Poverty and Subjective Poverty in Rural China

The most prevalent measurement of poverty is to set an objective absolute poverty line by the government, scholars or some organizations. Traditionally, absolute poverty measurement is based on a comparison of resources to needs. Thus, a family is identified as the poor if its resources short of the poverty threshold (Foster 1998). For instance, the most recent global poverty line recommended by the World Bank is US $ 1.90 expenditure per day per person (2011 purchasing power parity (PPP) price). Through the purchasing power parity (PPP), the global poverty line can be compared between different countries. However, the concept of the global poverty line has been strongly criticized by Deaton (2010). First, the poverty line is set by some experts which do not capture full information of the poor; Second, the prices collected by the International Comparison Program (ICP) are national average prices, which are different from those the poor face, as the expenditure patterns of the poor often differ the aggregate patterns; Third, each country (region) has different consumption patterns due to different food, culture, and traditions (Deaton 2010; Deaton and Dupriez 2011; Kim et al. 2018).

In addition to the global poverty line, many countries and areas are prone to set a national poverty line (an income or a consumption poverty line) based on their economic and social reality. The poverty lines are often adjusted according to economic development levels. Likewise, this type of traditional poverty line, mainly measuring absolute poverty, is very sensitive to different research designs. On the one hand, the household surveys, which are the main information sources for the poverty line setting, often cannot obtain accurate income or consumption information from the surveyed families. One typical example is the national sample survey of India in 1998. The Indian government replaced the traditional survey of 30-days food consumption with one of 7-days food consumption, which resulted in a sharp increase of food consumption expenditure per month. Due to the survey method change, the poor population in India reduced by nearly 175 million. Besides, according to Deaton (2001), the estimated deviation would be greater if we use the standard nutritional approach in which the poverty line is calculated by the costs of minimum nutrition requirement. Deaton and Drèze (2009) discussed the India poverty line by estimating the demand for calories, and the result showed that a family would prefer not to undertake heavy-labor work when the economic condition of the family improves, and hence the demand for calories declines. Thus, if the poverty line is based on calories demand, the incidence of poverty, on the contrary, would increase. On the other hand, if there are a lot of poor people live near the poverty line, a slight change in the poverty line would incur a great impact on the heads counting of the poor (Deaton and Heston 2010; Ravallion 2015). In addition, the cross-sectional survey data entail statistical errors. For instance, there is always a difference between the household’s consumption population and the survey population (Zhou and Yu 2014; Yu and Abler 2016). In most cases, the survey population is greater than the consumption population, and it leads to the underestimation of food consumption per capita.

Besides, the relative poverty line is also one of the most important objective poverty lines, which is widely used in developed countries. Compared with the absolute poverty line, the relative poverty line mainly focuses on the people who have some money but still no enough money to afford anything above the basics. Conventionally, it is useful for showing the percentage of the population who has been relatively left behind.

Apart from the several poverty measurements mentioned above, some literature focuses on the multidimensional poverty based on the “capability poverty” theory of Sen (1996) and presents a multidimensional poverty index that included education, health, living conditions, and so on (Alkire and Foster 2011; Alkire and Seth 2015). In practice, the multidimensional poverty index (MPI) is also widely used for measuring economic development levels globally, and the Human Development Index of the United Nations Development Programme (UNDP) is a typical application.

Income (consumption) or multidimensional poverty measurements belong to the category of objective poverty and are often measured by an objective poverty line threshold set by politicians, scholars or other authorized organizations. However, these objective poverty measures often ignore individual heterogeneities of wellbeing, result in large deviations in poverty headcounts, and have been subject to much criticism due to the aforementioned reasons.

Townsend (1979) further pointed out that the fundamental flaw of objective poverty is that it is difficult to scientifically define the non-material needs. Van Praag (1968) suggested that objective poverty is a patriarchal style measure, as the poverty line is decided by bureaucrats or experts, ignoring the real perception of the poor. Different people have a different understanding of poverty. Consequently, some who are not objective poverty may feel poor, while some who are objective poverty may not feel poor (Mahmood et al. 2018). Meanwhile, the information, provided from objective poverty, is very limited for policymakers particularly in an affluent society, which requires additional subjective information from the polls (Veenhoven 2002; Klasen et al. 2016). Hence, Deaton (2010) directly suggested that “why don’t we just ask people?”, since the people themselves have a very good idea of whether or not they are poor.

