Introduction

Non-communicable chronic diseases (NCDs) have grown in importance and currently represent the main cause of death and economic burden globally (WHO-World health organization, 2016). In the context of China, there has been a rise in the incidence of NCDs and the average age at onset is declining (SCC-State Council of China, 2022). NCDs have become a prominent social and public health problem in China. The rise in the incidence and prevalence of NCDs not only has impacts on premature mortality and morbidity, it also has implications for an array of labour market outcomes: first, the decision to be or not to be in the labour force, i.e. labour force participation, is influenced by the incidence of NCDs; second, for labour force participants, average hours of work may be impacted through the presence of NCDs; and finally, hourly wages paid to labour force participants may also be affected by the presence of NCDs that may erode workplace productivity. While these labour market effects take place among individuals with NCDs, there are also potential spillover effects among household members that may result in changes in their labour force behaviours in order to recoup any erosion in income.

Previous literature on the impact of NCDs on the economy is extensive and includes both macro-economic studies of the impact of NCDs on the economy, in general (Nugent, 2008; Suhrcke & Rocco et al., 2007), and micro-economic studies that trace the effects of NCDs on individual behaviours and health status, (Stephen et al., 2018; Murphy et al. 2020; Xie, 2011).

The economic burden caused by NCDs and the associated reduction in life expectancy have been shown to have a negative impact on macroeconomic growth (Ban’o, 1996; Daar et al., 2007; Suhrcke & Rocco et al., 2007). Much of the empirical literature has focused on the impact of specific diseases and their impacts on macroeconomic growth. For example, cardiovascular diseases (Suhrcke and Urban, 2010), COPD-Chronic obstructive pulmonary disease (Zafari et al., 2021), etc . In addition, there is a large body of literature that examines the micro-economic impacts of NCDs on the economy from the perspective of individuals. This work has tended to focus on the impacts of NCDs on labour supply, labour income, and their impacts on household and individual consumption (Gertler and Gruber, 2002; Gertler et al., 2009; Wagstaff, 2007; Abegunde and Stanciole, 2008; Murphy et al., 2020). The general set of findings from these studies has been the negative impact of NCDs on economic well-being.

At present, there is a paucity of work assessing the economic impact of NCDs in low- to middle-income countries, and the studies that do exist often lack consideration of the mediating role played by labour supply on the incomes of those with NCDs. The onset of NCDs weakens the ability of individuals to work, thereby inhibiting their labour productivity, and accordingly, dampens their stream of income.

There are three key determinants to an individual’s weekly income: labour force participation; average weekly hours of work; and average hourly wages . Zhang & Zhao et al. (2009) studied the impact of NCDs on labour force participation rates in Australia. They demonstrated that the presence of NCDs significantly reduced the rate of labour force participation. While there are studies examining labour force participation, we are unaware of similar studies from low- to middle-income countries that assess the effect of NCDs on working hours or hourly wages. Additionally, there may be differential effects of each type of NCD on the various labour market outcomes and individual income. Furthermore, the studies that have examined the impact of specific NCDs, such as diabetes, on labour market outcomes have tended to treat NCDs as if they were exogenous (Bastida & Pagán, 2002; Pedron et al., 2019). However, because the incidence of NCDs is related to factors such as age and gender, and in turn, these factors are related to labour market outcomes and individual income, the treatment of NCDs as an exogenous variable in regression analyses may lead to estimation bias.

Consequently, the purpose of this study is to assess the effects of NCDs on individual incomes and the manner in which those effects are mediated through various labour market behaviours after controlling for potential confounding. To guide empirical research on the economic impact of NCDs on individual income, it is helpful to understand the potential causal relationships between disease onset and the associated economic consequences, particularly those related to labour market behavours. To this end, this paper uses two waves (2016 and 2018) of the China Families Panel Survey (CFPS) to analyze the impact of NCDs on various labour market outcomes, and in turn, on individual income using propensity score matching and difference in difference (PSM-DID) methods to control for potential confounding.

Study framework and study methods

Conceptual framework and study hypotheses

The purpose of this study is to analyze the impacts of the presence of NCDs on individual income and the way that effect is mediated through various labour market behaviours.

