significant, positive yet small income gradient in a child’s subjective health status. In particular, estimates from ordered probit models show that an one logpoint increase in family income is associated with a child being 2.1% less likely to be in poor health relative to good or very good health. Furthermore, the size of the income coefficient remains relatively similar in comparison with results of Currie et al. (2007), therefore suggesting that the

(30)

the past decade. A second important finding is that there is no evidence on an increasing income gradient in child age in England, in contrast to results found by Case et al. (2002) for the United States and Currie and Stabile (2003) for Canada, but similar to those of Currie et al. (2007), who also find no systematic increase in the income gradient in both their raw income-child health gradient and ordered probit models.

Third, estimates from extended ordered probit models suggest that the mother’s education and employment status are possible mechanisms by which lower income leads to worse health in children. More specifically, the inclusion of additional control variables on parental employment and education status explain approximately 14.3% of the income gradient for the whole sample of children. This is in contrast to findings by Currie et al. (2007), who find no evidence that parental employment status plays a role in the importance of the income-child health gradient. Another finding presented in this research is that children whose genetic parents have a limiting chronic health condition are more likely to be in poor health than children who have healthy parents. Furthermore, this paper provides new evidence on the role of nutrition in protecting the health of children, and suggests that regular consumption of fruit leads to better health.

Finally, and most importantly, no evidence is found for a family income gradient in objective measures of child health (height, obesity and high blood pressure), with the exception of a child’s birth weight. These results, together with findings of a very small income gradient in the subjective child’s health status, propose the similar conclusion to Currie et al. (2007) that family income is not a major determinant of child health in England for both periods 1997-2002 and 2005-2010. As suggested by Currie et al. (2007), a potential explanation for this conclusion is that the National Health Service of England succeeds in providing proper healthcare for children in both poorer and wealthier families. Nevertheless, further research would be required to prove this hypothesis.

(31)

5. Discussion

As with all quantitative research, the results of this paper are subject to several data limitations. First of all, the use of subjective child health status could create bias in the estimation of the family income-child health gradient. Respondents to the child health questions in the HSE survey (either the child’s parent or the child him-/herself) may interpret the health response scales in different ways or give misguided answers depending on their own expectations of the child’s health. Previous research on adult health has shown that biases in self-reported health are common under adults, resulting in an underestimation of a person’s true health status and therefore biased research results (Apouey and Geoffard, 2014). The internal validity of the estimates on the child health-income gradient should therefore be considered with caution.

A second limitation of this paper is its representativeness for the entire English population under the age of 16. As mentioned in Section 2.2, more than 50% of children are omitted from the original HSE dataset because information on their parents and/or family income is missing. As a consequence, the results on the income gradient in both subjective and objective child health status could possibly only hold for a specific group of children and not be generalized for the entire population.

Finally, this research on the income gradient in child health is not prospective and might be subject to endogeneity problems as a result of omitted explanatory variables. This study does therefore not establish causality between family income and child health, but only correlations between all included variables. Furthermore, it is possible that there are other variables regarding family, parental or child characteristics that are not included in the econometric models or are simply unobservable, resulting in a bias in the coefficient on family income. These omitted or unobservable variables might explain why family income is not a major determinant of child health in England and why it remains difficult to find the exact mechanisms by which income affect a child’s health. Future research might therefore be interested in investigating different health related behaviours, e.g. hygiene habits of children, specific dietary habits of the family or physical exercise of the child, as possible mechanisms behind the income gradient.

(32)

Bibliography

Adams, P., Hurd, M., McFadden, D., Merrill, A., & Ribeiro, T. (2003). Healthy, wealthy, and wise? Tests for direct causal paths between health and socioeconomic status. Journal of Econometrics , 112, 3-56.

Allison, P. D. (2001). Missing Data. Quantitative Applications in the Social Sciences, Sage University Paper , 136.

Almond, D., & Currie J. (2011). Killing Me Softly: The Fetal Origins Hypothesis. Journal of Economic Prospect , 25 (3), 153-172.

Apouey, B. J., & Geoffard, P. Y. (2014). Child health and access to health care in France: Evidence on the role of family income. Revue d’Epidémiologie et de Santé Publique , 62 (3), 179-190.