There is a call for subjective poverty. The meaning of utility, a basic concept in economics, is defined as the subjective perception of self-welfare, but this is largely neglected in many studies, particularly in poverty measurement. Thus, some studies started to shed light on subjective poverty. They believe that social individuals possess the most fruitful information for themselves so that they can be the best persons who can judge whether they are in a poverty status or not. Combined with several drawbacks of objective poverty, the subjective poverty concept is beneficial to poverty identification and policy design (Ravallion and Lokshin 2002; Deaton 2010; Allen 2017; Zhou and Yu 2017; Deaton 2018). For example, Pradhan and Ravallion (2000) used the satisfaction of consumption to measure the subjective poverty status; Mahmood et al. (2018) compared the subjective poverty and objective poverty of Pakistan and found that the objective poverty cannot fully reflect the subjective poverty.

Empirically, subjective poverty is mainly identified through questionnaire surveys, evaluating social individuals for their welfare condition and minimum needs. According to the different identifications, there are three main types of questionnaire settings: namely, Income Evaluation Question (IEQ) (Van Praag 1968), Minimum Income Question (MIQ) (Goedhart et al. 1977) and Centre for Social Policy Question (CSP)¹ (Deleeck and Van den Bosch 1992). Early applications of the subjective poverty questions are mainly conducted in some Western industrial countries and regions. Van Praag et al. (1982) analyzed the subjective poverty line for 8 EU countries by the Income Evaluation Question (IEQ), and they found that the subjective poverty line of city residents was relatively higher than others. Danziger et al. (1984) and Colasanto et al. (1984) used the Minimum Income Question (MIQ) to study the subjective poverty of the United States, and the finding showed that the subjective poverty line was higher than the objective poverty line set by the government. Based on the Minimum Income Question (MIQ), Garner and Short (2005) proposed the Minimum Spending Question (MSQ) to study the subjective poverty of the United States with use of the data of Survey of Income and Program Participation (SIPP) and suggested that the subjective poverty line based on MIQ was higher than MSQ.

However, the subjective poverty in rural China has not been well studied though it is increasingly important from the policy perspective. There are a few exceptions or related studies in urban China. The earliest study is conducted by Gustafsson et al. (2004), who used the Minimum Income Question (MIQ) to investigate the subjective poverty in urban China and found that the subjective poverty line was close to the objective poverty line set by the Chinese government. Bishop et al. (2006) also used the Minimum Income Question (MIQ) to study the subjective poverty of different areas of China based on the data of the Chinese Household Income Project (CHIP). However, CHIP does not directly contain the Minimum Income Questions, and the MIQ used in their study was estimated through alternative indicators. Zuo and Yang (2013) discussed the implications of subjective poverty measurement for anti-poverty policy in China theoretically, suggesting that subjective poverty possesses both instrumental value and intrinsic value. However, there are no direct studies for subjective poverty in rural China, though more than 90% of the poor live in rural areas in China.

Currently, the poverty counting in rural China is still mainly based on an objective poverty line: the national poverty line. The current national poverty line is 3000 yuan net income per year (equals US $2.3 per day²) set in 2016. The Chinese government is campaigning for eliminating absolute poverty by 2020. Beyond then, the subjective poverty measurement would be a better way to reflect the poverty perception of people in rural China. Therefore, in order to fill the gap in the research of subjective poverty, this study adopts the method of MIQ to measure subjective poverty in rural China and attempts to provide implications for poverty-reduction policies for China beyond 2020.

The data used in this paper is a nationally representative survey of 2025 rural households in five provinces of China (Jiangsu, Sichuan, Shannxi, Jilin, Hebei) in 2016, namely China Rural Development Survey (CRDS), collected by the Center for Chinese Agricultural Policy, Chinese Academy of Science. The sample was selected as the following steps. First, we selected five provinces to represent the five major agro-ecological zones in China: Jiangsu is a representative sample province of the eastern coastal region; Sichuan is a sample of the south-west region; Shaanxi is a sample of the north-west region; Jilin is a sample of the north-east region; Hebei is a sample of the central region. Second, according to the per capita gross value of industrial output (GVIO),³ we divided all counties into five groups for each province, and then randomly selected one from each group. Following this procedure, we randomly selected two towns from each county, and two villages from each town, and then selected 20 sample households from each village. Finally, we collected a nationally representative sample of 2025⁴ households’ information for the year of 2015.

