Health Disparities Within and Between Socioeconomic
Groups in The Netherlands: The Role of Social Isolation,
Education, and Income
Abstract
This thesis examines disparities in self-assessed health status and BMI between and within socioeconomic groups. A descriptive analysis of this within-group variation of health outcomes in different income and education groups is provided by assessing the role of social isolation, education, and income. By using data from 4,738 respondents of the Longitudinal Internet Studies for the Social sciences (LISS) panel, the Ordinary Least Squares (OLS) results firstly confirm the socioeconomic gradient in BMI and self-assessed health status. Secondly, there is substantial variation in BMI and self-assessed health within education and income groups. In general, this variation is largest within the lowest socioeconomic group and increases as one moves up the socioeconomic ladder. Furthermore, there is a clear negative relationship between social isolation and self-assessed health status within income and education groups. Moreover, education can explain self-assessed health and BMI within some income quintiles, and vice versa. Policies to reduce social isolation should be focused on health potential within socioeconomic groups and pay particular attention to the most deprived groups.
Keywords: health disparities, BMI, socioeconomic status, education, income, social isolation
Name: Natasha Spijker Student number: S2711184 Supervisor: prof. dr. J.O. Mierau Co-assessor: dr. R.M. Jong-A-Pin
Course code: EBM877A20, EBM205A20
Organization: University of Groningen, Faculty of Economics and Business Date: 05-06-2020
2
1. Introduction
A positive relationship between socioeconomic status and various health outcomes and health behaviors has been well established in the literature (Alessie et al., 2019, Attanasio & Emmerson, 2003; Barbeau et al., 2004; Braveman et al., 2010; Brunello et al., 2015; Conti et al., 2010; Cutler & Lleras-Muney, 2010; Kivimäki et al., 2020; Lleras-Muney, 2005; Marmot et al., 1978; Marmot et al., 2020; van den Berg et al., 2006; van den Berg et al., 2009). However, health disparities do not only exist between, but also within socioeconomic groups (Ferrer & Palmer, 2004). Furthermore, variations in health are largest within low socioeconomic groups (Ferrer & Palmer, 2004). Especially at the lower end of the socioeconomic spectrum, this implies that some individuals have very good health, whereas others within the same socioeconomic group exhibit poor health. The sources of this variation are not extensively studied in the literature. This thesis therefore attempts to increase our understanding of health differences within socioeconomic groups by providing a descriptive analysis.
3 level if differences in health status are reduced within socioeconomic groups (de Boer et al., 2019; de Boer et al., 2020). Thus, reducing health disparities within socioeconomic groups has benefits both at the individual, and at the societal level.
This thesis focuses on self-assessed health status and BMI as health outcomes in order to identify health disparities between and within socioeconomic groups. Education and income groups will serve as socioeconomic status indicators. Furthermore, social isolation, education, and income will be examined as potential sources of health disparities within socioeconomic groups. Education and income are indicators of socioeconomic status themselves, and their positive association with health outcomes has therefore already been proven (Barbeau et al., 2004; Braveman et al., 2010; Brunello et al., 2015; Conti et al., 2010; Cutler & Lleras-Muney, 2010; Lleras-Muney, 2005). However, education and income capture different dimensions of socioeconomic status (Duncan et al., 2002). Hence, this also makes it interesting to investigate how education and income interact, i.e., whether education is responsible for health disparities within income groups and vice versa. Social isolation, on the other hand, is an inherently different potential source, which is most common within lower socioeconomic groups (Heritage et al., 2008; Weyers et al., 2008; Wilkinson & Marmot, 2003). Social isolation is negatively related to health outcomes (Heritage et al., 2008; Holt-Lunstad et al., 2010; Holt-Lunstad et al., 2015; Luo et al., 2012; Weyers et al., 2008; Wilkinson & Marmot, 2003). Therefore, social isolation may be another factor that separates healthy from unhealthy individuals within socioeconomic groups.
Based on the literature, several hypotheses are established to answer the following research question: To what extent can social isolation, education, and income explain health disparities
within socioeconomic groups?
4
2. Literature Review
This section outlines the most relevant literature regarding the socioeconomic gradient in health, disparities in health status within socioeconomic groups, and factors potentially contributing to this variation, respectively.
2.1 Health disparities between socioeconomic groups
The socioeconomic gradient in health has been observed in the literature for over 40 years. The early study of British civil servants by Marmot et al. (1978) already shows a negative association between occupation and mortality from coronary heart disease (CHD). Mortality among men in the lowest occupational class appeared to be 3.6 times higher than that among men in the highest occupational class. Occupational class also appears to have a negative relationship with risk factors of coronary heart disease. For instance, smoking and high blood pressure were more prevalent among men in the lowest occupational class, whereas physical activity was less prevalent. The authors even concluded that occupational class was a better predictor of CHD mortality than the main risk factors of coronary heart disease.
5 socioeconomic circumstances and health outcomes later in life has been discovered in many studies (Alessie et al., 2019; van den Berg et al., 2006; van den Berg et al., 2009).
Even today, in 2020, the socioeconomic gradient in health remains an issue and an ongoing topic in the literature. For instance, Kivimäki et al. (2020) find that socioeconomic status is associated with risk for 32.1% of the studied health conditions. Furthermore, in a recent report by Marmot et al. (2020) on health equity in England it is argued that the socioeconomic gradient in life expectancy and healthy life expectancy continues to exist today. Additionally, the socioeconomic gradient in health has become steeper over time, which implies that socioeconomic health disparities have increased (Marmot et al., 2020).
Despite the extensive welfare state, the socioeconomic gradient in health also persists in the Netherlands. For instance, lower socioeconomic status is positively associated with many measures of morbidity (Mol et al., 2005), but also with the use of multiple healthcare services and comorbidity (Droomers & Westert, 2004). Furthermore, higher rates of mortality are more prevalent in lower socioeconomic groups (Schrijvers et al., 1999; van Lenthe et al., 2004) and this also holds for rates of poor self-rated health and acute myocardial infarction (van Lenthe et al., 2004). The socioeconomic gradient in health has been observed for healthcare costs as well (de Boer et al., 2019; de Boer et al., 2020). Furthermore, a study by Knoops and van den Brakel (2010) shows that there are substantial differences in life expectancy and healthy life expectancy between low and high income groups. Finally, the substantial socioeconomic health disparities in the Netherlands have not reduced during the past decade and have even increased for some indicators (Broeders et al., 2018).
2.2 Health disparities within socioeconomic groups
6 up the socioeconomic ladder. Furthermore, the healthiest individuals in each socioeconomic group are comparable to one another, but the unhealthiest individuals in the highest socioeconomic group are healthier than the unhealthiest individuals in the lowest socioeconomic group (i.e., differences in within-group variation is largely due to differences in the worst health states between income groups). Hence, some of the individuals within low socioeconomic groups appear to be in very good health, while others in the same group exhibit very poor health. Besides Ferrer and Palmer (2004), de Boer et al. (2019) briefly touch upon differences in healthcare costs within socioeconomic groups and conclude that reducing variations in healthcare costs within socioeconomic groups can yield cost savings of 2.4%.
