Due to its profound intrinsic value, health is usually regarded as an important criterion to evaluate the development of a society. However, health differences among different groups of people are an objective reality. Health inequality arises when there are avoidable differences in health or differences in major social determinants for health among groups of people with different social advantages (wealth, power, or social status). Health inequality is an important part of social inequality and presents a serious challenge (Guo and Xie 2009) in both wealthy and impoverished countries, thus becoming a major research topic in the social sciences over the past 20 or 30 years.
Although the majority of previous research indicates significant associations between socioeconomic status (SES) and health (e.g., Feinstein 1993; Mackenbach et al. 1997; Wang 2011), the conclusions were based primarily on adult samples, and no consensus for the elderly group has been reached. Some research suggests that the relationship between SES and health would be strengthened among the elderly, but others point out that the relationship could be weakened. In addition, most previous studies were based on people from developed countries, and their conclusions may not be applicable to people in developing countries since there are significant differences in the level of economic development, social systems, and population structure between developing and developed countries. For example, the SES-health relationship may present different forms between different countries or areas. Given the above considerations, the current study focuses on the middle-aged and elderly groups in China, investigating the potential differences in the SES-health relationship between the two groups. Does the difference in health among groups of people with different SES expand or shrink during the elderly stage? Is there a significant difference in the SES-health relationship among different regions, that is, does the health difference among groups of people with different SES get larger or smaller as the area becomes wealthier?
Literature review and research question
Ever since the release of the “Black Report” (Black 1981), health inequality has received increasing attention in academic research. The research indicates that social stratification by health exists in almost all societies, i.e., the average health of people with high SES is better than that of people with low SES (Feinstein 1993; Mackenbach et al. 1997). The following factors constitute health advantages for high-SES groups: low risk of health impairment due to superior working and living conditions (Evans and Kantrowitz 2002; Liu and Tang 2004), advantages in acquiring health care knowledge and in making use of medical information and technology as a result of superior educational background (Glied and Lleras-Muney 2008), advantages in accessing and utilizing medical treatment and public health resources and services (Victora et al. 2000; Xie 2009; Meng 2007), and healthier lifestyles (Wang 2012). Nevertheless, health differences among different SES groups may be the consequence of “selection,” i.e., healthier people tend to attain high SES while less healthy people tend to move to the lower social classes (Dahl 1996; West 1991; Wang 2011), hence leading to greater differences in health among different SES groups.
Although the positive correlation between SES and health has already been well established, no consensus on how this correlation may change with age has been ascertained. Previous studies found that the health differences among different SES groups grew larger prior to the initial period of the middle-aged and elderly stage, but the differences decreased during the elderly period (Beckett 2000; House et al. 1990, 1994). Some researchers define this argument as the “Convergence Hypothesis” (Lowry and Xie 2009). The reason that health differences gradually become smaller during the elderly stage may be that in the elderly stage, the differences in psychological and social risk factors (such as a lack of social relations and social support or the deprivation of the feeling of being in control) faced by people with different SES gradually reduced or even disappeared (House et al. 1994; Lantz et al. 1998). However, the explanation may be the gradual enhancement of the determining effects of biological factors on health; these determining effects may even dominate the effects of socioeconomic factors (Mirowsky and Ross 2008). However, a large number of research works indicate that the effect of SES on health accumulates through the entire human life process; instead of shrinking, health differences among groups of people with different SES expand as age increases, and health inequality during the elderly stage is more severe than during the middle-aged stage (Dupre 2008; Lowry and Xie 2009; Lynch 2003; Mirowsky and Ross 2005; Ross and Wu 1996). Some researchers label this argument as the “Cumulative Advantage Hypothesis” (Lowry and Xie 2009). To summarize, no consensus has been reached on the relationship between SES and health during the elderly stage, and the most current conclusions are based on the empirical evidence from developed countries, which demonstrates the need for ongoing research on this problem in a different socioeconomic setting such as China.
