Abstract

Background

After more than 40 years of growth, the income of Chinese residents has greatly increased; however, the problems of overweight and obesity among residents have become increasingly prominent.

Methods

We used data from the China Family Panel Studies (CFPS) to study the relationship between residents鈥� income and obesity using the instrumental variable (IV) method.

Results

The impact of income on residents鈥� body mass index (BMI) is an inverted U-shape; that is, when income is low, BMI significantly rises with income, and when per capita income exceeds 57,066 yuan in 2023 prices (equivalent to 8,098 dollars), further increases in income will lead to a decrease in BMI.

Conclusions

The results suggest that the impact of income on resident obesity may be related to dietary behaviour and health investment. Although income increases the likelihood of health expenditure and exercise, it has an inverted U-shaped effect on whether residents consume fish, meat, fried or pickled foods, and dine out. That is, in the low-income stage, income mainly increases consumption of unhealthy foods such as fish, meat, and fried and pickled foods, leading to a significant increase in BMI. In the high-income stage, residents reduce consumption of large amounts of fish and meat, pay more attention to healthy diet and healthcare, and increase exercise, which leads to a decline in BMI and an overall inverted U-shaped impact of income on obesity. Further heterogeneous analysis showed that income has a greater impact on obesity among rural residents, those aged 50 years and older, and those with low education levels. Finally, this study provides relevant suggestions for the prevention and control (P&C) of obesity among urban and rural residents.

Introduction

Obesity is prevalent globally, with the number of overweight and obese individuals across all age groups continuously expanding; this figure is expected to rise further, making obesity one of the most serious global public health issues. According to data from the World Health Organization (WHO), the prevalence of overweight or obesity among children and adolescents aged 5鈥�19 years grew more than fourfold from 4% in 1975 to 18% in 2016. The 2023 edition of the Global Obesity Map released by the World Obesity Federation estimates that by 2035, more than four billion people worldwide will be overweight or obese, accounting for 51% of the global population.

Similarly, China faces the serious problem of obesity. With rapid economic and social development as well as lifestyle changes, the prevalence of overweight and obesity among Chinese residents has risen rapidly. According to data from the Chinese Residents鈥� Nutrition and Chronic Disease Status Report (2020), the overweight rate among Chinese residents aged 18 and older is 34.3%, with an obesity rate of 16.4%. China has become the country with the highest number of obese individuals globally [1].

Obesity is an established risk factor for chronic non-communicable diseases such as cardiovascular disease, type 2 diabetes, and certain cancers [2,3,4], and can also have negative impacts on individual employment and wage income [2,3,4,5] increasing societal healthcare costs [5]. According to Qin & Pan (2016) [6] the estimated annual medical expenses related to overweight and obesity in China are as high as 24.35听billion RMB and could potentially reach 418听billion RMB by 2030, accounting for approximately 22% of the total national healthcare expenditures [7]. Therefore, effective prevention and control (P&C) of overweight and obesity, as well as improving residents鈥� health, have become important public policy issues.

Obesity and overweight are complex conditions influenced by genetic, economic, and social factors. Income, as a significant factor, has long been a focal point of debate in the academic community regarding its relationship with obesity. Early studies have primarily found a positive correlation between income and obesity, with supporting evidence from low-to middle-income countries [8,9,10,11,12]. However, some scholars argue that there is a negative correlation between income and obesity [13,14,15,16]. Moreover, the impact of income on obesity may be non-linear because the measurement of obesity in terms of weight (or BMI) has unique characteristics compared with many other health outcomes. At a lower weight (or BMI), an increase in weight (or BMI) suggests an improvement in health status, whereas at a higher weight (or BMI), a decrease in weight indicates an improvement in health. Lakdawalla & Philipson (2009) [10] revealed that the derivative of an individual鈥檚 preferred steady-state weight with respect to income follows a U-shaped curve, where an initial rise in income may lead to weight gain, but this effect may reverse with further income increases, potentially reducing weight. Grecu & Rotthoff (2015) [11] described this trend as an obesity Kuznets curve. Similar conclusions have been drawn in studies using macro data [12,13,14,15,16].