2.3 Potential sources of health disparities within socioeconomic groups
The sources of health disparities within socioeconomic groups are still unclear. One of the potential sources suggested by Ferrer and Palmer (2004) is variation in individual characteristics, such as genetics, social support, and psychological traits, that can lead to differences in health outcomes. These factors can affect the susceptibility to certain diseases or conditions. Ferrer and Palmer (2004) speculate that there is an interaction between socioeconomic status and such individual characteristics, where the effect of socioeconomic status on health is exacerbated when individual characteristics are unfavorable. Wilkinson and Marmot (2003) also recognize that it is both material and psychosocial factors that help explain health disparities. The potential sources of health disparities within socioeconomic groups discussed here, are social isolation and two dimensions of socioeconomic status: education, and income.
2.3.1 Social isolation. Among the psychosocial factors is social isolation. The
7 The presence of such psychosocial factors is most common within lower social classes. For instance, the extent to which one receives social and emotional support is dependent on socioeconomic status (Weyers et al., 2008; Wilkinson & Marmot, 2003). Furthermore, low-income individuals are more likely to feel alone and are less likely to have friends (Heritage et al., 2008). As there is a negative relationship between social isolation and health, social isolation may be a factor that also separates healthy individuals from unhealthy individuals within socioeconomic groups.
2.3.2 Different dimensions of socioeconomic status. As mentioned earlier,
socioeconomic gradients in health are well-observed in the literature. However, less is known about the interaction between two indicators of socioeconomic status. For example, investigating the relationship between education and health within income groups or investigating the relationship between income and health within education groups. There are several different dimensions of socioeconomic status, but this discussion will focus on the dimensions captured by education and income.
8 In conclusion, education may be a factor that separates healthy from unhealthy individuals within income groups, and vice versa. This is because both education and income are positively related to health outcomes, but the relationship between education and income is imperfect.
3. Hypotheses
Based on the findings from the literature discussed in the previous section, several hypotheses can be established. The first two hypotheses relate to the existence of health disparities between and within socioeconomic groups. The last hypotheses relate to the sources of health disparities within socioeconomic groups.
First of all, it is clear from the literature that there exists a socioeconomic gradient in health, i.e., individuals with a higher socioeconomic status experience better health than individuals with a lower socioeconomic status. Therefore, the first hypothesis reads as follows.
Hypothesis 1: Individuals with a high socioeconomic status (i.e., high income or high education level) are healthier (i.e., have a lower BMI and a higher self-assessed health status) than individuals with a low socioeconomic status in the Netherlands.
Second, some evidence of health disparities within socioeconomic groups is found in the literature. Hence, the second hypothesis is stated as follows.
Hypothesis 2: There are substantial variations in health indicators (i.e., BMI and self-assessed health status) within socioeconomic groups (i.e., income groups or education groups) in the Netherlands, and this variation is higher is low socioeconomic groups than in high socioeconomic groups.
9
Hypothesis 3a: Within socioeconomic groups (i.e., income groups or education groups), individuals who are not socially isolated are healthier (i.e., have a lower BMI and a higher self-assessed health status) than individuals who are more socially isolated.
Hypothesis 3b: Within income groups, individuals with a high education level are healthier (i.e., have a lower BMI and a higher self-assessed health status) than individuals with a low education level.
Hypothesis 3c: Within education groups, individuals with a high income are healthier (i.e., have a lower BMI and a higher self-assessed health status) than individuals with a low income.
The aforementioned hypotheses will contribute to answering the research question: To what
extent can social isolation, education, and income explain health disparities within socioeconomic groups?
4. Data
This section describes the data used to analyze the research question. Section 4.1 will describe the source of the data. The different variables will be discussed in Sections 4.2 and 4.3. Finally, Section 4.4 specifies the final sample.
4.1 Source
Data will be used from the LISS (Longitudinal Internet Studies for the Social sciences) panel, administered by CentERdata (Tilburg University, The Netherlands). The panel consists of approximately 4,500 households, which amounts to approximately 7,000 different individuals participating in the panel. These panel members comprise a representative sample of Dutch individuals and participate in monthly internet surveys. Furthermore, participants also complete yearly surveys, so important changes in their lives can be detected. Persons aged 16 and over and who are capable of participating in the surveys are deemed eligible.
10 datasets will be utilized: Background Variables, Health, Social Integration and Leisure, and
Family and Household.
4.2 Health outcomes
The main dependent variables are self-assessed health status (SAH), which is frequently used in the literature, and Body Mass Index (BMI). The former aims to capture overall health status, whereas the latter captures health behavior. To construct self-assessed health status, the subjects were asked the following question: How would you describe your health, generally speaking? The five answer options range from poor to excellent. This variable is transformed into a binary variable that takes a value of 1 if the respondent reported to be in good, very good, or excellent health, and 0 if the respondent reported to be in moderate or poor health. BMI is not included in the original dataset, but is easily constructed by dividing weight (kg) by the square of height (m).
Some outliers were detected regarding weight and height. More specifically, 16 individuals reported a height of less than 1.00 meter and one person reported a height of 2.69 meters. Similarly, three persons reported having a weight of less than 10 kilograms and two persons reported a weight of more than 700 kilograms. These individuals are excluded from the analysis.
4.3 Independent variables
This section discusses the independent variables, consisting of key variables of interest and control variables.
4.3.1 Key variables of interest. To test the aforementioned hypotheses, two indicators
of socioeconomic status are employed throughout the analysis. The first indicator is the highest level of education with diploma. Originally, the six answer categories for regular education range from primary school to university. Additional categories are other, not (yet) completed
any education, and not yet started any education. The first six answer categories are combined
into three categories: low, middle, and high. Low refers to either primary school or intermediate secondary education, middle refers to higher secondary education, preparatory university education, or intermediate vocational education, and high refers to either higher vocational education or university. Respondents who initially answered other, not (yet) completed any
11
other is largely uninformative, as it is difficult to assign individuals who completed some other
type of education to one of the three new categories. Furthermore, individuals who reported not to have completed any education or not to have started any education are excluded, as a minimum completed education level is required in the Netherlands. As all respondents are aged 16 or over, they should have at least completed primary school. Therefore, it is unlikely that the individuals reporting one of these two categories really have not completed any type of education. A total of 224 individuals reported one of these three categories.
The second indicator of socioeconomic status is quintiles of equivalized gross monthly household income. Individuals who reported a gross household income below the minimum level of social security benefits are excluded from the analysis. This amounts to €982.79 for singles and single parents and to €1403.98 for married couples or cohabitants (Rijksoverheid, 2016). A total of 215 individuals are excluded based on these minimum values. Furthermore, observations for which one person in the household has an income above €20,000 are identified as outliers. 10 individuals reported such high incomes and are therefore removed from the dataset. Moreover, four individuals with incomes between 10,000 and 20,000 are removed from the dataset due to incorrect responses for income. In these cases, an income above 10,000 was reported while having completed a low education level, which is improbable. Next, equivalized income is calculated to adjust for differences in household size and composition. This is done using the modified OECD equivalence scale (OECD, n.d.).1 Using these standardized incomes, income quintiles are created, where the first quintile consists of the 20% of individuals with the lowest incomes and the fifth quintile consists of the 20% of individuals with the highest incomes.