Previous research has pointed out that the SES-health relationship is affected by a country’s social, political, and economic conditions (Lowry and Xie 2009), that is, regional socioeconomic conditions can change the mechanism by which individual SES relates to health. At the country level, some research indicates that the effects of individual SES on health are stronger in developed countries because socioeconomic factors have become the major determinant for health in these countries (Wilkinson 1997). Other research indicates that although the rate of mortality and morbidity among people with low SES was higher in all Western European countries, the inequalities were greater in Sweden and Norway, whereas the differences in the mortality rate among people with different SES in France were the highest among all the Western European countries (Mackenbach et al. 1997). No significant differences in the level of health inequalities were detected among people of different social classes, while differences in the level of inequalities by gender prevailed in Latin America and Caribbean countries (Dachs et al. 2002). In Asia, differences in the effects of SES on health were found within Thailand and Philippines (Zimmer et al. 2004). Differences in the effects of SES on health can occur in different regions of the same country. Since the socioeconomic background of a region has a profound impact on individual health (Pickett and Pearl 2001; Robert 1998; Yen and Syme 1999), the degree to which individual SES and health are associated can be adjusted (Bassuk et al. 2002), thereby making a difference in the impact on groups of people with different SES (Robert 1998). Therefore, if an individual with low SES lives in an area with poor socioeconomic conditions, the health risk factors could double: heath differences among people with different SES could increase in a region with poor socioeconomic conditions; in contrast, if an individual with low SES lives in a region with improved socioeconomic conditions, the resulting feelings of deprivation may have a negative impact on health (Ellaway et al. 2012), therefore widening the gap between the health of different SES groups.
In addition, previous research indicates that the SES-health relationship may differ in different health indicators (Huurre et al. 2005), resulting from the differences in the connotations, properties of the measure of each health indicator, and effects of the external social factors. In social science research, the measuring indicators of health conditions include self-rated health, physical functional status, incidence of disease, and depressive symptoms, among others. Physical functional status and the incidence of disease are relative objective indicators, which are more sensitive in response to socioeconomic factors (Sun et al. 2003). For instance, the research finds that health disparities caused by the socioeconomic status is the largest for some of the most preventable diseases (Phelan et al. 2004). Self-rated health is among the most commonly used and most popular health indicators and is a more tolerant, accurate measure of health conditions and risk factors (Idler and Benyamini 1997). Unlike most health indicators, however, self-rated health relies on the process of subjective cognition. Self-rated health is not only affected by the objective health conditions of an individual but also influenced by individual feelings, cognitive framework, and socioeconomic background. It is an interactive and constructive process involving the objective health condition and the subjective cognition (Jylhä 2009). Differences in self-rated health may not be in a one-to-one relationship with the “objective” difference in health among different groups of people (Dowd and Zajacova 2010). For example, individuals with low literacy or from areas with poor medical treatment and public health conditions are likely to report better health conditions than they actually had because of their lack of knowledge about their risks (Dowd and Zajacova 2010). Most elderly people have a positive attitude toward their health condition; even those who live in nursing homes have positive ratings for their health conditions (Hooyman and Kiyak 2011). As a result, compared to the differences in some objective health indicators, the differences in self-rated health among people with different SES are relatively smaller, especially among the elderly group.
From the above literature review, we find that previous studies on health inequalities mostly targeted developed countries; only a few studies were based on health inequalities in China and they appeared quite recently. The question of whether the consensus obtained from empirical evidence in Western developed countries can be applied to China merits further investigation. In the meantime, we address whether health differences among people with different SES expand or shrink during the elderly stages. Ongoing research on this problem in the context of China’s socioeconomic background could provide new empirical evidence to resolve this question. Additionally, although previous studies took into account the effects of socioeconomic factors on individual health conditions, little has been put forward on whether significant health differences exist in different areas among people with different SES. Finally, the majority of the previous research uses a single health measurement indicator, which overlooks the multidimensionality and totality of health.
Based on the above considerations, this study explores the relationship between SES and different health measurement indicators. Through the use of multilevel data, it examines whether significant differences in the relationship exist among people of different age groups or from different regions. Specifically, this study focuses on the following problems:
-
1.
Although the socioeconomic background in China is quite different from that in developed countries, several factors, including the unbalanced development of urban and rural areas, wider income gaps between people, an incomplete medical insurance system, and the gradual social structure differentiation since the implementation of the reform and open policy, aggravate health inequalities in China. The first research question is thus to further ascertain whether significant differences exist among Chinese people aged over 45 in physical functional status, levels of depression, and self-rated health across different groups of people with different SES.
-
2.
This study investigates whether significant differences occur due to age groups among the health differences of people with different SES. Specifically, will the gap between the health of different SES groups expand or shrink in the elderly stage? Does the age pattern of the SES-health relationship differ with different choices of health indicators?
-
3.
In view of the fact that the areal social and economic conditions are not only important influential factors for the individual’s health but also related to the individual’s SES, this study further examines the areal pattern of the SES-health relationship. Across areas with different levels of wealth, will the pattern of SES-health relationship be different? With the improvement in an area’s wealth, will the gap between the health of different SES groups expand or shrink?