Despite the extensive research on this topic, this study has some limitations. First, there is no consensus among Chinese studies, and the main database used was the China Health and Nutrition Survey database, with the latest data available up to 2015, which might not accurately reflect the most recent situation in China. Second, endogeneity in empirical studies is a critical issue. However, existing studies have not adequately addressed endogeneity, leading to potentially biased estimations. Third, there is a lack of empirical research on the mechanisms through which income influences obesity. While some studies indicate that income may affect individual obesity through the intake of fats, proteins, carbohydrates, dietary diversity, and knowledge, these findings might not be comprehensive.

We used data from the China Family Panel Studies (CFPS) for 2012, 2014, 2016, and 2018 and employed the instrumental variable (IV) method for empirical research. We aimed to analyse the impact of income on obesity among urban and rural residents, investigate the mechanisms at play, and explore the heterogeneous effects among different population groups. The goal was to deepen our understanding of how income influences obesity and provide empirical evidence to inform policy interventions for obesity P&C.

This article offers several innovations. First, it expands the analysis of the mechanisms through which income affects residents鈥� obesity. In contrast to previous studies focusing on diet and nutrition, we found that income may influence residents鈥� obesity not only through dietary habits but also by increasing health investments (such as non-medical healthcare spending and the likelihood of engaging in exercise). Second, by considering factors such as place of residence, age, and years of education, we carefully examined variations in the impact of income on obesity among different population groups, thereby enriching the research outcomes on the relationship between income and obesity. Third, in terms of research methodology, this article addresses the endogeneity issue of income. The potential endogeneity of income may arise from unobserved heterogeneity and possible reverse causality. Employing the IV method to address the endogeneity of income enhanced the reliability of the estimation outcomes.

This article is structured as follows: The second section presents the hypotheses, the third section describes the data and models used in the empirical research, and the fourth section contains the empirical findings. Finally, conclusions and policy recommendations are provided in fifth section.

Theoretical analysis and hypotheses

We drew upon on the neoclassical theory of obesity proposed by Lakdawalla & Philipson (2009) [10] to elucidate the relationship between income and obesity. Health is a normal good, and an increase in income has a positive impact on health [17,18,19,20,21,22]. However, for adults, health measured by weight (or BMI) has unique characteristics; at lower BMI levels, an increase in BMI signifies an improvement in health, whereas at higher BMI levels, a decline in BMI suggests an improvement in health. Therefore, as income rises, the impact of income on weight (or BMI) transitions from positive to neutral, ultimately becoming negative [23] and forming an inverted U-shaped relationship between income and obesity. Accordingly, we proposed the following hypothesis:

H1

There is an obesity Kuznets curve where the impact of income on obesity demonstrates an inverted U-shaped relationship. Specifically, as income grew, residents鈥� BMI initially increased. When income reaches a certain level, its impact on BMI becomes negative.

According to consumption theory, individual consumption is constrained by income. With rising income, individuals鈥� dietary behaviours and non-medical healthcare costs may change, thereby affecting their BMI.

At lower income levels, a rise in income leads to a more relaxed budget, enabling people to afford more nutritious food. They may consume more high-fat foods and high-protein meat products such as fried snacks, meat, fish, and pickled foods, resulting in a nutritional transition in their diet [5]. However, an increase in income also provides residents with more opportunities for dining out and experiencing richer flavours, and food consumed outside the home may contain more oil, salt, calories, and be of lower quality [24, 25], potentially leading residents to consume more fat and energy [22, 26]. Thus, at lower-income levels, a rise in income may lead residents to consume excess energy, resulting in an increase in BMI. With a rise in income, the opportunity cost of health grows, prompting residents to pay more attention to health issues, raise their health consciousness [27, 28], and increase monetary and time expenditure on physical exercise. For example, they may invest in fitness equipment and health products and engage in physical exercise. Importantly, at lower income levels, owing to limited health knowledge, the impact on body weight may be limited. Hence, diet plays a primary role in the increase in BMI with rising income.

In the high-income stage, after income reaches a certain level, residents may purchase and consume high-quality healthy foods [29, 30]. This could be because increased income provides residents with more access to health and nutritional knowledge, raising the opportunity cost of unhealthy choices. As such, residents adjust their dietary structure, reducing the consumption of high-fat and fried, processed unhealthy foods, decreasing the frequency of dining out, reducing overall calorie intake, and consequently lowering their BMI [31]. Additionally, increasing health investments such as healthcare costs and exercise by residents may also help to lower their BMI.

Therefore, we formulated the following hypothesis:

H2

The inverted U-shaped impact of residents鈥� income on obesity may be achieved by altering dietary behaviours and increasing health investments.