Conceptually, individual income, Y, is derived from two sources: employment income, E, and other income sources, O. The former depends on the product of three variables: first, the binary decision to be in or not to be in the labour force, L; second, the average number of hours of work over a given time frame for those in the labour force, H; and finally, the average hourly wage paid to labour force participants, W. Consequently, individual income is defined as

$$Y = E + O = L \ast H \ast W + O$$

This study will assess how weekly individual income is impacted both directly and indirectly by the presence of NCDs. The focus of this paper is on the impact of chronic diseases on earned income. Unless otherwise specified, the income referred to in this article refers to earned income, including both full- and part-time income, but excluding asset income (rent, stock dividends, interest, etc.), income from donations and other sources of income. Five main hypotheses are advanced. The presence of NCDs will:

H1: Significantly reduce weekly individual income;

H2: Significantly reduce the probability of labour force participation;

H3: Significantly reduce the number of weekly hours worked;

H4: Significantly reduce hourly wages; and holding labour market behavours constant

H5: The presence of NCDs will significantly reduce individual income.

A challenge posed with the use of linear regression methods to evaluate the impact of NCDs on income and labour market behaviours arises through potential sample selection biases (for hours of work and hourly wages) and potential endogeneity of NCDs per se. To address these statistical challenges, we use sample matching methods to evaluate the impact of NCDs on individual income and labour market behaviours. We logarithmically transform personal income and hourly wages that are right-skewed and working time that is left-skewed to ensure asymptotic normality and to reduce potential heteroscedasticity.

Methods

To assess the impact of NCDs on individual income and labour market behaviours, we divide our sample into two groups: a treatment group; and a comparison group. Individuals with NCDs fall into treatment group while those without NCDs fall into the comparison group. Due to the non-randomness of natural experiments, it is easy to cause endogeneity due to confounding factors and potential sample selection bias. For example, the average age of the treatment group (i.e., individuals with NCDs) is significantly greater than that of the comparison group, and age is significantly related to individual income. Additionally, some unobserved factors related to gender (whether smoking, drinking, eating habits, etc.) may increase exposure to NCDs and also affect individual income and labour market behaviours.

Rosenbaum and Rubin (1983) proposed propensity score matching (PSM) methods, based on the work of Rubin (1974), To operationalize these methods, we first select covariates that may lead to endogenous effects, such as age, gender, etc., and then match the treatment group members (individuals with NCDs) with similar comparison group members (individuals without NCDs) based on these covariates. We then use the selected samples for comparison. Rosenbaum and Rubin (1983) showed that matching the treatment group and the comparison group with appropriately selected covariates can eliminate the confounding factors and potential sample selection biases in the data. The propensity score matching and difference-in-difference (PSM-DID) method proposed by Heckman et al (1997, 1998) was developed based on the work of Rosenbaum and Rubin (1983). It’s a common way of controlling for potential confounding and selection bias coming from time trend data. We use this method in this paper. PSM-DID has two advantages, one is to eliminate confounding factors through sample matching methods (PSM), and the other is to control for differences between groups that cannot be observed and do not change with time through time differencing. These two advantages ensure that the difference between the treatment group and the comparison group as estimated by PSM-DID is more reliable than the use of linear regression analyses.

There are two steps taken to operationalize the PSM-DID method. In the first step, we divided the sample into treatment and comparison groups by year. The treatment group included individuals of working age with NCDs. The comparison group comprised working-age individuals without NCDs who had similar characteristics (age, sex, etc.) to those in the treatment group. Propensity scores were estimated from the treatment and covariate variables. The comparison group was then selected to match the treatment group through the use of the propensity score. After grouping, the samples of individuals in each group were divided into three categories, one for individuals who were without NCDs in both 2016 (base period) and 2018 (observation period), and another for individuals who reported NCDs in both 2016 and 2018. The third category only comprised individuals who reported NCDs in 2018 and their absence in 2016 as there were no individuals who reported NCDs in 2016 and their absence in 2018. In the second step to operationalize the PSM-DID, comparisons were made between time-dependent differences for the treatment and comparison groups. The method requires the computation of a dummy variable Ncd to distinguish between individuals with NCDs and individuals without NCDs. Ncd = 1 represents those individuals with NCDs, and Ncd = 0 represents those without NCDs. A dummy variable Time is used to distinguish between the base period and the observation period. Time = 0 represents the base period, while Time = 1 represents the observation period. The difference-in-differences model was specified as follows:

$$Y_{it} = \beta _0 + \beta _1Ncd_{it}\, * \,Time_i + \beta _2Ncd_i + \beta _3Time_i + \gamma X_{it} + \varepsilon _{it}$$