As argued by Duncan et al. (2002), education is one of the mostly used indicators of socioeconomic status, while economic indicators, such as household income, are less frequently used. Duncan et al. (2002) mention several reasons why education level is a good indicator of socioeconomic status. First, an individual’s working life and economic circumstances are to a large extent determined by the level of education completed. Second, information about education is relatively trustworthy. Third, reverse causality is not an issue in later life. Still, this
1 Each household member receives a weight as follows: 1.0 for the first adult, 0.5 for the second adult and each
12 is not to say that education is an exogenous variable. Endogeneity may still be an issue, as unobserved heterogeneity cannot be accounted for. Besides being a good indicator of socioeconomic status, education is also an indicator of early-life conditions. However, even though one’s education level partially determines one’s economic circumstances, such as income, it is not itself an indicator of standard of living. Therefore, by using both education and income as indicators of socioeconomic status, different dimensions of socioeconomic status are captured, and their effects can be compared.
The final key independent variable is a measure of social isolation. This measure is derived from the following statement, for which the respondents had to indicate whether it applied to them: “I miss having people around me.” Possible answers are yes, more or less, and no. The first two answer categories have been combined to form a group of individuals who are, at least to some extent, socially isolated. This variable has thus been transformed into a binary variable that takes a value of 1 if the respondent answered yes or more or less and 0 if the respondent answered no.
4.3.2 Control variables. Two variables will be included as controls. First, the age of
the respondent will serve as a control variable. This is necessary, as individuals will typically experience more health problems at older ages (Grossman, 1972). However, it is important to note that it is impossible to differentiate the effect of a respondent’s age from possible period or cohort effects. The interpretation of an age effect is different from that of a period or cohort effect (Mason et al., 1973; Blanchard et al., 1977). Therefore, as the effect of the age variable might also contain period or cohort effects, the coefficient of the age variable cannot actually be interpreted as the effect of age. Nevertheless, this variable solely serves as a control variable and will therefore not be interpreted. To account for the nonlinearity in this relationship, the square of age is added. Finally, sex (1 = male, 0 = female) will be controlled for to take health differences between men and women into account.
4.4 Sample
13 reduced by 1,562 observations. Hence, the final sample consists of 4,738 observations. Figure A1 in Appendix A shows the process of arriving at the final sample in detail.
Table 1 shows descriptive statistics2 for the variables. The second column shows the data for all observations, while the other columns show the data for the different income quintiles. Table 2 shows the same information per education level. Some preliminary conclusions can be drawn from the tables. First, regarding the means, the gradient is clearly visible when comparing self-assessed health status between the different income quintiles and education levels. The share of individuals who reported to be in good, very good, or excellent health increases as income or education level increases. Second, the gradient is also visible for BMI: it decreases as income or education level increases, even though it slightly increases between the first and second income quintile. The share of overweight individuals and the share of obese individuals follows a similar pattern. Third, the share of socially isolated individuals generally decreases as income or education level increases, as expected. However, when moving from the second to the third income quintile, the share of socially isolated individuals increases somewhat.
2 The coefficient of variation is presented in this table, but will be discussed in detail in Section 6.2.
Table 1. Descriptive statistics per income quintile (equivalized)
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Variable All obs. €1710) (€354- (€1710-€2300) (€2300-€2933) (€2933-€3850) €12667)
(€3850-Mean Health 0.825 0.741 0.784 0.828 0.872 0.899 BMI 25.82 26.35 26.39 25.96 25.48 24.92 Social isolation 0.253 0.310 0.252 0.262 0.228 0.211 Age 52.77 55.36 55.91 53.56 51.04 47.94 Male 0.467 0.421 0.445 0.476 0.492 0.504
Share within educational group
Low 0.254 0.420 0.335 0.212 0.171 0.131
Middle 0.355 0.395 0.397 0.371 0.312 0.303
High 0.391 0.186 0.268 0.417 0.517 0.566
Concentration Index education 0.343 0.366 0.342 0.356 0.394 0.429 Share overweight (BMI > 25) 0.510 0.558 0.558 0.513 0.493 0.427 Share obese (BMI > 30) 0.156 0.188 0.191 0.151 0.138 0.110 Coefficient of variation Health 0.461 0.592 0.526 0.456 0.383 0.336 (0.380) (0.439) (0.412) (0.378) (0.334) (0.302) BMI 0.182 0.185 0.194 0.187 0.165 0.169 (4.701) (4.882) (5.128) (4.850) (4.210) (4.203) Observations 4,738 948 948 948 947 947
14 Next, education and income follow the expected pattern. The share of low educated individuals decreases as income increases, whereas the share of high educated individuals increases. Similarly, the share of individuals within lower income quintiles decreases as education level increases, whereas the share of individuals within higher income quintiles increases. The Concentration Index for education is calculated by summing the squares of the shares within each educational group and shows the extent to which education is concentrated within income quintiles. The Concentration Index for income is calculated in a similar way. The index for education is highest in the last income quintile, where education is concentrated around the highest level. The index for education is lowest in the second income quintile, where individuals are most equally distributed amongst education groups. In the first income quintile education is concentrated around the lowest two levels. Relatively few individuals have completed a high education level within this income quintile, compared to the two lowest education levels, which results in a higher index for this quintile. Similar conclusions can be drawn from Table 2. Within the lowest education group income is most concentrated around the first two quintiles, whereas in the highest education group income is most concentrated around the last two quintiles. The
Table 2. Descriptive statistics per education level
Variable All obs. Low Middle High
Mean Health 0.825 0.761 0.825 0.866 BMI 25.82 26.57 25.98 25.19 Socially isolated 0.253 0.283 0.260 0.227 Age 52.77 58.96 49.97 51.29 Male 0.467 0.407 0.473 0.502
Share within income quintile
Quintile 1 0.200 0.331 0.222 0.095
Quintile 2 0.200 0.264 0.223 0.137
Quintile 3 0.200 0.167 0.209 0.213
Quintile 4 0.200 0.135 0.175 0.265
Quintile 5 0.200 0.103 0.170 0.290
Concentration Index income 0.200 0.236 0.203 0.227 Share overweight (BMI > 25) 0.510 0.569 0.523 0.459 Share obese (BMI > 30) 0.156 0.199 0.171 0.113 Coefficient of variation Health 0.461 0.561 0.461 0.393 (0.380) (0.427) (0.380) (0.341) BMI 0.182 0.196 0.185 0.165 (4.701) (5.195) (4.797) (4.164) Observations 4,738 1,203 1,684 1,851
15 index is lowest for the middle education group, implying that within this group individuals are most equally distributed amongst the different quintiles. In conclusion, education level and income move in the same direction, as expected.