Furthermore, different population groups exhibit variations in health literacy, social pressures, and external environments, leading to potential differences in how income affects obesity. For instance, urban and rural residents have different living environments and dietary habits, resulting in variations in how income influences obesity in urban and rural areas [32]. Age groups may differ in terms of health literacy, resource access, and utilisation. Younger generations may possess more comprehensive health literacy and use advanced resources and technology to avoid obesity [33]. Due to early life experiences with hunger, elderly individuals may consume more nutrients in their later years, leading to a greater likelihood of obesity [34]. Education level can affect individual health awareness and choices, with highly educated individuals potentially being less affected by income in terms of obesity. Women may face greater social pressure than men and are more likely to face social repercussions for being overweight [11, 35], potentially resulting in gender-based differences in how income influences obesity. Given these factors, we developed the following hypothesis:

H3

The effect of income on obesity may vary among populations based on residence, age, education level, and gender.

Data and research design

Data

We obtained the data from the CFPS, a nationwide, large-scale, interdisciplinary social tracking survey conducted by the China Social Science Survey Centre at Peking University. The survey covers 25 provinces, municipalities, and autonomous regions in China and includes all members of the sampled households. The CFPS administered a formal baseline survey in 2010, gathering information from 14,960 households. Follow-up surveys were carried out nationwide in 2012, 2014, 2016, 2018, and 2020. The CFPS collects information at the individual, household, and community levels, reflecting changes in the Chinese society, economy, population, education, and health. We chose to use CFPS data because they comprise a large sample size and have strong representativeness.

Since many decisions related to body size are made at the household level, we used average household income as a proxy for income. Starting in 2012, the CFPS design team made significant adjustments to the household economic questionnaire, which rendered the income variables in the 2010 survey and post-2012 questionnaires incomparable. As such, we only used data from 2012 onwards. Given the possible impact of the COVID-19 pandemic in 2020 on the estimation outcomes, we ultimately chose to use data from the 2012, 2014, 2016, and 2018 surveys. Although the duration of the data used only encompassed six years, China has undergone enormous socioeconomic shifts in recent years, far exceeding those experienced by other developed nations over the past several decades. Even with just six years of data, we can provide preliminary evidence for the Kuznets curve of obesity.

As underage and elderly populations have differences in childhood experiences and metabolic factors compared to adults, and for convenience in comparing results with other studies focusing on adult health, we selected data for adults between the ages of 18 and 65. Therefore, we excluded pregnant women, students, and participants with missing control variables. Additionally, following a practice established in the literature [36,37,38], we removed a small number of extreme value samples to eliminate their interference and only retained samples with a BMI ranging from 15 to 60. Furthermore, to ensure the accuracy and reliability of the results, we truncated the continuous variable income by 0.5% at the upper and lower tails to mitigate the influence of outliers. In total, we obtained 82,113 samples.

Empirical model

We primarily investigated the impact of income on obesity among residents. Considering the potentially non-linear relationship between income and obesity, we referred to relevant literature and built the following model:

$$\begin{gathered}\:{Y_{it}} = \alpha {\:_0} + \alpha {\:_1}Incom{e_{it}} + \alpha {\:_2}Income_{it}^2 \hfill \\+ \beta {X_{it}} + \gamma {\:_t} + {P_p} + {\varepsilon _{it}} \hfill \\ \end{gathered} $$

(1)

where Y_it represents the obesity level of individual i in period t, Income_it denotes the average household income of individual i in period t, and $\:{\varvec{X}}_{\varvec{i}\varvec{t}}$refers to the other control variables that impact obesity. To enhance the model鈥檚 explanatory power and eliminate the influence of unobserved time and regional characteristics on individual obesity, we included the year dummy variable 纬_t and the province dummy variable P_p in the model. 蔚_it indicates the stochastic error term. If 伪₁ is significantly positive and 伪₂ is significantly negative, this validates H1.

To examine the mechanism through which income affects individual obesity, we adopted the method proposed by Cutler & Lleras-Muney (2010) [39] and replaced the dependent variable in Eq.听(1) with the mechanism variable of interest. We set the model as follows:

$$\begin{gathered}\:Mechanis{m_{it}} = \beta {\:_0} + \beta {\:_1}incom{e_{it}} + \beta {\:_2}income_{it}^2 \hfill \\+ \delta {X_{it}} + \mu {\:_t} + \zeta {\:_p} + \nu {\:_{it}} \hfill \\ \end{gathered} $$

(2)

where Mechanism_it represents the mechanism variable of interest. The other variables are defined as in Eq.听(1).