In this formulation, the dependent variable, Yit, represents the outcome of interest, in this study it represents income, working hours, hourly wages, etc. The independent variable, Xit, represents the vector of covariates, including age, gender, education level, etc. (see below for details) that are expected to be associated with variations in the outcome variable(s). The disturbance term is represented by εit. Based on the DID model, the coefficient, β1, on the interaction term, Ncd * Time, reflects the impact of chronic diseases on the explained variables such as individual income and working hours.

The main advantage of the DID method is that it proposes the bias caused by the Pretreatment difference between chronic disease individuals and non-chronic disease individuals. The PSM method is used to reduce the confounding caused by the difference in characteristics between those with NCDs and those without NCDs. To ensure that are findings are robust, we employ two methods in the fifth part to test the robustness of the core assumptions.

The main purpose of this paper was to study the average treatment effect of the treated (ATT) on individual income and various labour market outcomes between the treatment and comparison groups in order to ascertain the impact of NCDs.

Variables

We use PSM-DID methods to conduct a series of regression analyses to answer several hypotheses raised in section 3.1. First, the impact of NCDs on individual income. In this question, patients with NCDs are set as the treatment group, and other populations are set as controls. The key explanatory variable is a dummy variable, N, that is set as unity when individuals have any NCD and zero otherwise. The dependent variable is logarithmically transformed individual income to deal with the skewness in the data and the covariates used to control for other influential effects on income include education level, gender, age, economic region where the family is located, and whether individuals are urban or rural residents.

The second hypothesis (H2) concerns the impact of NCDs on the probability of labour force participation. The dependent variable is the binary outcome to be or not to be a member of the workforce and the covariates are similar to those used to assess H1. Likewise, the third hypothesis (H3) concerns the impact of NCDs on weekly hours of work. The dependent variable is censored and represents the number of weekly hours of work for those that participate in the labour market and the covariates are similar to those used to assess H1. The fourth hypothesis (H4) examines the impact of NCDs on hourly wages. The dependent variable is censored and represents hourly wage for those in the labour force, and again, the covariates are similar to those used to assess H1.

Because our study concerns the impact of NCDs on various labour market outcomes, we restricted the sample to those of working age only (i.e., 18 to 60 years of age for men and 18–55 years of age for women as the legal age for retirement in China for women and men is 55 and 60, respectively.

There are a wide range of geographic socio-economic disparities in China, Liu et al. (2004) and there exist significant differences in income and living habits among different economic regions. To accommodate these differences, we defined and included in our analyses a regional economic variable, region, to control for the heterogeneity of economic regions. The variable region is a dummy variable, representing the economic region of residence for each sample participant. While it was possible to capture each of the thirty provinces from which the sample was drawn, that unit of analysis would have included some provinces with very few participants. Instead, we used four regional dummy variables that are commonly used in policy analyses: Eastern regions; Western regions; Central regions; and regions in the North-East of China. The Eastern region is generally considered the most developed region of China and the Western region is the least developed.

There are also differences in income levels and consumption habits between urban and rural areas (Li and Han 2011). Therefore, we defined an additional dummy variable, urb, to control for differences between urban and rural areas. This variable was set at unity for those participants who resided in urban regions and zero otherwise.

The level of educational attainment plays an important role in determining r individual income, and there exits an extensive literature demonstrating the association between educational attainment and the occurrence of NCDs (Rheault et al., 2019; Coskun & Bagcivan, 2021). Given these effects, the educational level of each participant was used as an additional control variable. The China Families Panel Survey (CFPS) data divides each residents' educational level into eight groups, namely illiterate, primary school, junior high school, high school and vocational college, junior college, undergraduate, master and doctorate. We used this classification scheme to define a rank variable edu = 1, 2, …, 8 corresponding to the eight educational levels.