Finally, as a lot of observations are missing, it is important to compare the characteristics of these individuals to the characteristics of the individuals included in the final sample. Table A1 in Appendix A is similar to Table 1, but the different columns show the data throughout the formation of the final sample. As of the second column, i.e. after eliminating outliers for height and weight, the descriptive statistics generally stay relatively constant until the final sample is formed. However, t-tests reveal that regarding some characteristics, the individuals included in the final sample are significantly different from the individuals who are excluded based on missing data. For instance, BMI is significantly higher in the final sample (t(5,491) = 2.02, p = .04; M = 25.82, SD = 4.70 compared to M = 25.45, SD = 4.78). The age of the individuals included in the final sample is significantly higher as well (t(5,497) = 10.82, p < .001; M = 52.77, SD = 17.65 compared to M = 45.34, SD = 17.08). This also holds for the share of men in the final sample (t(5,497) = 2.35, p = .02; M = 0.47, SD = 0.50 compared to M = 0.42, SD = 0.49). No significant differences are found for self-assessed health status, social isolation, education level, and gross household income. The implications of differences in characteristics between the final sample and the missing observations are discussed in Section 7.
5. Methodology
A descriptive analysis will be performed to analyze the research question. Furthermore, linear regression models will be estimated to test Hypotheses 1, 3a, 3b, and 3c. To test Hypothesis 1 and 3a, the model that will be estimated is presented in Equation 1.
ℎ𝑒𝑎𝑙𝑡ℎ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒! = 𝛼 + 𝑺𝑬𝑺!"𝜷 + 𝒄𝒉𝒂𝒓 !
"𝜸 + 𝑠𝑜𝑐𝑖𝑎𝑙 𝑖𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛
!𝛿 +
(𝑠𝑜𝑐𝑖𝑎𝑙 𝑖𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛! × 𝑺𝑬𝑺!")𝜻 + 𝜖!.
In this equation, ℎ𝑒𝑎𝑙𝑡ℎ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 denotes either self-assessed health status or BMI. Furthermore, 𝒄𝒉𝒂𝒓 is a vector of control variables (i.e., sex, age, and age²), and 𝑺𝑬𝑺 is a vector of either the different education levels or the income quintiles. The middle categories (i.e., a
16 middle education level and the third income quintile) will serve as reference categories. As usual, 𝛼 denotes the constant, 𝜖 is the error term, and 𝑖 denotes the individual.
Equation 1 will be estimated separately for both indicators of socioeconomic status, and for both health outcomes. The model will gradually be expanded. In the first specification only a measure of socioeconomic status is included as independent variable. In the second specification, control variables are added. Then, social isolation will be included. Finally, the interaction between social isolation and socioeconomic status will be added.
In order to test Hypotheses 3b and 3c, a model will be estimated in which both indicators of socioeconomic status are included, as well as the interaction between both indicators, and interaction terms between social isolation and each indicator of socioeconomic status. This model is presented in Equation 2.
ℎ𝑒𝑎𝑙𝑡ℎ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒! = 𝛼 + 𝒆𝒅𝒖!"𝜼 + 𝒊𝒏𝒄𝒒!"𝜽 + 𝒄𝒉𝒂𝒓!"𝜸 + 𝑠𝑜𝑐𝑖𝑎𝑙 𝑖𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛!𝛿
+ (𝑙𝑜𝑤 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛!× 𝒊𝒏𝒄𝒒!")𝝑 + (ℎ𝑖𝑔ℎ 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛
! × 𝒊𝒏𝒄𝒒!")𝜾
+ (𝑠𝑜𝑐𝑖𝑎𝑙 𝑖𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛!× 𝒆𝒅𝒖!")𝜿 + (𝑠𝑜𝑐𝑖𝑎𝑙 𝑖𝑠𝑜𝑙𝑎𝑡𝑖𝑜𝑛! × 𝒊𝒏𝒄𝒒!")𝝀 + 𝜖!
In this equation, 𝒆𝒅𝒖 is a vector of education levels and 𝒊𝒏𝒄𝒒 is a vector of income quintiles. Equation 2 will also be estimated for both health outcomes and will be gradually expanded. In this case, the first specification contains both indicators of socioeconomic status, control variables, and social isolation. The second specification will add the interaction between both socioeconomic status indicators. The third specification replaces the interaction between the socioeconomic status indicators by an interaction term between social isolation and the education levels. This is repeated in the fourth specification for the income quintiles. The fifth specification is the full model, which includes all interaction terms.
Equations 1 and 2 will be estimated by Ordinary Least Squares (OLS) when the health outcome is BMI. In the case of self-assessed health status, a Linear Probability Model (LPM) will be estimated. Compared to a Logit model or a Probit model, the estimated coefficients of a LPM are easier to interpret.3 Robust standard errors will be used to account for the presence of
3 Generally, the average marginal effects are comparable between the three different models (Verbeek, 2017).
17 heteroskedasticity.4 Furthermore, due to the potential presence of endogeneity, the purpose of the analysis is not to find any causal relations. Endogeneity will be discussed in Section 7. Finally, in order to test Hypothesis 2 the coefficients of variation of BMI and self-assessed health status will be analyzed and compared for each income quintile and for each education level. The coefficient of variation is calculated by dividing the sample standard deviation by the sample mean, and therefore measures relative variability. As an alternative assessment of health disparities within socioeconomic groups, the methods of Ferrer and Palmer (2004) will be followed. These methods and the corresponding results are discussed in Appendix B.
6. Results
Section 6.1 examines the relationship between socioeconomic status (i.e., education level or income quintiles) and health outcomes (i.e., BMI and self-assessed health status), while Section 6.2 examines health disparities within socioeconomic groups. Finally, Section 6.3 investigates whether social isolation, education, and income can explain health disparities within socioeconomic groups.5
6.1 Health disparities between socioeconomic groups
The coefficient estimates of the OLS regressions for BMI are presented below in Table 3.6 First of all, when education level is the indicator of socioeconomic status, it can be observed that the coefficient of low education is statistically significant in the first three specifications. The level of significance decreases as more variables are added to the model and the size of the coefficient has also reduced substantially. This implies that the observed correlation between low education and BMI can partly be explained by the variables that are added to the model. The coefficient of high education is statistically significant at the 1% level. Furthermore, the coefficients of the two education levels imply that there is a negative relationship between education level and BMI. More specifically, if an individual has a low education level, this person’s BMI is
4 Both the Breusch-Pagan test and the White test indicate that heteroskedasticity is present when BMI is the health
outcome. For self-assessed health status, no tests were performed, as heteroskedasticity is always present in an LPM (Verbeek, 2017).
5 All coefficients will be interpreted under the ceteris paribus assumption.
6 In order to analyze Hypothesis 1, the results presented in the third and seventh specification will be discussed, as
18 approximately 0.364 higher than if this person would have a middle education level. Similarly, having a high education level is associated with a BMI that is approximately 0.907 lower compared to having a middle education level. However, when the income quintiles serve as the indicator of socioeconomic status, the pattern is less clear. When more variables are added to the model, the significance of the first four income quintiles disappears. In the seventh specification the coefficient of the fifth income quintile is statistically significant at the 1% level. If an individual has an income in the range of the fifth income quintile, this person’s BMI is approximately 0.723 lower than if this person would have an income in the range of the third income quintile.