How to measure subjective poverty is a key issue in this study. As aforementioned, there are three main methods to identify the subjective poverty for social individuals: IEQ, MIQ, CPS. Compared with IEQ and CPS methods, MIQ is more easily for the respondents to understand and more feasible in the survey. Due to these advantages, Minimum Income Question (MIQ) is widely used in the subject poverty research all over the world (e.g. Van Praag et al. 1982; Gustafsson et al. 2004; Bishop et al. 2006). Thus, this study adopts the Minimum Income Question to identify the subjective poverty in rural China. The survey question reads: “Please offer an income amount below which you will feel poor for a household as yours”. We believe such a number offered by the respondents entails all information about their individual living condition, subjective well-being, and regional development level. This is a typical MIQ for estimating a subjective poverty line for this household.

In order to control for the effect of family size, the subjective poverty line per capita is computed by dividing the self-reported minimum income by the family size. Based on the mean value of individual subjective poverty standards, the subjective poverty status for each rural household can be identified: 1 if the per capita real income surpasses the subjective poverty standard and 0 otherwise. Besides, we could also define the depth of subjective poverty, which is computed by dividing the gap between the subjective poverty standard and per capita real income by the subjective poverty standard.⁵ In addition, to reveal the difference between the subjective poverty and objective poverty, this study measures the objective poverty by two means: the national poverty line (2800 yuan per year in 2015)⁶ and the global poverty line (the US $1.90 per day⁷).

The explanatory variables mainly include the demographic and socioeconomic characteristics of the rural household (characteristics of the household head, characteristics of the household, human capital of the household, material capital of household, social capital and major irregular expenditure of the household). Specifically, (1) the characteristics of a household head includes head’s age, head’s gender, marital status, head’s education level, whether the head is a village leader, and whether the head is a party member; (2) the characteristics of a household include per capita income, family size, number of elders, number of children and number of labor forces; (3) the human capital includes average health condition of family members, and average education level of family members; (4) the material capital includes land size, house value, productive asset value, and consumption asset; (5) the social capital is measured by two questions: “how many friends or relatives working in the government” and “how many friends or relatives working as managers in the enterprise”; (6) the major irregular expenditure of a household includes education expenditure, medical expenditure, gift expenditure, and wedding expenditure. Table 1 presents the definitions of all variables involved in this study.

Table 2 shows the descriptive statistics of the subjective poverty, the objective poverty, and other information of the rural household. It is clear that the prevalence rate of subjective poverty is 0.44, while the rates of objective poverty are 0.22 (National objective poverty line) and 0.20 (Global objective poverty line), respectively. It indicates that the objective poverty measurement cannot reflect the subjective poverty comprehensively in rural China. It highlights the importance of a study on subjective poverty in rural China.

Looking at characteristics of the household head, the average age of the household heads is 57.84 years old; 88% of the heads are male and married; Their average school years is only 6.84⁸; 16% of them are party member; only 8% are village leaders. This shows a general picture of the demography in rural China, and aging is a problem facing rural China.

Regarding the characteristics of the household, the average family size is 4.14. Both the average numbers of elders and children in a household are 0.64, which shows that 36% are dependent population. The average number of labor forces is approximately 2.66, but the average school year is only 6.73, slightly lower than the number of household head.

The house value is the most precious asset in rural China, while the medical expenditure and the gift expenditure are the two largest irregular expenditures. As for the social capital, there are more friends or relatives working in government organizations (1.02) than acting as managers in enterprises (0.47).

The descriptive statistics clearly show some differences between subjective poverty and objective poverty. Table 3 reveals the subjective poverty lines in rural China which are compared with national and global poverty lines. The average subjective poverty standard for rural households is 8297 yuan, which can be used as a nationally representative Subjective Poverty Line, much higher than the national (objective) poverty line (2800 yuan) and global (objective) poverty line (2563 yuan) respectively. Specifically, the subjective poverty line is about 2.96 and 3.23 times of the national and global poverty lines respectively. It implies that the objective poverty line in China and the World Bank cannot well mirror the welfare level of rural households in China, given continuously high economic growth rates in the past 40 years.

Table 3 also shows the subjective poverty standards for five sample provinces. We have a similar finding that the subjective poverty standards for all five provinces are all higher than the objective poverty lines. The gap of Jiangsu province is the largest though the differences within the five provinces are not substantial, all-around three times. It is possible that the Jiangsu province is the richest region within the five provinces.