Age was captured as a continuous variable and gender, gend, was defined as unity for men and zero otherwise. These variables were included as additional control variables in all analyses.

Data source and study samples

The China Families Panel Survey (CFPS) is a longitudinal survey that has been conducted every two years since 2010, and the sampled households cover most provinces, municipalities and autonomous regions of China. The original data included individuals of all ages, but in our study of labour market activity, we restricted the sample to only those individuals in work age. The 2018 survey represents the latest publicly available data. The questionnaire used in each wave of the CFPS varied slightly. For example, weekly hours of work hours was only asked of participants since 2014. As a result, we selected two years of survey data, namely 2016 and 2018, as the questionnaire over those two most recent waves has been relatively stable.

Individuals with NCDs in the survey data refer to the individuals diagnosed with at least one chronic disease in the current period. CFPS data divides chronic diseases into 21 categories and 131 diseases (http://www.isss.pku.edu.cn/cfps/wdzx/sjwd/1357972.htm). The first category is infectious chronic diseases, including hepatitis B, HIV, tuberculosis, measles and so on. The second category is parasitic diseases, including schistosomiasis, malaria and so on. NCDs in this paper refer to patients with the third largest category of malignant tumors, up to the last category of unspecified genetic diseases, unspecified stones and other diseases. See the detailed diseases list in the appendix. Some individuals suffer from a variety of NCDs, but the survey data only counts the two most important NCDs for each individual. For individuals with multiple NCDs, we use data on the chronic disease that is reported to have had the greatest impact on the life of the respondent.

Over the survey period, because some individuals may die or be lost to follow-up, the analysis sample will be unbalanced. To study the impact of NCDs on individual income, we combined the two survey waves (2016 and 2018) to form a balanced panel. To balance the data, we used the unique identification number for each individual and only included pairs of observations, i.e., individuals who were respondents to both survey waves. Because the purpose of this study was to analyze the impact of NCDs on the income of individuals of working age, only observations on individuals aged between 18 and 60 years for men and 18 and 55 for women were retained. In addition, other exclusion criteria were employed, such as, observations with missing data on the occurrence of NCDs and observations that were outliers with respect to individual income. In the original data, there were 58,504 respondents in 2016 and 58,179 respondents in 2018. There were 30,465 respondents who responded to both survey waves and were of working age. There were 2148 respondents who had missing data on the occurrence of an NCD and were therefore excluded. Additionally, there was one respondent with outlier income in 2016 of 10.3 million Rmb, just over $1.5 million US in 2016. After imposing these inclusion and exclusion criteria, the resulting analysis sample was composed of 28,317 individuals that responded to both the 2016 and 2018 waves of the CFPS and were of working age. In 2016, there were 4616 (16.3%) respondents with at least one NCD and 23,701 individuals without a NCD. The equivalent proportions for 2018, were 16.7% and 83.3%, respectively.

Results

Descriptive statistics

Basic descriptive statistics for our sample stratified by survey year are reported in Table 1. The most common NCD is Cardiovascular and cerebrovascular. The proportion of women in the group with NCDs was significantly lower than that in the group without NCDs. In addition, the average age, employment rate, etc. of the group with NCDs and the group without NCDs also have significant differences. Individuals with a NCD reported significantly (p = 0.002) lower income than those without a NCD. Since the average age (p < 0.001), education level, and gender of individuals with NCDs were also significantly different from those without an NCD, it is necessary to control for these covariates in our main analyses to avoid potential endogeneity bias when analyzing the impact of NCDs on personal income.