To verify the inverted U-shaped impact of income on BMI, we initially employed ordinary least squares (OLS) regression. However, potential endogeneity may arise owing to omitted variables and reverse causality [35]. To address the endogeneity of income in the estimation, we employ an instrumental variable (IV) approach. Generally, institutional changes could be used as potential valid instruments [40]. Following Ren et al.(2019) [41], we use the minimum wage as an instrument for income. The minimum wage is expected to be a valid instrument for two reasons. First, it has a significant direct impact on household income, covering both urban and rural workers. Many rural laborers have shifted to off-farm sectors including migration to cities. The data from the Seventh National Census shows that the proportion of China鈥檚 population living in urban areas is 63.89%, and 33.86% of rural labor force are working or residing in cities, and they are generally the main source of income for rural households. Second, the minimum wage can be considered exogenous as local governments have flexibility in setting it, and it varies across counties and over time [42], so it should not directly affect individual health. China first issued the minimum wage regulation in 1993 and updated it in 2004, with features such as mandatory updates every 2 years, coverage of all types of enterprises and workers, and increased penalties [43]. These features make the minimum wage a relevant and valid instrument for income in the health estimation.

Variable setting

The dependent variable

We used BMI as a continuous quantitative measure of obesity. BMI is an internationally recognised indicator of body weight and health and falls within the healthy range [44].The core independent variable.

Income is the key independent variable. Referring to related literature [41], we calculated total household income per capita by dividing total household income by the household population. Total household income includes wages, business, property, transfers, and other income earned over the past year. For the empirical analysis, we adopted per capita income adjusted using the consumer price index for provincial residents as the base in 2012.

The control variables

Referring to pertinent literature [22, 31], the control variables include individual characteristics such as gender, ethnicity, household registration, education level, and marital status. Given the non-linear impact of age on individual BMI, we included age and age squared in the model. Additionally, we included variables related to work status because work type is a chief factor affecting energy consumption, with significant differences in energy spending between physical and mental labour. We controlled for the effect of work activities on obesity by including individual occupational variables. Finally, we included variables related to health status and behaviours in the model such as self-rated health, smoking status, and alcohol consumption.

The mechanism variables

As mentioned earlier, income may affect obesity by changing dietary behaviours and increasing health investments. Owing to limited data availability in terms of dietary behaviour, we focused on the impact of income on the consumption of high-fat and high-protein foods (such as fried snacks, meat, and fish), processed foods, and the costs of dining out over the past week. The specific mechanism variables include the consumption of fried snacks, meat, fish or other aquatic products, processed foods, and the costs of dining out. Regarding health investments, we used non-medical healthcare costs and whether individuals exercised as proxy variables.

Table听1 shows the descriptive statistics of each variable.

Table 1 Meaning and descriptive statistics of the main variables

Results

Baseline regression results

To examine the non-linear effect of income on BMI, we first conducted baseline regression using income, income squared, and BMI. Table听2 presents the results, with columns (1), (2) and (3) reporting the outcomes of the OLS regression. Then, owing to potential endogeneity issues stemming from omitted variables and reverse causality, we employed the minimum wage and its square as IVs for the regression. To ensure the validity of the IVs, we conducted tests for weak instruments and overidentification. The weak instrument tests indicated that the Cragg-Donald F statistic was greater than the critical value for a 10% error in both specifications, thus rejecting the null hypothesis of weak instruments (The detailed results of the first stage can be found in Appendix a). This suggests that the IVs were valid.

Table 2 Benchmark regression results of income鈥檚 impact on BMI

Columns (4), (5) and (6) report the 2SLS regression outcomes. Columns (1) and (4) include only income and income-squared variables, columns (2) and (5) incorporate other individual-level control variables, and columns (3) and (6) add year and province dummy variables to the specifications in (2) and (5). All regression results show that income and income squared are significantly different from zero at the 1% level, with the income coefficient being significantly positive and the income squared coefficient being significantly negative. This implies that as income rises, BMI initially increases but then declines after income reaching a certain level, denoting an inverted, U-shaped relationship between income and BMI. Thus, H1 is supported. Comparing the results from the OLS and IV regressions, we found that the magnitudes of the coefficients for income and income squared obtained from the IV regression were larger in absolute value than those obtained from the OLS regression. Considering Column (6) as an example, the coefficients for income and income squared are 1.514 and 鈭掆��0.162, respectively, whereas the corresponding coefficients from the OLS regression in Zhao and Zheng (2019) [45] are 0.843 and 鈭掆��0.311. This finding suggests that neglecting endogeneity may severely underestimate the impact of income on obesity. Hence, in the subsequent analysis, we report the results of the IV regression.