Besides, we also compared the subjective poverty standard with real per capita net income for each province and the nation as well. The result shows that the average subjective poverty standard is lower than the average per capita income, about 76% of the average real income. It shows that the gap between the subjective poverty line and the real net income is correlated with the income level again. The higher the income is, the larger the gap is. The largest gaps are from Shaanxi and Jiangsu, with relatively higher income.

Once we have a nationally representative Subjective Poverty Line, we can use it to count the subjective incidents. Table 4 compares the incidences of subjective poverty and objective poverty and their difference in rural China as well. Overall, the prevalence rate of subjective poverty is 0.44, which means that almost half of rural households feel that they are in subjective poverty status. However, the prevalence rate of objective poverty are only 0.22 based on the national poverty line and 0.20 based on the global poverty line, respectively. It is clear that the prevalence of subjective poverty is twice as much as objective poverty with the national poverty line, indicating that even if China eliminates the absolute poverty by 2020, there are still plenty of rural households who fall into the subjective poverty.

After analyzing the subjective poverty for five provinces, we can draw a similar conclusion: there are more rural households in the status of subjective poverty than in the objective poverty. The gaps between the prevalence of subjective poverty and objective poverty for Jilin province are the largest (0.29 and 0.32, respectively for the national and global poverty lines).

Table 5 specifically sheds light on the comparison between the poor households differently measured by the subjective and objective poverty standards. This would offer us a deeper understanding of the subjective poverty in rural China. As Table 5 shows, 449 rural households fall into the objective poverty based on the national poverty line, and 1576 households are non-poor. Among the 449 objective poor, 434 households also feel subjectively poor, about 96% of the objectively poor, which means that nearly all the objective poor fall into the subjective poor with a few exceptions. As the national poverty line is relatively low, it is comprehensible that these extreme poor subjectively feel they are poor as well. In this case, the target of the objective poor is also an ideal way to reduce the subjective poverty, particularly for the extremely poor people.

Nevertheless, 467 rural households who are not identified as the objective poor fall into the subjective poverty status, sharing 29% of the non-poor. Similar with Mahmood et al. (2018), it reveals a reality that plenty of rural households who feel subjectively poor are not identified as the poor with the national poverty line. Thus, if we only focus on the objective poor, we ignore a large part of the subjective poor. A similar conclusion can be drawn from the objective poverty line measured by the global poverty line. Specifically, 97% of the objective poor who fall into the subjective poverty status while 31% of the non-objective-poor feel that they are in the subjective poverty status. Such a finding once again highlights the importance of the concept of subjective poverty for poverty reduction in rural China beyond 2020.

Subjective poverty is measured by a subjective statement so that each individual has a different understanding of the subjective poverty line. Table 5 also compares the subjective poverty standards for the objective poor and non-poor. The subjective poverty standard for the objective poor is 7876.87 yuan, while the standard for the objective non-poor is 8416.92 yuan when we use the national poverty line. Clearly, the objective non-poor have a higher subjective poverty standard. A similar result is found when we use the global poverty line.

Furthermore, there is an asymmetry between subjective and objective poverty. In our sample, 449 households are identified as the objective poor according to the national poverty line. Within these 449 households, 369 (or 82%) confirmed that they were also the subjective poor, and the rest 80 (18%) thought they did not belong to the subjective poor. A similar result could be yielded even we use the global poverty line. There are some poor who are even below the national/global poverty line, but they do not think they are poor. It is possible that the poverty could be transitionary, or they live a simple life.

In order to further investigate the characteristics of the subjective poor, Table 6 presents the comparison of the demographic and socioeconomic characteristics between the subjective poor and non-poor. From the last column of Table 6, we can find that there are some significant differences in the demographic and socioeconomic characteristics between the subjective poor and non-poor. In particular, for the subjective non-poor, household head’s age is younger and their education level is higher than the subjective poor. In terms of the characteristics of the household, per capita income of the subjective non-poor is higher than the subjective poor. Moreover, there are more family members, children, and labor forces in the subjectively non-poor households, while the (subjective) poor households have more elders to support. The human capital condition of the subjective non-poor is better than the subjective poor, having more healthy members and higher education experience. The subjective non-poor households also have a higher value of the house, productive asset, consumption asset, and more land size than the subjective poor. Regarding the major irregular expenditure, the subjectively non-poor households spend more money on the gift and wedding expenditures, while the subjectively poor households spend more money on the medical expenditure. It shows that medical insurance is not well established in rural China, and the poor suffer from medical expenditure due to extremely serious diseases. Finally, we do not find a significant difference between the subjective non-poor and poor households in the aspect of social capital.