Table 1 Basic statistics of individuals with and without NCDs.
Full size table

Sample matching effect test

In this paper, the method of kernel matching was selected for sample matching. After sample matching, there were 534 individuals in the treatment group and 5476 individuals in the comparison group when we did not stratify by sex. These groups were used to ascertain the impact of NCDs on individual income. When the analysis was stratified by sex, there were 274 men in the treatment group and 3167 men in the comparison group. In the regression assessing the impact of NCDs on the income of women, there were 207 women in the treatment group and 2188 in the comparison group. To test for the matching effect, we conducted a balancing test on the samples before and after matching. From the test results (see appendix), there is no difference in gender, educational attainment, and other characteristics between the treatment and comparison groups after sample matching. There are still significant differences in age alone, but the age difference between the treatment group and the comparison group was greatly reduced when compared with the full sample. For example, in the regression study of the impact of NCDs on income regardless of sex, the difference in age between those with NCDs and those without NCDs was 7.81 years, but in the matched samples the difference fell to 0.862 years. In the sex stratified regression analysis, the age difference between the treatment and comparison groups for both sets of analysis (i.e., for men and women) was less than 1 year compared to that in the matched analysis of 0.958 and 0.768 years for men and women, respectively. See Appendix for balancing tests results. In addition, these results were maintained when we used either radius matching or kernel matching.

The impact of NCDs on labour force income

In this paper, PSM-DID was used to estimate the average treated effect (ATT) for each explanatory variable, and the kernel matching method was used to match the treatment and control samples. The results are shown in Table 2.

Table 2 PSM-DID test results of the economic impact on labour force.
Full size table

The results demonstrate that individual income for those with NCDs was significantly lower by 19.2% than that for individuals without NCDs. There are three mechanisms that account for the decline in income: first, labour force participation was lower by 5.1% (p = 0.0044) among those with NCDs compared to their counterparts; second, average weekly hours of work were lower by 5.5% (p = 0.0024) for those with NCDs; and finally, average hourly wages were 8.8% (p = 0.0031) lower for those with NCDs. Consequently, approximately 45%, 29%, and 26% of the decline in earned income due to the presence of NCDs were associated with lower wages, fewer hours of work, and reduced labour force participation, respectively.

In terms of gender, after holding labour market behaviours constant, the income of men with NCDs was 20.1% lower than their counterparts without NCDs. In the case of women, incomes were 15.2% lower among women with NCDs than their counterparts. Labour force participation and average hourly wages were significantly lower for men with NCDs, but their average weekly hours of work were not significantly lower than men without NCDs. In the case of women, average weekly hours of work were significantly lower among women with NCDs, but both labour force participation and average hourly wages were not significantly lower than their counterparts without NCDs.

We further analyzed the impact of specific diseases on individual income for those with NCDs. We classified NCDs into six categories: cardiovascular and cerebrovascular diseases; musculoskeletal diseases; digestive system diseases; respiratory system diseases; diabetes; and other diseases. We examined the impact of each set of NCDs on individual income with the results shown in Table 3.

Table 3 PSM-DID test results of patient income by disease.
Full size table

Table 3 demonstrated the differential effect of NCDs on individual income. Musculoskeletal diseases were shown to have the largest impact on reducing individual income, while digestive system diseases had the smallest negative impact.

Robustness test

Two methods were used to test the robustness of the main study conclusions: one used the PSM method; and the other used the method of intercepting some samples to carry out the robustness test.

Robustness test of withdraw some samples

The labour market behavior of individuals with NCDs was shown by Sapkota et al. (2021) to be sensitive to family income. Sapkota et al. (2021) demonstrated that in families with very low incomes, some individuals with chronic diseases will increase their working hours at the expense of their health to maintain family income and ensure sufficient funds for medical expenditures. In contrast, when family income is sufficiently high, the presence of NCDs will tend to result in withdrawal from the labour market to allow individuals to recuperate. To verify the robustness of our results, we exclude 10% of individuals at each end of the family income distribution based on income reported in 2016. Using the personal ID, we removed the corresponding observations from the 2018 wave and repeated the PSM-DID test to identify the effect of NCDs on individual income and various labour market behaviours. The results are shown in Table 4. From the results of Table 4, we demonstrate the robustness of our earlier estimates. Similarly, when we repeat the analysis by excluding 5% of individuals at each end of the family income distribution in 2016, the personal income of those with NCDs was 18.5% (p = 0.003) lower than for those without NCDs. This compares well to the results from the full sample where income was 19.2% (P = 0.0019) lower. The reduction in labour force participation, average weekly hours of work, and average hourly wages remain statistically significant and similar to our earlier estimates.