Based on the coefficients in Column (6), the turning point at which BMI shifts from rising to falling is determined to be an income level of 46,728 yuan (in 2012 prices) or 57,066 yuan in 2023 prices (equivalent to 8,098 dollars) according to the cumulative inflation rate based on China鈥檚 Consumer Price Index. By 2023, the per capita disposable income of Chinese residents reached 39,218 yuan (5,565 dollars in 2023 prices). This implies that at the current average income level, without further strengthening P&C efforts, the BMI of most people in China may continue to increase.

Robustness tests

Changing the dependent variable into a dummy variable

According to standards for determining adult weight, we created two binary variables to indicate whether individuals were obese (obesity鈥�=鈥�1 when BMI was greater than or equal to 28 and obesity鈥�=鈥�0 if otherwise) or overweight / obese (overweight鈥�=鈥�1 when BMI was greater than or equal to 24 and overweight鈥�=鈥�0 if otherwise). We then performed IV probit regression; the results are reported in columns (1) and (2) of Table听3.

Changing the dependent variable into an ordinal multiclass variable

Using an ordinal multiclass variable based on different BMI ranges, we assigned values to the dependent variable such that a BMI鈥�<鈥�18.5 took a value of 1 (underweight), a BMI鈥夆墺鈥�18.5 and BMI鈥�<鈥�24 took a value of 2 (normal weight), a BMI鈥夆墺鈥�24 and a BMI鈥�<鈥�28 took a value of 3 (overweight), and w BMI鈥夆墺鈥�28 took a value of 4 (obese). We then estimated the ordinal variable using the IV 2SLS regression; the results are shown in Column (3) of Table听3.

Generalised method of moments (GMM)

Considering the possibility of heteroscedasticity or autocorrelation in the error term, we employed the generalised method of moments (GMM) estimation to conduct IV regression. From Table听3, we can see from that regardless of whether the dependent variable is obesity, overweight / obesity, the ordinal variable, or estimated using GMM estimation, the coefficients for income and income squared are significantly positive and significantly negative, respectively, indicating an inverted U-shaped relationship between income and obesity rates, overweight/obesity rates, and BMI. This outcome confirms the robustness of the results.

U-test method

In empirical studies, when identifying U-shaped or inverted U-shaped relationships, relying solely on the significance of the quadratic term coefficient and the estimated inflection point within the data range may lead to a misjudgement [46]. This is because, when the actual relationship is convex and monotonous, the model estimation may incorrectly produce an inflection point and a U-shaped or inverted U-shaped relationship. To avoid this misjudgement, Lind and Mehlum developed the U-test command based on the framework proposed by Sasabuchi (1980) [47] to accurately test for a U-shaped or inverted U-shaped relationship between the two variables. We used this command to examine the inverted U-shaped relationship between income and obesity. The calculated inflection income point was 46,630 yuan, which fell within the data range, and the null hypothesis of monotonic or U-shaped was strongly rejected at the 1% level (The detailed results can be found in Appendix b). Furthermore, the slopes in the interval were negative, indicating that the effect of income on obesity had an inverted U-shape.

Table 3 Robustness tests

Replace the minimum wage with a city-level one

Given the large number of migrant workers in rural areas of China, the relatively low minimum wage standards in some regions may not fully account for the migration-related issues. Therefore, we use the city-level minimum wage standard as an instrumental variable to alleviate this problem. As shown in column (1) of Table听4, the coefficients of income and income squared are still significantly positive and significantly negative, respectively, indicating an inverted U-shaped relationship between income and BMI.

Replace the income with wage

To further conduct robustness checks, we use wage income as the explanatory variable. As shown in column (2) of Table听4, the coefficients of wage income and its squared term are significantly positive and significantly negative, respectively, indicating an inverted U-shaped relationship between wage income and BMI.

