Older Adults’ Mental Health in China: Examining the Relationship Between Income Inequality and Subjective Wellbeing Using Panel Data Analysis

Abstract

Although people in China are experiencing rapid economic growth, higher income, and better living standards, the level of subjective wellbeing has not risen correspondingly. According to the ‘Easterlin Paradox’, economic growth does not necessarily bring about improvement in wellbeing, because an important part of happiness comes from making comparisons. This study investigates the relationship between income inequality and subjective wellbeing in China, focusing on older adults between 60 and 90 years. Empirical evidence is drawn from the Chinese Health and Nutrition Survey 2006, 2009, and 2011 waves. Using county-level fixed-effects estimation, the analyses show that generally, income inequality is negatively associated with subjective wellbeing, net of individual income. The association between inequality and wellbeing varies between people with rural or urban household registration status, and between people ranked within different income deciles. The association between inequality and wellbeing is stronger for people with urban household registration status, and for people ranked within higher income deciles.

Appendices

Appendix 1

See Table 5.

In this analysis, six groups of occupational positions are constructed based on CHNS’ original occupation variable (B4). They are:

• 1. The service class includes “senior professional/technical” (1), “administrator/executive/manager” (3), and “army officer/police officer” (8).

• 2. The non-manual worker category include “junior professional technical” (2) and “office staff” (4).

• 3. The group of skilled workers/supervisor include “skilled worker” (6), “ordinary soldier, policeman” (9), “driver” (10) and “athlete, actor, musician” (12).

• 4. The category of semi-skilled/non-skilled workers includes “non-skilled worker” (7), “service worker” (11).

• 5. Farmers are as originally defined, coded 5 in the original CHNS data.

• 6. Other occupation includes the rest of the original occupation: 13–16.

The constructed variable of sector in the analysis has grouped the 8 values in CHNS’ original sector variable (B6) into 6 groups:

• 1. The state sector includes “government” (1), “state service/institute” (2) and “state-owned enterprise” (3).

• 2. The collective sector includes “small collective enterprise” (4) and “large collective enterprise” (5).

• 3. The sector of family farming is the origin variable of “family contract farming” (6).

• 4. The individual enterprise sector is the origin variable “private, individual enterprise” (7).

• 5. Three-capital enterprise remained the same, coded from (8).

• 6. Others includes “unknown” (9).

Table 5 Fixed-effects models on wellbeing. Including occupational class and employment sector

Appendix 2

See Table 6.

Table 6 Sensitivity test with missing income variable replaced by random value between 0 and 420 yuan

Appendix 3

See Table 7.

Table 7 County fixed-effects with multiple imputation (MI) and maximum likelihood with missing value (MLMV)

Appendix 4: Individual Income Versus Household Income?

In this study, we chose to use individual income instead of household income for several reasons. First, the rapid process of urbanisation in China has diversified the economy, including the occupational structure in rural areas. Although a large proportion of rural people work in the agricultural sector, many also receive individual income, working in townships and other non-agricultural sectors. This is also the case for older adults. Among surveyed individuals in rural areas, 20% were working with ‘family contract farming’, whereas others worked in the state (20%), collective enterprises (47%), private enterprises (11%) and a few in three-capital enterprises (.3%). For urban older adults, pensions are the most important source of income. For rural elderly, income from their own labour is the most important financial source.

Second, household income may also contain an indirect measure of social support or family network. Because our main interest is in income, we chose to examine how much money individual older adults earn, as well as control for household size, which provides an approximation for the size of an individual’s family network.

We have also tested models based on controlling for individual and household income (see Table 8). The results are very similar. This has also improved the robustness of our models, showing the importance of income inequality on older adults’ subjective wellbeing.

Table 8 County fixed-effects with individual versus household income

Appendix 5: County Fixed-Effects with Different Samples

We have included analysis with separated samples: see Tables 9, 10, 11, 12 and 13. In all models, log Gini is correlated significantly with income inequality. For women, people with lower education, and people with rural household registration, this correlation is significant at 10%. The size of the coefficient does not significantly vary. For models including different samples in Tables 12 and 13, the coefficient does not differ significantly from the original model. The effect is particularly strong for people with high education, and people with an urban household registration, meaning that income inequality is even more harmful to the wellbeing of these two social groups.