Table 4 The robust test (exclude some samples) of the NCDs economic impact on labour force.
Full size table

The robustness test results of the impact of specific diseases on personal income are shown in Table 5. From the results of Table 5, the impact of specific types of NCDs on income also change slightly. The occurrence of musculoskeletal diseases still resulted in the largest reduction in personal income (25.9%, p < 0.0001), compared with the results of full sample regression at 21.5% (P < 0.0001). While the occurrence of digestive system diseases still has the smallest reduction in personal income (6.6%, p = 0.0135). Compared with the results of full sample regression, this value is 6.9% (P = 0.012).

Table 5 The robustness test (exclude some samples) of the different diseases impact on personal income.
Full size table

Robustness test of PSM method

The DID method needs to satisfy the parallel trends assumption. The length of the data series available in this paper was not enough to test the parallel trend of the samples. Instead, we used PSM methods to the same samples to test the main hypothesis in this paper. The results are shown in Table 6 that show that chronic diseases have a significant impact on personal income, the rate of labour force participation and hours of work, which is similar to the results obtained by using the PSM-DID method. These findings demonstrate the robustness of our study findings.

Table 6 The robust test(PSM test) results of the economic impact on labour force.
Full size table

Discussion

Summary of the study finding

This paper used the PSM-DID method to assess the impact of NCDs on the income of working-age individuals, and further analyzed the impact of NCDs on various labour market behaviours, specifically: labour force participation; average weekly hours of work; and average hourly wages. Three key sets of findings are to noted.

(1) Working-age individuals with NCDs reported significantly lower personal incomes than their counterparts. We identified several mediating factors that together contribute to this decline in income. First, there was a tendency for individuals with NCDs to withdraw from the labour force. Compared with those without NCDs, the labour force participation rate for individuals with NCDs was 5.1% lower than their counterparts. Second, the average number of hours of work for individuals with NCDs was 5.5% lower than that for individuals without NCDs. Third, average hourly wages received by those with NCDs was 8.8% lower than that for individuals without NCDs. Together, these labour market outcomes resulted in a 19.2% decline in personal income for those with NCDs relative to their counterparts. These findings demonstrate the magnitude of the income reduction for individuals with NCDs as summarized in the raw differences. The test results (appendix d) of covariates show that both age and gender are significantly related to the prevalence of NCDs. Specifically, the average age of individuals with NCDs was greater than that for those without NCDs. Also, the proportion of men with NCDs was significantly higher than that for women with NCDs. Moreover, as age is positively correlated with personal income (Lu et al., 2022), and the average income of men is higher than that for women, there was a need to control for age and gender to ensure that we did not underestimate the erosion in personal income attributable to the presence of NCDs.

(2) There were differential effects of NCDs on the personal income of men and women. Men suffer a larger decline in income in the presence of NCDs than women, 20.1% v 15.2% (Table 2). This differential was expected given the lower labour market outcomes of women compared to those for men. Furthermore, the main source of the decline in income for men was attributed to the significant reductions in both labour force participation and hourly wages, while the decline for women was primarily attributable to lower hours of work. NCDs have differential effects on the income of men and women. There are two potential reasons for these effects. First, there are significant differences in the industries and employment categories in which men and women are employed. For example, US Department of Labour survey data for 2004 shows that the proportion of women in the tool and fastener manufacturing industry was only 4%, while the proportion of women in secretarial positions was high at 97%. In China, there are also differences between men and women in their choice of employment industry (National Bureau of Statistics of China, 2020). For example, women account for only 11.5% of individuals employed in the construction industry, while the proportion of women employed in the health services industry was 66.5%. Second, the types of NCDs that arise among men and women are also different. We showed that the proportion of men with cardiovascular and cerebrovascular diseases were greater among men than women, while the opposite takes place with respect to respiratory diseases. The data reported in Table 3 demonstrates the differential effect of different types of NCDs on personal income. Consequently, the differential exposure to different types of NCDs and their different effects on personal income along with differences in patterns of employment by sex, accounts for the differences in the effects of NCDs on personal income and labour market behaviours for men and women.