Data and statistical methods for analysis
-
(i)
Data
The data was obtained from the national baseline survey in the “China Health and Retirement Longitudinal Study (CHARLS)” during 2011.Footnote 1 A random sample was taken among individuals in the family aged over 45. The CHARLS baseline survey covers 450 villages and neighborhoods in 150 counties or districts across the country. In addition to the data from individuals and their family members, the study collected information on the communities. After deleting the cases not eligible to the study purpose, the sample size of the study was 12,246, consisting of individuals from 149 county-level units.
-
(ii)
Variables and method of measurement
Three dependent variables were collected in the study: physical functional status, depressive symptoms, and self-rated health. Physical functional status includes the activities of daily living (ADL) and instrumental activities of daily living (IADL), encompassing seventeen activities of daily living in total.Footnote 2 Each of the activities had four possible responses: (1) with no difficulty, (2) with difficulty but can still be completed, (3) with difficulty and need help, and (4) cannot be completed. The four choices were assigned the values of one through four; the total score of physical functional status for each person was obtained by summing component scores across the 17 activities. This study treated the total score as a continuous variable ranging from 17 to 68; higher scores indicate worse physical activity status.
In CHARLS, the depressive symptoms of respondents are measured based on ten items, nine of which are included in the current study.Footnote 3 Each item consists of four choices: (1) rarely or none, (2) not too much, (3) sometimes or half of the time, and (4) most of the time. We assigned each choice a value ranging from one to four and obtained the total score of depressive symptoms for each individual by summing the component scores over nine activities. The study regards the total score of depressive symptoms as a continuous variable ranging from nine to 36, with higher scores indicating worse depressive symptoms.
Self-rated health measures the respondents’ answers to the question, “How do you consider your current health condition?” We grouped the answers “very good,” “good,” and “ordinary” as one categoryFootnote 4, namely self-rated “good” (coded as zero), and the answers “not good” and “very bad” as another category, self-rated “not good” (coded as one).
The explanatory variable in this study is SES. Education, occupation, and income are commonly used measures of SES, but some studies indicate such measurements are primarily applicable to developed countries; whether they can be applied to developing countries merits further study (Zhu and Xie 2007). Given the influence of the specific household registration system in China, we also regard the household register as an indicator of SES. Although treating the household register as an indicator for SES is controversial, we agree with studies indicating that household register forms the most important determinant for the distribution of social resources and power (Wu and Treiman 2004), justifying its use as a measurement of SES (Zhu and Xie 2007). In the present study, household register is divided into two categories: (1) rural household register and (2) urban household register. Educational attainment is classified into four categories: (1) illiteracy, (2) elementary school levelFootnote 5, (3) junior high school level, and (4) high school level or above. Since the occupations among the middle-aged and elderly, Chinese are not dispersed, a fair number of zero counts would be expected if we divided the occupation into more categories, which in turn affects the model estimates. Therefore, we combined some occupation categories and obtained two major categoriesFootnote 6: (1) physical workers, including farmers and workers, and (2) nonphysical workers, including management, professional and technical personnel, officers, and business services personnel. The main occupation before retirement was recorded if the respondent had retired. In the current study, income was classified into four categoriesFootnote 7: annual household income per capita falling (1) below 2300 yuan, (2) between 2300 and 5000 yuan, (3) between 5000 and 10,000 yuan, and (4) above 10,000 yuan. Given the high correlations among these SES variables, we obtained a general evaluation of SES using the latent class analysis (LCA) method. By comparing goodness-of-fit statistics for different models, we identified the optimal model with three latent classes. In other words, given the chosen education, income, occupation, and household variables, respondents could be classified into three groups: high, medium, and low socioeconomic status.
In addition to individual SES, area income level was investigated as an important socioeconomic factor. We used the median household income per capita to measure the area income level, representing the degree of wealth in an area. We also included sex, age, marital status, health behaviors, and other variables in the present study.
-
(iii)
Analysis methods
For the two dependent variables of physical functional status and depressive symptoms, the study used multilevel linear regression models. For the dependent variable, self-rated health, we used multilevel models for binary responses. The basic strategy was moving from simple models to more complex models and from a few variables to more. We started with a simple model and gradually incorporated relevant independent variables. We first investigated the fixed components of the model and then added random components. We began with the lowest level and moved to higher levels.