Burgess, S., Propper, C., & Rigg, J. (2004). The impact of low-income on child health: evidence from a birth cohort study. CMPO Working Paper Series , 98 (4).

Case, A., Lee, D., & Paxson, C. (2008). The income gradient in children's health: a comment on Currie, Shields and Wheatley Price. Journal of Health Economics , 27 (3), 801-807. Case, A., Lubotsky, D., & Paxson, C. (2002). Economic Status and Health in Childhood: The

Origin of the Gradient. American Economic Review , 92 (5), 1308-34.

Condliffe, S., & Link, C. R. (2008). The relationship between Economic Status and Child Health: Evidence from the United States . The American Economic Review , 98 (4), 1605-1618. Contoyannis, P., Jones, A., & Rice, N. (2004). The dynamics of health in the British Household

Panel Study. Journal of Applied Econometrics , 19, 473-503.

Currie, A., & Stabile, M. (2003). Socioeconomic status and child health: why is the relationship stronger for older children? . American Economic Review , 93, 1813-1823.

Currie, A., Shields, M., & Price, S. (2007). The child health/family income gradient: Evidence from England. Journal of Health Economics , 26, 213-232.

Dowd, J. B. (2007). Early childhood origins of the income/health gradient: the role of maternal health behaviors. Social Science and Medicine , 65 (6), 1202-1213.

Frijters, P., Haisken-DeNew, J., & Shields, M. (2005). Estimating the causal effect of income on health: evidence from post reunification East Germany. Journal of Health Economics , 24, 997-1017.

(33)

Khanam, R., Nghiem, H. S., & Connelly, L. B. (2009). Child Health and the Income Gradient: Evidence from Australia . Journal of Health Economics , 28 (4), 805-817.

Li, Y., Ley, S. H., Tobias, D. K., Chiuve, S. E., VanderWeele, T. J., Rich-Edwards, J. W., … Qi, L. (2015). Birth weight and later life adherence to unhealthy lifestyles in predicting type 2 diabetes: prospective cohort study. BMJ : British Medical Journal, 351, h3672.

National Obesity Observatory (2011). Obesity and ethnicity. National Health Service

Propper, C., Rigg, J., & Burgess, S. (2007). Child health: evidence on the roles of family income and maternal mental health from a UK birth cohort. Health Economics , 16 (11), 1245 1269.

Sigler, J. A. (1983). Minority Rights: A Comparative Analysis. Westport, Connecticut: Greenwood Press.

Siponen, S. M., Ahonen, R. S., Savolainen, P. H., & Hameen-Anttila, K. P. (2011). Children's health and parental socioeconomic factors: a population-based survey in Finland. BMC Public Health , 11 (457).

Williams, R. (2006). Generalized ordered logit/partial proportional odds model for ordinal dependent variables. The Stata Journal, 6 (1), 58-82.

Williams, R. (2010). Estimating heterogeneous choice models with olgm. Department of Sociology, University of Notre Dame.

(34)

Appendix 1: Summary Statistics

Table 7: Summary statistics

Children aged 0-15

Variable Observations Mean Standard _deviation Min. Max.

Basic demographic variables

Age 12050 8.533 4.003 1 15 Male 12050 0.508 0.500 0 1 Black 12050 0.022 0.146 0 1 Asian 12050 0.044 0.205 0 1 Other ethnic minority 12050 0.006 0.077 0 1 Household size 12050 3.992 1.062 2 12 Health-indicating variables General health status 12050 1.433 0.614 1 4 Birth weight 12050 3.300 0.532 0.9 6.5 Height 10624 133.876 24.577 60.3 198.3 BMI 10457 18.490 3.690 - 59.6 Obese 10457 0.014 0.119 0 1 Diastolic pressure 5240 57.943 22.211 - 131 High blood pressure 5240 0.019 0.138 0 1 Number of days the child eats fruit each week

9480 2.057 1.734 0 21

Number of days the child eats vegetables each week

9480 3.112 2.193 0 23.8

Family income variable Adjusted total family income 12050 33454.9 27718.89 226.9 150,000 Parental variables Age mother 12050 37.380 6.120 - 56 Age father 12050 40.718 5.144 - 75 Mother absent household 12050 0.227 0.419 0 1 Father absent household 12050 0.501 0.500 0 1

The relationship between family income and Child Health : revised evidence from England

BACHELOR THESIS