(3) The main diseases confronted by individuals in the sample include cardiovascular and cerebrovascular diseases, musculoskeletal diseases, respiratory system diseases, digestive system diseases and diabetes. The economic impact of various diseases on personal income is statistically significant, but to varying degrees. Among them, musculoskeletal diseases have the greatest impact resulting in a 21.5% decline in income. The second most impactful set of NCDs are respiratory diseases, which were shown to reduce incomes by 20.4%, while cardiovascular and cerebrovascular diseases lowered incomes by 17.4%, diabetes by 9.8%, and digestive system diseases had the smallest impact on incomes at 6.9%. Different types of diseases have different consequences, and the causal mechanism of these differences needs to be further studied. An intuitive reason is that different diseases have differential harms to health and to the ability to engage in labour market behaviours, resulting in differential effects on labour force participation, hours of work, and hourly wages.

In Contrast to Prior Literature

To date, we have not found literature that have adopted an analytical framework similar to that used herein. A few studies have examined NCDs in general, but most focus on only a single specific NCD (Jaspers et al., 2015). These studies report common findings that the presence of NCDs are associated with a reduction in income and an increase in both health service utilization and health expenditures. Together, these effects result in a heavy economic burden to individuals and their families. Similarly, the empirical study by Goel et al (2018) found that 47% of farmers or labourers included in the study with a cervical spine injury lost their jobs with an associated reduction in income and increased health expenditures.

The impact of diabetes on labour market participation rate has been well studied (Vijan et al., 2004; Harris, 2009; Latif, 2009; Lin, 2011; Rumball-Smith et al., 2014; Zafari et al., 2021). But there are few studies on the impact of diabetes on income. Kraut et al. (2001), based on a sample set of diabetes patients in Manitoba, Canada, studied the labour market participation rate and income changes of diabetes patients. Among the 608 diabetes patients in the sample, 242 had complications, and the income of diabetes patients with complications was 72% of that of healthy individuals. The income of diabetes patients without complications was not significantly different from that of healthy individuals. Therefore, overall, the income of diabetes patients is equivalent to 89% of that of healthy individuals. This is similar to the results of this paper (the income of diabetic patients decreased by 9.9%).

Moreover, there is also a lack of systematic research on the impact of Cardiovascular and cerebrovascular, respiratory system disease and digestive system disease on patient income and labour market behavior. In general, this paper verified the negative impact of NCDs on patient income and labour market behavior and conducted quantitative analysis, opening a new direction for subsequent research.

Limitations

There are several limitations to note. First, our study was limited in its use of self-reports from only two waves (2016 and 2018) of the China Families Panel Survey (CFPS). While we only used the two most recent waves of the CFPS in order to ensure the availability of data to consistently worded questions, the resulting time series data is time limited. It is therefore difficult for us to verify the stability of our conclusions and to generalize the findings more broadly. Consequently, it is challenging to assess the long-term impacts of NCDs on personal income, and on the associated labour market outcomes without recourse to further waves of survey data. This is part of our future research plans.

Second, occupational variables, such as manager, professional, labourer, etc were not included as covariates in our analysis, because the CFPS data lacked such information. In future work, it would be useful to identify and control for unique, person-specific occupational data on the places of employment and other variables to control of potential confounding and for the assessment of occupation specific policy implications.

Finally, the study was limited to China, and while China is a vast and populous middle-income nation, there are limitations in generalizing our findings to other jurisdictions. Despite this observation, many of the methods employed and our conceptual framework outlined are amenable for application to the issue of the impact of NCDs on income and associated labour market outcomes in other jurisdictions.

Conclusions

This paper represents an empirical analysis of the impact of NCDs on income and labour market outcomes of working-age individuals in China. The results of this paper provide reference for policy formulation in several regards. When our results are combined with estimates of the trends in NCDs, forecasts of the economic burden of NCDs can be derived in terms of the effects on individual income and labour market outcomes. These estimates may in turn be used to evaluate the costs and benefits of proposed NCD preventive and curative interventions. Likewise, the findings derived herein may be used to inform policy prescriptions that address the significant erosion to economic and personal well-being associated with the presence of NCDs.