For all models, county average income is significantly correlated with wellbeing, indicating the importance of economic development on wellbeing. Being healthy is correlated with better health, but for people with urban household registration the association is not significant. This may be due to the privileged healthcare programmes for urban citizens. When excluding people with the lowest level of wellbeing, the effect of health also loses its significance.

Finally, individual income only has an impact on wellbeing for women, for those with higher education, for those with urban household registration, when excluding the lowest scores for wellbeing, and excluding those with the lowest income.

Table 9 Gender

Table 10 Education

Table 11 Household registration

Table 12 Wellbeing

Table 13 Income

Appendix 6: Control for Adaptive Expectations

Adaptive preferences can have important impact on wellbeing. This robustness check tests the relation between inequality and wellbeing when controlling for adaptive preferences, using changes over time as a proxy.

Model A in Table 14 shows the fixed-effects model with standardised wellbeing as the outcome variable. In this model, we added a variable of income change between waves ($\Delta income = income_{t + 1} - income_{t}$). Both inequality and income changes correlate significantly with wellbeing. Although controlling for income change, the association between income inequality and wellbeing remains solid.

Table 14 Fixed-effects models with income changes and income development included as controls

The coefficient of inequality (− .715) has increased in size, compared to the original coefficient (− .540). One explanation is that income changes have reduced the unexplained variability. Another explanation of the increased Gini is multicollinearity. Income change is highly correlated with individual income, with a Pearson’s R equals to .81. This violates the assumption of instance independence and may raise serious problems.

Therefore, we have also constructed a set of dummy variables to measure income development throughout the three waves See Table 15. If income has increased from a previous wave to a subsequent wave ($Income_{t + 1} - Income_{t} > 0$), the income change is then defined as positive. Similarly, a negative change means that income has decreased between two waves ($Income_{t + 1} - Income_{t} < 0$). A constant income development is indicated by a stable income ($Income_{t + 1} - Income_{t} = 0$). Changes in income between waves are divided into the following categories:

Table 15 Income development: changes in income in period 2006–2009 and 2009–2011

These dummies illustrate the development of income changes. The development can be stable without a turning point, for example, for those whose income increased both in the 2006–2009 and 2009–2011 periods. Some other individuals experienced unstable development with a turning point, as, for example, those whose income increased in 2006–2009, but flattened in 2009–2011.

By using these dummies, we can assess the eventual adaptive preference change. If adaptive expectation plays a significant role in wellbeing, we may assume that, if the income development has been consistent over all three waves, people would have less reason to adjust their expectations/preferences based on economic reasons. On the other hand, if a turning point is observed, the expectations/preferences of the corresponding group of people are more likely to change, in order to face uncertain economic situations.

Incorporating the dummy variables into the fixed-effects model does not influence the relationship between Gini and wellbeing (Model C in Table 14). The size of the coefficient has been reduced minimally. In fact, model C and model A are very similar (almost identical). Compared to people with continuously increased income during all three waves (the reference group), wellbeing decreased significantly for people with constant or decreased income during the first two waves, but experienced changes (both positive and negative) in the third wave (categories 4, 6, 7, and 8).

A final sensitivity test is to separate the sample into two groups: people who have experienced a turning point (categories 2, 3, 4, 6, 7, and 8 in Table 15), and those who have not (categories 1, 5, and 9 in Table 15). Fixed-effects models are carried out to test the relation between inequality and wellbeing separately for these two samples. In Table 16, Model D is the fixed-effects model with the sample for people experiencing changes/turning point, while Model E is estimated for people with the same development trend throughout the three waves. The log Gini is stronger for people with stable income development, which may confirm the effect of adaptive preferences on wellbeing. However, the relation between inequality and wellbeing remains strong and significant for both samples, which is in line with our findings.

Table 16 Fixed-effects models with separated samples of income changes

One alternative is to divide the sample further, into people with a top turning point (upward curve) and people with a bottom turning point (downward curve). However, less than four percent of the observations experienced an upward curve. With such a small sample size, it is difficult to incorporate the fixed-effects models.