19
Table 3. OLS coefficient estimates (dependent variable: BMI)
(1) (2) (3) (4) (5) (6) (7) (8) Education level Low 0.591*** 0.374** 0.364* 0.458** (0.190) (0.187) (0.187) (0.213) High -0.787*** -0.918*** -0.907*** -0.872*** (0.152) (0.148) (0.147) (0.165) Income quintile First 0.393* 0.375* 0.358 0.533** (0.224) (0.220) (0.220) (0.244) Second 0.431* 0.355 0.358 0.720*** (0.229) (0.225) (0.225) (0.250) Fourth -0.476** -0.326 -0.313 -0.097 (0.209) (0.203) (0.202) (0.220) Fifth -1.037*** -0.744*** -0.723*** -0.456** (0.208) (0.201) (0.199) (0.219) Personal characteristics Male 0.499*** 0.510*** 0.513*** 0.477*** 0.488*** 0.480*** (0.130) (0.130) (0.130) (0.130) (0.130) (0.130) Age 0.277*** 0.279*** 0.279*** 0.257*** 0.260*** 0.259*** (0.022) (0.022) (0.022) (0.0216) (0.022) (0.022) Age² -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Social isolation Socially isolated 0.364** 0.507* 0.358** 1.145*** (0.159) (0.274) (0.159) (0.400)
Educ. level x social isolation
Low x socially isolated -0.343
(0.428)
High x socially isolated -0.135
(0.360)
Inc. quintile x social isolation
First x socially isolated -0.685
(0.529)
Second x socially isolated -1.401***
(0.543)
Fourth x socially isolated -0.837
(0.510)
Fifth x socially isolated -1.093**
(0.506)
Constant 25.98*** 18.15*** 17.99*** 17.95*** 25.96*** 18.41*** 18.24*** 18.08*** (0.117) (0.504) (0.506) (0.506) (0.158) (0.519) (0.519) (0.517)
Observations 4,738 4,738 4,738 4,738 4,738 4,738 4,738 4,738
R-squared 0.014 0.078 0.079 0.079 0.014 0.073 0.074 0.076 Notes: Robust standard errors are in parentheses (*** p < .01, ** p < .05, * p < .10). Specification (1) only includes the education levels. Specification
20
Table 4. LPM coefficient estimates (dependent variable: SAH)
(1) (2) (3) (4) (5) (6) (7) (8) Education level Low -0.064*** -0.041*** -0.038** -0.045*** (0.015) (0.016) (0.016) (0.017) High 0.041*** 0.045*** 0.041*** 0.039*** (0.012) (0.012) (0.012) (0.013) Income quintile First -0.088*** -0.082*** -0.077*** -0.066*** (0.019) (0.019) (0.019) (0.020) Second -0.044** -0.038** -0.039** -0.058*** (0.018) (0.018) (0.018) (0.020) Fourth 0.044*** 0.037** 0.033** 0.021 (0.016) (0.016) (0.016) (0.017) Fifth 0.071*** 0.056*** 0.049*** 0.042*** (0.016) (0.016) (0.016) (0.016) Personal characteristics Male 0.018* 0.015 0.014 0.017 0.013 0.014 (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) Age -0.005*** -0.006*** -0.006*** -0.004** -0.005*** -0.004*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) Age² 0.000 0.000* 0.000* 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Social isolation Socially isolated -0.122*** -0.132*** -0.117*** -0.140*** (0.014) (0.023) (0.014) (0.031)
Educ. level x social isolation
Low x socially isolated 0.024
(0.037)
High x socially isolated 0.010
(0.032)
Inc. quintile x social isolation
First x socially isolated -0.032
(0.045)
Second x socially isolated 0.075*
(0.045)
Fourth x socially isolated 0.051
(0.042)
Fifth x socially isolated 0.029
(0.042)
Constant 0.825*** 1.005*** 1.058*** 1.061*** 0.828*** 0.983*** 1.038*** 1.040*** (0.009) (0.038) (0.038) (0.038) (0.012) (0.039) (0.039) (0.039)
Observations 4,738 4,738 4,738 4,738 4,738 4,738 4,738 4,738
R-squared 0.012 0.027 0.046 0.046 0.023 0.036 0.054 0.056 Notes: Robust standard errors are in parentheses (*** p < .01, ** p < .05, * p < .10). Specification (1) only includes the education levels. Specification
21
6.2 Health disparities within socioeconomic groups
To assess health disparities within socioeconomic groups, Table 1 and Table 2 are consulted. When comparing the coefficient of variation of self-assessed health status between the different income quintiles in Table 1, it can be observed that it is highest in the first income quintile and lowest in the fifth income quintile. The first income quintile, therefore, shows the most variation in self-assessed health and this variation decreases when moving to higher income quintiles. A similar conclusion can be drawn from Table 2, where the most variation is observed within the low education group and the least variation in the high education group. Therefore, there is a clear negative relationship between socioeconomic status and variability in self-assessed health status. However, for BMI this pattern is only visible in Table 2. In this case, the coefficient of variation is highest in the low education group and lowest in the high education group. From Table 1 it can be concluded that variation in BMI fluctuates between income quintile. For instance, the most variation is found within the second income quintile and the least variation within the fourth income quintile. Therefore, there is a negative relationship between education level and variability in BMI, but not between the income quintiles and variability in BMI. As mentioned in Section 5, Appendix B provides an alternative assessment of health disparities within socioeconomic groups by following the methods of Ferrer and Palmer (2004).
6.3 Potential sources of health disparities within socioeconomic groups
6.3.1 Social isolation. To find out whether social isolation is responsible for healthdisparities within7 socioeconomic groups, it is required to assess the average marginal effects (i.e., the combined effects based on the baseline coefficient and the interaction terms) of social isolation for each education level and for each income quintile. These average marginal effects are presented in Table 5 and are based on the fourth and eighth specification from Table 3 and Table 4.
7 The purpose of this thesis is to examine whether social isolation is responsible for health disparities within
22
Table 5. Average marginal effect of social isolation conditional on education or income
(BMI) (SAH) (BMI) (SAH)
Education level Income quintile
Low 0.164 -0.108*** First 0.461 -0.172*** (0.330) (0.029) (0.348) (0.032) Middle 0.507* -0.132*** Second -0.256 -0.064** (0.274) (0.023) (0.367) (0.032) High 0.372 -0.122*** Third 1.145*** -0.140*** (0.232) (0.022) (0.400) (0.031) Fourth 0.308 -0.089*** (0.319) (0.029) Fifth 0.052 -0.110*** (0.309) (0.029)
When BMI is the dependent variable, the average marginal effect of social isolation in the middle education group is statistically significant at the 10% level. Conditional on having a middle education level, if an individual is socially isolated this person’s BMI is approximately 0.507 higher than when this person would not be socially isolated. In the third income quintile the average marginal effect of social isolation is statistically significant at the 1% level. Conditional on having an income in the range of the third income quintile, an individual’s BMI is approximately 1.145 higher when this person is socially isolated than when this person would not be socially isolated. However, the average marginal effects of social isolation in the other education groups and income groups are insignificant. Therefore, apart from the middle education level and the third income quintile, there is no relationship between social isolation and BMI within socioeconomic groups.