Above all, we fit a random intercept model with no explanatory variables to examine whether there were significant cluster effects, i.e., model 1.
$$ {\mathrm{y}}_{ij}={\beta}_0+{\mu}_{0j}+{e}_{ij} $$
or
$$ \log \left(\frac{\pi_{ij}}{1-{\pi}_{ij}}\right)={\beta}_{\mathbf{0}}+{\mu}_{\mathbf{0}j} $$
where i denotes a level-one unit, namely the respondent; j denotes a level-two unit, namely a county-level unit; y
ij
is the score of physical functional status or depressive symptoms; π
ij
is the probability of self-rated health as “not-good”; and μ
0j
denotes the residual of the level-two unit, representing the area effect.
Second, we added individual-level explanatory variables to model 1, including sex, age, marital status, and smoking, and obtained model 2. We incorporated individual-level SES variables into model 2, i.e., dummy variables representing medium and highest SES and obtained model 3. Model 4 was obtained by adding interaction terms between SES and age in model 3. Model 5 was obtained from model 4 with additional level-two explanatory variables, namely area income level. On the basis of model 5, model 6 considers the random effects due to SES variables, namely a random coefficient model in which the relationship between SES and dependent variables can vary in different level-two units. Based on model 6, cross-level interaction terms, namely the interaction of SES variables and area income level for investigating whether SES-health relationships are affected by the area income level variable, were added to produce model 7. Model 7 is written as:
$$ \begin{array}{l}{y}_{ij}={\beta}_0+{\beta}_1{\mathrm{ses}}_{2ij}+{\beta}_2{\mathrm{ses}}_{3ij}+{\beta}_3{\mathrm{ses}}_{2ij}\times {\mathrm{age}}_{ij}+{\beta}_4{\mathrm{ses}}_{3ij}\times {\mathrm{age}}_{ij}+\\ {}{\beta}_5{\mathrm{ses}}_{2ij}\times {\mathrm{countyinc}}_j+{\beta}_6se{s}_{3ij}\times {\mathrm{countyinc}}_j+\\ {}{\displaystyle \sum_{k=7}^p{\beta}_k}{x}_{kij}+\left({u}_{0j}+{u}_{1j}{\mathrm{ses}}_{2ij}+{u}_{2j}{\mathrm{ses}}_{3ij}+{e}_{0ij}\right)\end{array} $$
or
$$ \begin{array}{l} \log \left(\frac{\pi_{ij}}{1-{\pi}_{ij}}\right)={\beta}_0+{\beta}_1{\mathrm{ses}}_{2ij}+{\beta}_2{\mathrm{ses}}_{3ij}+{\beta}_3{\mathrm{ses}}_{2ij}\times {\mathrm{age}}_{ij}+{\beta}_4{\mathrm{ses}}_{3ij}\times {\mathrm{age}}_{ij}+\\ {}{\beta}_5{\mathrm{ses}}_{2ij}\times {\mathrm{countyinc}}_j+{\beta}_6{\mathrm{ses}}_{3ij}\times {\mathrm{countyinc}}_j+\\ {}{\displaystyle \sum_{k=7}^p{\beta}_k}{x}_{kij}+\left({u}_{0j}+{u}_{1j}{\mathrm{ses}}_{2ij}+{u}_{2j}{\mathrm{ses}}_{3ij}\right)\end{array} $$
where ses2ij
and ses3ij
variables are two dummy variables, denoting medium and highest SES; countyinc
i
denotes area income level; β
3ses2ij
× age
ij
and β
4ses3ij
× age
ij
denote the interaction terms between SES and age; ses3ij
× countyinc
i
denotes the cross-level interaction, i.e., the interaction between SES and area income level; the coefficients of ses2ij
and ses3ij
are random at level two; and the intercept is treated as a random variable. The covariance matrix for the intercept and the coefficients, denoted by Ω2, often needs to be computed. The term Ω1 is used to represent the covariance matrix at level one and comes with a single variance term at level one.
$$ {\varOmega}_2=\left(\begin{array}{lll}{\sigma}_{u0}\hfill & {\sigma}_{01}\hfill & {\sigma}_{02}\hfill \\ {}{\sigma}_{01}\hfill & {\sigma}_{u1}\hfill & {\sigma}_{12}\hfill \\ {}{\sigma}_{02}\hfill & {\sigma}_{12}\hfill & {\sigma}_{u2}\hfill \end{array}\right)\kern1em {\varOmega}_1={\sigma}_{e0}^2 $$
In general, the estimation for multilevel linear regression models usually proceeds with the maximum likelihood method. Comparison of nested model proceeds with −2LL, the deviance statistic for significance testing. We estimated all model parameters using the lme4 packages available in the R statistical software.Footnote 8