Nevertheless, we have tested two subgroups that had sufficient sample observations. Most of the observations have experienced continuously increased income throughout all three waves (83.76%). For this group, wellbeing decreases with .63 standard deviations with one percent increment of the Gini coefficient, controlled for all other covariates. The next largest group is people who have experienced increased income in 2006–2009, but decreased income in 2009–2011 (10.30%). In other words, a group of people experienced a downward curve in income development. For this group, wellbeing decreases by .60 standard deviations with each percentage increment of Gini, holding all other variables constant.

Appendix 7: Control for Household Assets

Wealth is an important factor for an individual’s subjective wellbeing. Three variables are constructed to approach wealth in this robustness test. These are: (a) total value of items (that are available from the survey) owned by the household; (b) value of the dwelling a person currently lives in, if it is also owned by her/him (minus bank loan); and (c) the sum of (a) and (b).

The items owned by the household are:

• 1. Household electrical appliances, including VCR, television, washing machine, refrigerator, air conditioner, sewing machine, electric fan, computer, camera, microwave oven, electric rice cooker, pressure cooker, telephone, cell phone, VCD or DVD player, and satellite dish;

• 2. Means of transportation, including tricycle, bicycle, motorcycle, moto-tricycle, and automobile;

• 3. Farm machinery, including tractor, garden tractor, irrigation equipment, power thresher, household water pump, and others;

• 4. Household commercial equipment, including cooking equipment, carpentry equipment, haircutting equipment, sewing machine, small machine shop tools or equipment, and others.

The descriptive statistics are shown below. Notice that, in order to provide a more intuitive interpretation of the coefficient, we have divided the wealth-related variables by 1,000,000 (Table 17).

Table 17 Descriptive statistics for the wealth variables

These proxies for wealth are tested as control variables with fixed-effects estimations. These variables do not influence the relation between wellbeing and inequality, and the results are very similar to the original model in the manuscript (see Table 18). Therefore, it strengthens our conclusion.

Table 18 Fixed-effects models predicting wellbeing, controlled for wealth

Appendix 8: Some Additional Discussions

8.1 Marital Status

In Model 2 (Table 2 in the article), marital status is not significantly associated with wellbeing, and this may be caused by the small sample size for the group of single older adults. Out of a total of 3311 observations included in the models, only 112 observations (87 distinctive individuals) are single. This is only one per cent of the total sample size. With so few observations, estimations lose power. Because of the small sample size, there are also large variations within this group of single people. The standard error for the single group (.44) is more than ten times larger than that for the group of married people. Therefore, there is no significant difference between single and married people in terms of their level of wellbeing.

It is worth noting that in the robustness check (Table 4), although marital status is still not significant, the effect size of this variable seems to be more correct—the coefficient of having a single status turned out to be negative.

8.2 The Family Structure

In this study, we did not find a significant correlation between household size and older adults’ wellbeing. One hypothesis is that income inequality may confound this relationship. Income inequality can cause higher levels of stress for older adults because it contributes to more pressure, a more fast-paced lifestyle, competition at work and lower levels of social cohesion (Wilkinson 2006). At the same time, bonds among household members may be weakened by greater inequality (Ge and Shu 2001).

Family members living in a place with high inequality may also face more pressure from an increasingly competitive work environment and thus spend less time with the elderly. For urban citizens, inequality means a high-tempo, stressful lifestyle and a competitive labour market. For rural residents, higher inequality may symbolise a higher degree of neoliberal thought and greater workforce mobility. In places where inequality is high, more younger family members in poorer rural households may find work in nearby cities and thus spend less time with the elderly.

Furthermore, family structures have changed; nuclear families are on the rise (Whyte 2003), a large population of rural migrant workers has moved away from their older parents (Whyte 2010), and there is a decline in patrilocal and multigenerational co-residence (Zeng and Wang 2004; Hu and Peng 2015). Increasing inequality has brought about enormous changes for older adults, not all of which are positive. It would be interesting for further studies to examine how inequality correlates with wellbeing for different types of households, taking migration into account.

Finally, how much time children spend with their parents could be an important indicator for measuring the effects of older adults’ social networks. Questions about contact hours were included in the survey, but there are important limitations that prevented us from using this data in the analysis. Only women under 52 years old who were ‘married, widowed, or divorced’ were asked about how much time they spend taking care of their parents. This has excluded all men’s contact with their parents, as well as data about older women. It is also difficult to calculate the total amount of time each older adult has received from all children and children-in-law, since the data can only identify family relationships within each household.