Regarding self-assessed health status, the results are more convincing. The average marginal effect of social isolation is statistically significant at the 1% level within all education groups. Conditional on having a low education level, the probability of being in good health is approximately 0.108 lower when an individual is socially isolated than when this person would not be socially isolated. The reduction in this probability is approximately 0.132 and 0.122 within the middle and high education group, respectively. Hence, within each education group there exists a negative relationship between social isolation and the probability of being in good health. A similar conclusion can be drawn when considering the income quintiles. The average marginal effect of social isolation is statistically significant at the 1% level for all income quintiles, except the second, for which the average marginal effect is significant at the 5% level.
Notes: Robust standard errors are in parentheses (*** p < .01, ** p < .05, * p < .10). The (BMI)
23 Conditional on having an income in the range of the first income quintile, the probability of being in good health is approximately 0.172 lower when an individual is socially isolated than when this person would not be socially isolated. The reduction in this probability is approximately 0.064, 0.140, 0.089, and 0.110 within the second, third, fourth, and fifth income quintile, respectively. Therefore, within each income quintile a negative relationship between social isolation and the probability of being in good health is present.
Finally, Figures A2 up to and including A5 in Appendix A show the relationship between socioeconomic status (i.e., education levels or income quintiles) and health outcomes (i.e., BMI or self-assessed health), conditional on social isolation. From Figure A2 it can be observed that, for instance, individuals with a low education level who are not socially isolated have a higher BMI than individuals with a high education level who are socially isolated (F(1, 4,729) = 17.50,
p < .001). The same holds when comparing non-socially isolated individuals with a middle
education level to socially isolated individuals with a high education level (F(1, 4,729) = 7.71,
p = .01). In Figure A3 the pattern is less clear. Regarding self-assessed health status in Figures
A4 and A5, the opposite is true. When comparing non-socially isolated individuals within lower socioeconomic groups to socially isolated individuals within higher socioeconomic groups, the former group generally has better predicted health than the latter.
6.3.2 Different dimensions of socioeconomic status.
24
Table 6. Average marginal effects of education conditional on income
(BMI) (BMI) (SAH) (SAH)
low edu high edu low edu high edu
Income quintile First 0.268 -0.742* 0.003 0.028 (0.350) (0.421) (0.032) (0.038) Second 0.551 -0.554 -0.064* 0.039 (0.394) (0.385) (0.033) (0.031) Third 0.449 -1.019*** -0.111*** -0.015 (0.461) (0.339) (0.036) (0.025) Fourth -0.330 -0.778** 0.080** 0.059** (0.389) (0.303) (0.032) (0.026) Fifth 0.594 -0.782*** -0.028 0.008 (0.520) (0.280) (0.035) (0.021)
First of all, it can be observed that low education cannot explain BMI within income quintiles, since all average marginal effects are insignificant. However, the average marginal effect of high education is significant within almost all income quintiles, though at different levels. For instance, conditional on having an income in the range of the first income quintile, an individual’s BMI is approximately 0.742 lower when this person has a high education level than when this person would have a middle education level. The reduction in BMI is approximately 1.019, 0.778, and 0.782 within the third, fourth, and fifth income quintile, respectively. The average marginal effect of high education within the second income quintile is insignificant. This implies that conditional on having an income in the range of the second income quintile, there is no relationship between high education (compared to middle education) and BMI.
Considering self-assessed health status, the coefficient estimates are shown in Table A3 in Appendix A. The average marginal effect of the education levels, conditional on the income quintiles, are shown in Table 6. The last two columns are based on the second specification in Table A3.
The results show that the average marginal effect of low education is statistically significant at the 10% and 1% level within the second and third income quintile, respectively. For instance,
Notes: Robust standard errors are in parentheses (*** p < .01, ** p <
25 conditional on having an income in the range of the second income quintile, the probability of being in good health is approximately 0.064 lower when an individual has a low education level than when this person would have a middle education level. The reduction in this probability is approximately 0.111 within the third income quintile. The average marginal effect of low education within the fourth income quintile is statistically significant at the 5% level, but has the unexpected positive sign. The average marginal effect of low education is insignificant within the first and last income quintile. Hence, the negative relationship between low education (compared to middle education) and self-assessed health is only observed within the second and third income quintile. Next, the average marginal effect of high education is statistically significant at the 5% level within the fourth income quintile. Conditional on having an income in the range of the fourth income quintile, the probability of being in good health is approximately 0.059 higher when an individual has a high education level than when this person would have a middle education level. Within the other income quintiles the average marginal effect of high education is insignificant. Hence, only within the fourth income quintile there exists a positive relationship between high education (compared to middle education) and self-assessed health status.
6.3.2.2 Income quintiles within education groups. The average marginal effects of the income quintiles, conditional on the education levels, are shown in Table 7. The first four columns correspond to the second specification in Table A2, whereas the last four columns correspond to the second specification in Table A3.
Table 7. Average marginal effects of income conditional on education
(BMI) (BMI) (BMI) (BMI) (SAH) (SAH) (SAH) (SAH)
first second fourth fifth first second fourth fifth
Education level Low -0.116 0.103 -0.968** -0.557 -0.000 -0.011 0.150*** 0.102** (0.442) (0.480) (0.479) (0.594) (0.038) (0.040) (0.039) (0.043) Middle 0.066 0.001 -0.188 -0.701* -0.113*** -0.058** -0.040 0.020 (0.373) (0.368) (0.365) (0.360) (0.029) (0.027) (0.028) (0.025) High 0.344 0.467 0.053 -0.464* -0.071** -0.005 0.033 0.042* (0.390) (0.358) (0.271) (0.254) (0.035) (0.029) (0.022) (0.022) Notes: Robust standard errors are in parentheses (*** p < .01, ** p < .05, * p < .10). The (BMI) columns correspond to
26 Regarding BMI, the average marginal effects of the first and second income quintile are not statistically significant within any of the education groups. The average marginal effect of the fourth income quintile is statistically significant at the 5% level within the low education group, but insignificant within the middle and high education groups. Conditional on having a low education level, an individual’s BMI is approximately 0.968 lower when this person has an income in the range of the fourth income quintile than when this person would have an income in the range of the third income quintile. The average marginal effect of the fifth income quintile is statistically significant at the 10% level within the middle and high education group. Conditional on having a middle education level, an individual’s BMI is approximately 0.701 lower when this person has an income in the range of the fifth income quintile than when this person would have an income in the range of the third income quintile. Within the high education group this reduction in BMI is approximately 0.464. Hence, the first two income quintiles cannot explain BMI within education groups (compared to the third income quintile). However, there is a negative relationship between the fourth or fifth income quintile (compared to the third income quintile) and BMI within some education groups.
27 marginal effect of the fifth income quintile is statistically significant at the 5% level within the low education group and at the 10% level within the high education group. Conditional on having a low education level, the probability of being in good health is approximately 0.102 higher when an individual has an income in the range of the fifth income quintile than when this person would have an income in the range of the third income quintile. The increase in this probability is approximately 0.042 within the high education group. Furthermore, within the low education group, the coefficient of the fourth income quintile is larger than that of the fifth income quintile, but this difference is not statistically significant (F(1, 4,719) = 1.58, p = .21). In conclusion, conditional on education level, there is a relationship between some of the income quintiles (compared to the third income quintile) and self-assessed health. However, there is no clear and consistent positive relationship between income and self-assessed health status within education groups.