Replace the income with equivalent income defined by OECD

To fully consider the economies of scale within households, we use the OECD鈥檚 equivalent income standard [48] as a substitute for income. As shown in column (3) of Table听4, the coefficients of the equivalent income standard and its squared term are significantly positive and significantly negative, respectively, indicating an inverted U-shaped relationship between wage income and BMI. This result confirms the robustness of the findings.

Table 4 Robustness tests

Control individual fixed effects

To enhance the robustness and interpretability of the research results, learn from literature [41], we control for individual fixed effects. As shown in column (4) of Table听4, after controlling for individual fixed effects, the coefficients of income and income squared are still significantly positive and significantly negative, respectively, indicating an inverted U-shaped relationship between income and BMI, confirming the robustness of the results.

Mechanism analysis

The baseline results show that income had an inverted U-shaped impact on BMI; however, what are the mechanisms through which income affects obesity rates? As analysed earlier, income increases may change individuals鈥� dietary behaviours and increase health investments, thereby influencing obesity. Next, we explored the mechanisms by which income impacts obesity.

Impact of income on dietary behaviour

We assessed dietary behaviour using five indicators in terms of whether individuals consumed (1) fried snacks, (2) meat, (3) fish or other aquatic products, and/or (4) processed foods, and (5) dining out costs over the past week. Table听5 presents the regression outcomes for the impact of income on dietary behaviour using IV regression. The coefficients of income and income squared are significantly positive and negative, respectively, at the 1% level. This implies that, as income rises, the probability of consuming fried snacks, meat, fish or other aquatic products, processed foods, and dining out initially increases and then declines. In other words, as income rises, individuals can afford to consume more high-salt, high-fat, high-protein fried foods. This leads to excessive nutritional intake and an increased BMI. However, after reaching a certain income level, enhanced health consciousness may lower the probability of consuming high-fat, high-salt, and unhealthy foods, resulting in decreased nutritional intake and BMI. We observed similar results for dining out costs, where an increase in income was associated with higher expenditures. Dining out at restaurants and hotels usually involves heavier use of oil and salt, which increases the risk of obesity [49]. However, as income surpasses a certain level, people may become more focused on healthy eating and social quality, resulting in a reduced frequency of dining out and leading to a decrease in BMI.

Table 5 Mechanism analysis: eating style

The impact of income on health investments

Next, we examined the impact of income on health investments. Table听6 presents the results of the impact of income on non-medical healthcare costs and exercise. As income rises, both per capita non-medical healthcare costs and the probability of engaging in exercise increase. This indicates that as income rises, individuals are more likely to allocate resources to health-related expenses, such as purchasing fitness equipment and health supplements and engaging in physical exercise. These measures are beneficial for increasing energy expenditure and reducing BMI.

In summary, with a rise in income, individuals鈥� spending on health and the probability of engaging in exercise grow, which can help to improve health and reduce weight. However, in the low-income stage before the turning point, the impact of income on BMI may be limited because of factors such as limited health knowledge and lower exercise intensity. Therefore, the impact of income on dietary behaviour may play a dominant role in obesity, resulting in an initial positive and negative relationship between income and BMI. This supports H2.

Table 6 Mechanism analysis: health investments

Heterogeneity analysis

Previous estimates have demonstrated the average impact of income on BMI in adults aged 18鈥�65. Next, we conducted heterogeneous studies focusing on different genders, residential areas, ages, and education levels.

Heterogeneity analysis based on gender and residential area

The results in columns (1) and (2) of Table听7 show that the impact of income on BMI for individuals of different genders follows a pattern of first being positive and then negative and is significant at the 1% level. Although there are slight differences in the coefficients of income and income squared between male and female subgroups, the empirical p-values used to test for intergroup coefficient differences indicate the rejection of significant disparities. This contrasts with the findings of Li et al.(2021) [31]) and Kong & Qi (2017) [32]. While women face greater social pressure and are more likely to face social punishment for being overweight [35], this influence may apply primarily to urban areas with less pressure on rural women, particularly older female residents.

Table 7 Heterogeneity analysis: gender and residence

The outcomes in columns (3) and (4) of Table听7 show that the impact of income on BMI is first positive and then negative for rural and urban residents. Intergroup coefficient difference tests revealed significant differences in the sizes of the coefficients between the two groups. Figure听1 illustrates the overall impact of income on BMI and its effects on the residents of different residential areas. In Fig.听1, we can see that the inverted U-shaped impact of income on BMI is steeper for rural residents (the red curve) compared to the flatter curve for urban residents (the green curve). This aligns with new trends in urban and rural obesity over the past decade [50]. These differences could be related to overall education levels, sports infrastructure, and dietary beliefs in different regions. Urban residents generally have higher education levels and greater health awareness, coupled with better sports facilities, making them more likely to control their diet, exercise, and maintain a healthy weight. In contrast, rural residents are transitioning from traditional diets rich in carbohydrates and fibre to high-fat, low-fibre diets [51, 52]. Moreover, there is a lack of awareness regarding the dangers of obesity, with a prevalent belief that 鈥榚ating well is a blessing, and obesity signifies good nutrition鈥� [53]. Hence, an increase in income is more likely to result in rural residents experiencing excessive nutrition, which exacerbates obesity.

Fig. 1

The difference in residence for the effect of income on BMI. Note: In order to visually reflect the impact of income on BMI, we did not consider other variables and only plotted the graph using the coefficients of income and income squared in the regression

Based on the calculated results, the inflection point for rural residents鈥� BMI transitioning from an increase to a decrease in response to income was 43,144 yuan (the red vertical line, 52,795 yuan in 2023 price, equivalent to 7,492 dollars), whereas for urban residents, the inflection point was 50,714 yuan (the green vertical line, 61,987 yuan in 2023 price, equivalent to 8,797 dollars). In 2023, the average disposable income per capita for rural residents was 21,691 yuan (equivalent to 3,078 dollars in 2023 price) and that of urban residents was 51,821 yuan (equivalent to 7,354 dollars in 2023 price). Although neither urban nor rural areas cross the inflection point, rural residents are farther away from the inflection income level. This suggests that the BMI of rural residents is likely to continue to rise, leading to a gradual increase in overweight and obesity rates.

Heterogeneity analysis based on age

Table听8 reports the analysis results of income on BMI across different age groups. We divided the population into three subgroups: 18鈥�29 years, 30鈥�49 years, and 50鈥�65 years. Income鈥檚 impact on BMI follows an inverted U-shaped pattern across all three age groups. However, the results of the intergroup difference tests indicate significant differences in the coefficients. The inverted U-shaped impact of income on BMI is more pronounced in the 50鈥�65 age group. This can be explained by the thrifty phenotype hypothesis, which suggests that an individual鈥檚 health partly depends on his/her nutritional status during childhood [54] and the connection between early malnutrition and unhealthy outcomes in adulthood may be owing to gene activation [55, 56]. Individuals born during a period of resource scarcity, such as those in the 50鈥�65 age group, may be more likely to activate certain 鈥榯hrifty鈥� genes. If they later live in an environment of abundance, they may be more prone to obesity due to overeating in their later years. Figure听2 illustrates the relationship between income and BMI among residents in different age groups. For residents aged 50 to 65, the inflection point was 47,979 yuan (58,594 yuan in 2023 price, equivalent to 8,315 dollars), while it was around 4,5,000 yuan (54,956 yuan in 2023 price, equivalent to 7,799 dollars) for the other two groups.

Table 8 Heterogeneity analysis: age

Fig. 2

The difference in age for the effect of income on BMI. Note: In order to visually reflect the impact of income on BMI, we did not consider other variables and only plotted the graph using the coefficients of income and income squared in the regression. The inflection point for residents aged 50鈥�65 years is 47 979 yuan, whereas the inflection points for the other two groups are approximately 45,000 yuan

Heterogeneity analysis based on education level

As for education level, we divided the sample into three subgroups: primary school and below, junior high/middle school and high school, and college and above. The regression results are presented in Table听9. Income had no impact on BMI in the college or above groups. The baseline regression outcomes also indicate that under other constant conditions, residents with a college education or higher had the lowest BMI. This suggests that individuals with higher education levels have strong health awareness, understand the risks of obesity, can control their diet, and engage in moderate exercise; thus, income variations do not significantly affect their weight. Income had a significant impact on the BMI of individuals with a high school education or below, following an inverted U-shaped pattern. The shape was steeper for the lower-educated (primary school and below) groups and relatively flat for the middle and high school groups. Fisher鈥檚 combined test confirmed that the differences between the two groups were significant, implying that the impact of income on BMI was concentrated mainly in the lower-education group. Figure听3 illustrates the relationship between income and BMI among residents with a primary school education and below as well as those with a middle and high school education. The inflection points for the two groups are different, with an inflection point of 39,868 yuan (48,688 yuan in 2023 price, equivalent to 6,909 dollars) for residents with a primary school education and below, and 43,273 yuan (52,847 yuan in 2023 price, equivalent to 7,499 dollars) for residents with a middle and high school education. This suggests that individuals with higher education levels reach their peak BMI at higher income levels.