Finally, it must be noted that education is a predictor of income, which possibly leads to a problem of “bad controls” if both are included in the analysis. This can lead to biased estimates (Cinelli et al., 2019). Therefore, the choice has been made to focus on the results presented in Table 3 and Table 4 when examining Hypotheses 1 and 3a.
7. Discussion
This section discusses the findings, highlights policy implications, and addresses the most important limitations and directions for future research.
7.1 Discussion of the findings
28 convincing for self-assessed health status than for BMI, as BMI is sensitive to the socioeconomic indicator used.
Second, the findings from Section 6.2 reveal that substantial health disparities within socioeconomic groups exist and that these differ between socioeconomic groups. Regarding self-assessed health status, variation is largest in the lowest income group and decreases when moving to higher income quintiles. A similar conclusion holds for education. For BMI, variation is also largest within the lowest education group and decreases when moving to higher education groups. The same, however, cannot be concluded regarding income, as variation in BMI fluctuates between income quintiles. Therefore, Hypothesis 2 is fully supported for self-assessed health status, but only supported for BMI when education level is the socioeconomic status indicator. However, the results presented in Appendix B do indicate that there is also a negative relationship between within-group variability in BMI and socioeconomic status itself (i.e., also for income). The results in Appendix B also reveal that variation in BMI and self-assessed health status significantly differs between education groups and between income groups. A final conclusion that can be drawn from the results in Appendix B is that the difference in within-group variation in BMI between socioeconomic groups is largely due to differences in the worst health states between socioeconomic groups. Overall, the findings are in line with Ferrer and Palmer (2004). However, again the evidence is stronger for self-assessed health status due to the sensitivity of BMI to the indicator of socioeconomic status.
Third, the results from Section 6.3 provide an answer to the research question: To what extent
can social isolation, education, and income explain health disparities within socioeconomic groups? First, the results from Section 6.3.1 imply that within income quintiles and within
29 for education than for income. Finally, the findings from Section 6.3.2 have revealed that education level can explain BMI and self-assessed health status within some income quintiles, and vice versa. However, it depends on the level of education and the income quintile whether this is the case. Therefore, for both self-assessed health status and BMI Hypotheses 3b and 3c are partially supported.
As a final remark on the results, several robustness checks are performed.8 First, as students are unique in that they typically have a low income, but will generally report good health, the analysis has been reperformed for the sample without any students. The results remain similar. Second, the analysis has been performed with gross household income. As mentioned in Section 4, the income quintiles have been generated using equivalized income. To calculate the equivalent size of the household, weights had to be attached to each household member. Household members aged 14 or older are attached a different weight than children aged younger than 14. Sometimes respondents indicated that they had, for instance, two children living at home, but then only reported the age of one child. This has led to an inaccurate equivalent size for several households. However, including income quintiles based on gross household income instead of equivalized income does not change the results. A different specification includes individuals who reported not to have completed any type of education. In this alternative specification, these individuals are classified as having a low education level. The only difference compared to the main results, is that the negative relationship between the income quintiles and BMI is present in this case. Finally, logit models have been estimated for self-assessed health status and these yield similar results.
7.2 Policy implications
The main finding that can inform policy decisions is that within socioeconomic groups socially isolated individuals are less healthy than non-socially isolated individuals. Therefore, policies and interventions aimed at reducing social isolation and loneliness can help reduce health disparities within socioeconomic groups. One could think of, for instance, the promotion of more or stronger social connections, or the availability of alternative possibilities to engage in social interactions within the community. Furthermore, efforts to improve social skills at the individual level might be worthwhile. Moreover, improving the social climate not only within the community, but also at schools and the work environment may improve health and thereby
30 reduce health inequalities later on. These policies should focus on each socioeconomic group’s health potential, thereby taking a proportionate universalist approach (Broeders et al., 2018). This implies that policies should be universal, but pay particular attention to the most disadvantaged groups (Marmot et al., 2010). Additionally, as the socioeconomic gradient in health has once more proven to exist, education policies should be viewed as a vehicle to reduce health inequalities and improve overall health status in society. Subsidies on higher vocational education and university can help achieve this goal.
7.3 Limitations and further research
Several limitations should be taken into account. First and most importantly is the potential presence of endogeneity, which has proven to be difficult to circumvent. Potential unobserved factors may have caused biased estimates. Therefore, the findings presented in this thesis cannot be interpreted as causal effects. Future research should investigate the potential sources of endogeneity and take this into account by using instrumental variables techniques and/or panel data. Second, this study has been performed using a Dutch dataset and can therefore not be generalized to different countries as such. Results may differ in countries with different healthcare systems or different institutional environments. Hence, future research on health disparities within socioeconomic groups should be conducted in other countries to examine whether the results are robust to different settings.
31 individuals are identified as being socially isolated if they have indicated that they miss having people around them. However, again a different measure may yield different results. Examining different indicators of social isolation, such as the lack of close friendships or not being a member of any type of club, is a suggestion for future research.
Next, a lot of observations have been lost due to missing data. Only complete cases have been used throughout the analysis, which potentially leads to biased results. For instance, the individuals excluded were significantly younger than individuals included in the final sample. There is a positive, nonlinear relationship between age and income (Creedy & Hart, 1979; Lewis, 1989). By excluding relatively young individuals from the analysis, the positive relationship between income and self-assessed health status may be overestimated. Similarly, the strength of the relationship between income and BMI is likely to be understated. Therefore, the finding that there is no negative relationship between income and BMI may be due to the exclusion of relatively young individuals in the sample. Furthermore, younger generations have higher education levels than older generations (Bialik & Fry, 2019). Therefore, the exclusion of relatively young individuals may have caused the observed positive (negative) relationship between education and self-assessed health (BMI) to be underestimated (overestimated). Moreover, among the individuals excluded were more women than among the individuals included in the final sample. Many studies provide evidence of the existence of a gender-wage gap, in favor of men (Bayard et al., 2003; Becker, 1985; Blau & Kahn, 2017). If men have higher incomes, the exclusion of relatively many women may have led to an overestimated relationship between income and self-assessed health, and to an understated relationship between income and BMI. Future research may want to consider a more balanced sample or exploit ways to impute missing data.
32 quintile (Table 6, column 3) has a statistically significant, but positive sign. A suggestion for future research is to test whether this also occurs with different data, and if so, examine why the coefficient has the unexpected sign.
8. Conclusion
33
References
Alessie, R. J. M., Angelini, V., van den Berg, G. J., Mierau, J. O., & Viluma, L. (2019). Economic conditions at birth and cardiovascular disease risk in adulthood: Evidence from post-1950 cohorts. Social Science & Medicine, 224, 77–84.
https://doi.org/10.1016/j.socscimed.2019.02.006
Attanasio, O. P., & Emmerson, C. (2003). Mortality, Health Status, and Wealth. Journal of
the European Economic Association, 1(4), 821–850.
https://doi.org/10.1162/154247603322493168
Barbeau, E. M., Krieger, N., & Soobader, M.-J. (2004). Working Class Matters:
Socioeconomic Disadvantage, Race/Ethnicity, Gender, and Smoking in NHIS 2000. American
Journal of Public Health, 94(2), 269–278. https://doi.org/10.2105/ajph.94.2.269
Bayard, K., Hellerstein, J., Neumark, D., & Troske, K. (2003). New Evidence on Sex
Segregation and Sex Differences in Wages from Matched Employee‐Employer Data. Journal
of Labor Economics, 21(4), 887–922. https://doi.org/10.1086/377026
Becker, G. S. (1985). Human Capital, Effort, and the Sexual Division of Labor. Journal of
Labor Economics, 3(1/2), S33–S58. https://doi.org/10.1086/298075
Becker, G. S., & Chiswick, B. R. (1966). Education and the Distribution of Earnings. The
American Economic Review, 56(1/2), 358–369. Retrieved from www.jstor.org/stable/1821299
Bialik, K., & Fry, R. (2019, February). Millenial life: How young adulthood today compares
with prior generations. Retrieved from
https://www.researchgate.net/publication/337113826_Millennial_life_How_young_adulthood _today_compares_with_prior_generations
Blanchard, R. D., Bunker, J. B., & Wachs, M. (1977). Distinguishing aging, period and cohort effects in longitudinal studies of elderly populations. Socio-Economic Planning
34 Blau, F. D., & Kahn, L. M. (2017). The Gender Wage Gap: Extent, Trends, and
Explanations. Journal of Economic Literature, 55(3), 789–865. https://doi.org/10.1257/jel.20160995
Braveman, P. A., Cubbin, C., Egerter, S., Williams, D. R., & Pamuk, E. (2010).
Socioeconomic disparities in health in the United States: What the patterns tell us. American
Journal of Public Health, 100(S1), 186-196. https://doi.org/10.2105/ajph.2009.166082
Broeders, D. W. J., Das, H. D., Jennissen, R. P. W., Tiemeijer, W. L., & de Visser, M. (2018). Van verschil naar potentieel: een realistisch perspectief op de sociaaleconomische
gezondheidsverschillen, wrr-Policy Brief 7, Den Haag: wrr. Retrieved from
https://www.wrr.nl/publicaties/policy-briefs/2018/08/27/van-verschil-naar-potentieel.-een-realistisch-perspectief-op-de-sociaaleconomische-gezondheidsverschillen
Brunello, G., Fort, M., Schneeweis, N., & Winter-Ebmer, R. (2015). The Causal Effect of Education on Health: What is the Role of Health Behaviors? Health Economics, 25(3), 314– 336. https://doi.org/10.1002/hec.3141
Cinelli, C., Forney, A., & Pearl, J. (2019, August 14). Causal Analysis in Theory and Practice » A Crash Course in Good and Bad Control. Retrieved May 25, 2020, from http://causality.cs.ucla.edu/blog/index.php/2019/08/14/a-crash-course-in-good-and-bad-control/
Conti, G., Heckman, J., & Urzua, S. (2010). The Education-Health Gradient. American
Economic Review, 100(2), 234–238. https://doi.org/10.1257/aer.100.2.234
Creedy, J., & Hart, P. E. (1979). Age and the Distribution of Earnings. The Economic
Journal, 89(354), 280–293. https://doi.org/10.2307/2231602
Cutler, D. M., & Lleras-Muney, A. (2010). Understanding differences in health behaviors by education. Journal of Health Economics, 29(1), 1–28.
35 De Boer, W. I. J., Buskens, E., Koning, R. H., & Mierau, J. O. (2019). Neighborhood
Socioeconomic Status and Health Care Costs: A Population-Wide Study in the Netherlands. American Journal of Public Health, 109(6), 927–933.
https://doi.org/10.2105/ajph.2019.305035
De Boer, W. I. J., Dekker, L. H., Koning, R. H., Navis, G. J., & Mierau, J. O. (2020). How are lifestyle factors associated with socioeconomic differences in health care costs? Evidence from full population data in the Netherlands. Preventive Medicine, 130, 105929.
https://doi.org/10.1016/j.ypmed.2019.105929
Droomers, M., & Westert, G. P. (2004). Do lower socioeconomic groups use more health services, because they suffer from more illnesses? The European Journal of Public
Health, 14(3), 311–313. https://doi.org/10.1093/eurpub/14.3.311
Duncan, G. J., Daly, M. C., McDonough, P., & Williams, D. R. (2002). Optimal Indicators of Socioeconomic Status for Health Research. American Journal of Public Health, 92(7), 1151– 1157. https://doi.org/10.2105/ajph.92.7.1151
Ferrer, R. L., & Palmer, R. (2004). Variations in health status within and between
socioeconomic strata. Journal of Epidemiology and Community Health (1979-), 58(5), 381-387. https://doi.org/10.1136/jech.2002.003251
Grossman, M. (1972). On the Concept of Health Capital and the Demand for Health. Journal
of Political Economy, 80(2), 223–255. https://doi.org/10.1086/259880
Harmon, C., Oosterbeek, H., & Walker, I. (2003). The Returns to Education: Microeconomics. Journal of Economic Surveys, 17(2), 115–156.
https://doi.org/10.1111/1467-6419.00191
36 Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T., & Stephenson, D. (2015). Loneliness and Social Isolation as Risk Factors for Mortality. Perspectives on Psychological
Science, 10(2), 227–237. https://doi.org/10.1177/1745691614568352
Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social Relationships and Mortality Risk: A Meta-analytic Review. PLoS Medicine, 7(7): e1000316.
https://doi.org/10.1371/journal.pmed.1000316
Kivimäki, M., Batty, G. D., Pentti, J., Shipley, M. J., Sipilä, P. N., Nyberg, S. T., … Vahtera, J. (2020). Association between socioeconomic status and the development of mental and physical health conditions in adulthood: a multi-cohort study. The Lancet Public Health, 5(3), e140–e149. https://doi.org/10.1016/s2468-2667(19)30248-8
Knoops, K., & van den Brakel, M. (2010). Rijke mensen leven lang en gezond. TSG, 88(1), 17–24. https://doi.org/10.1007/bf03089530
Lewis, W. C. (1989). On the Relationship Between Age, Earnings, and the Net Discount Rate. Journal of Forensic Economics, 2(3), 69–77. https://doi.org/10.5085/0898-5510-2.3.69 Lleras-Muney, A. (2005). The Relationship Between Education and Adult Mortality in the United States. Review of Economic Studies, 72(1), 189–221. https://doi.org/10.1111/0034-6527.00329
Luo, Y., Hawkley, L. C., Waite, L. J., & Cacioppo, J. T. (2012). Loneliness, health, and mortality in old age: A national longitudinal study. Social Science & Medicine, 74(6), 907– 914. https://doi.org/10.1016/j.socscimed.2011.11.028
Marmot, M. G., Allen, J., Boyce, T., Goldblatt, P., & Morrison, J. (2020). Health Equity in
England: The Marmot Review 10 Years On. Retrieved from