Table 9 Heterogeneity analysis: education level

Fig. 3

The difference in education level for the effect of income on BMI. Note: In order to visually reflect the impact of income on BMI, we did not consider other variables and only plotted the graph using the coefficients of income and income squared in the regression. Income had no significant impact on the BMI of residents with a university degree or higher, which is not shown

In summary, these analyses demonstrate that income has varying effects on BMI across the genders, regions, age groups, and education levels.

Conclusion and policy implications

We utilised data from the CFPS of 2012, 2014, 2016, and 2018 to empirically analyse the impact of income on residents鈥� obesity using an IV approach. We explored the potential mechanisms and heterogeneous impacts of income on obesity across different population groups. The effect of income on residents鈥� BMI followed an inverted U-shaped pattern where at lower income levels, a rise in income led to a significant increase in BMI. However, when per capita income exceeded 46,728 yuan (57,066 yuan in 2023 prices, equivalent to 8,098 dollars), further income growth caused BMI to decline. The mechanism analysis suggests that this non-linear impact of income on obesity may be related to dietary behaviour and health investments. Specifically, at lower income levels, increased income primarily increased the consumption of fish, meat, fried foods, and pickled foods, leading to a significant rise in BMI. Conversely, at higher income levels, residents reduced their consumption of high-fat meat and focused more on healthy diets and exercise, resulting in a decline in BMI, reflecting an inverted U-shaped impact of income on obesity. Heterogeneity analysis revealed that income had a more significant impact on obesity among rural residents, individuals aged 50鈥�65, and those with lower education levels.

These findings provide insights for policy considerations. The national average disposable income of Chinese residents has not yet reached a turning point, indicating a continuous expansion of the obese population. As such, we developed the following policy recommendations:

First, various obesity P&C strategies should be implemented. Given the inverted U-shaped relationship between income and obesity, policymakers should adopt tailored P&C measures for different income groups. For lower-income groups, emphasis should be placed on raising awareness about healthy diets, promoting balanced nutrition, and avoiding excessive consumption of high-fat, high-energy foods. For high-income groups, the focus should be on encouraging healthy lifestyles as well as increased physical exercise and health investment.

Second, efforts to prevent and control obesity in rural areas should be strengthened. This study indicates that income has a greater impact on obesity among rural residents, necessitating heightened efforts for rural obesity P&C. Measures include improving public sports facilities in rural areas, conducting targeted health education campaigns, enhancing residents鈥� health awareness and self-management capabilities, and promoting the development of rural health industries to provide additional health service options for rural residents.

Third, we addressed obesity issues among the elderly and individuals with lower education levels. This study highlights those individuals aged 50 years and older as well as those with a primary education and below are more susceptible to the obesity effects of rising income. Specific health education and intervention plans should be formulated for these patients. Strategies include offering community health lectures, establishing senior citizen universities, promoting health knowledge, offering education/knowledge about scientific diets, and encouraging moderate exercise. Additionally, enhancing health monitoring and follow-up for these groups can help to promptly identify and intervene in obesity-related issues.

Finally, a healthy dietary culture should be promoted, and rational consumption should be guided. Income increases can impact residents鈥� dietary behaviours, particularly at lower income levels where the consumption of high-fat, high-energy foods may increase. Thus, a healthy dietary culture should be cultivated through diverse channels such as media campaigns, school education, advocacy for balanced nutrition, and concepts of moderate food intake. It is important to encourage the food sector to develop food options that provide consumers with healthier choices. In the catering industry, implementing nutrition labelling systems can assist consumers in making informed dietary choices.

This study has certain limitations. For the data, we relied on self-reported height and weight information, which may have contained measurement errors [57]. Future research could use measured or more accurate data to minimise errors [35, 58]. Moreover, owing to data availability, we explored only the mechanisms of dietary patterns and health investments, future studies should investigate other potential mechanisms.