In the next section, we are going to exercise econometric models to study the determinants of subjective poverty in rural China.

In the previous sections, we designed a questionnaire to reveal each household’s subjective poverty line and use the average value as the representative subjective poverty line for the nation to identify the subjective poverty status for each household. In order to study the determinants of subjective poverty in rural China, econometric models are specified as follows:

In Eqs. (1), (2), and (3), subscript \(i\) denotes the \(i\) th household. Specifically, Eq. (1) is to study the determinants of subjective poverty standard, and \({\text{SS}}\) denotes the subjective poverty standard reported by each rural household; Eq. (2) is to study the determinants of subjective poverty status, and \({\text{SP}}\) denotes the subjective poverty status of the rural household measured by whether the real income is higher than the nationally representative Subjective Poverty Line or not (0-higher, 1-lower); And Eq. (3) is to study the determinants of the depth of subjective poverty, and \({\text{SD}}\) denotes the depth of subjective poverty of the rural household.

In terms of the independent variables, \({\text{HD}}\), \({\text{HH}}\), \({\text{HC}}\), \({\text{MC}}\), \({\text{SC}}\), and \({\text{ME}}\) respectively stand for the characteristics of household head, the characteristics of household, human capital of the household, material capital, household social capital, and major irregular expenditures (including education expenditure, medical expenditure, gift expenditure, and wedding expenditure). The explanations for all the related variables are reported in Table 1. The terms \(\upvarepsilon\) are error terms following normal distributions with zero mean.

As SS (Subjective Poverty Standard) and SD (Subjective Poverty Depth) are continuous variables, OLS can be used for estimating Eqs. (1) and (3). In contrast, for Eq. (2), \({\text{SP}}\) is a dummy variable that denotes whether the rural household is subjectively poor (\({\text{SP}} = 1\) if the rural household is subjectively poor; \({\text{SP}} = 0\) if the household is not subjectively poor) with use of the subjective poverty line. Thus, a Probit model is used in the estimation of Eq. (2). Besides, as the data used in this study is cross-sectional, we report the robust standard errors to remedy the heteroscedasticity problem.

We also used the variance inflation factor (VIF) to check multicollinearity, and find that it is not an issue in the regressions (Mason et al. 1989).

Table 7 shows the results of the determinants of subjective poverty standard. Model 1 is the estimation of Eq. (1), indicating that the head's age, per capita income, family size, human capital, material capital, and major irregular expenditure play significant effects on the subjective poverty standard reported by the rural household. However, the subjective poverty standard might be affected by regional policies, customs, cultures, and geographical environments. To remedy this problem, we add county dummy variables in Model 2 to control for the unobservable regional effect. The two results are very similar, and the following discussion is mainly based on the estimation results from Model 2.

First, the coefficients for the head’s age and the family size within the category of demographic variables are statistically significant at 1%, and respectively are − 0.007 and -0.164. It implies that old household heads and large family size are less likely to feel subjectively poor. Specifically, when the age of a household head increases by one year, the subjective poverty line decreases by 0.7%. It is possible that a young family demands more money, e.g., for building a new house and supporting the education of children. When the household size increases by one member, the subjective poverty line decreases by 16.4%, perhaps due to the consumption of family public goods.

Second, the coefficient for per capita income is 0.022 and statistically significant at 5%, indicating per capita income of a household would increase the subjective poverty standard significantly. It is easy to understand that the more money a rural household has, the higher life quality they pursue. As a result, the subjective poverty standard obviously increases. Much literature points out that per capita income is an important factor that affects the subjective poverty standard (Kingdon and Knight 2006; Posel and Rogan 2014; Reyes-García et al. 2016; Mahmood et al. 2018).

The socioeconomic characteristics also significantly affect the subjective poverty standard. Particularly, the coefficient for the average education level of family members is 0.024 and statistically significant. It can be explained by the fact that a person with more education often has a higher expectation of good life quality. Similarly, the material capital, the house value, and the value of consumption assets also enhance the subjective poverty standard significantly.

For the variables related to major irregular expenditures, the coefficients for almost all variables are positive and significant, specifically including education expenditure, medical expenditure, and gift expenditure. One plausible interpretation is that, due to imperfections of the social security system in rural China, rural households face high education and medical expenditures, and they wish a high income to compensate for these necessary expenditures. As a result, it increases the subjective poverty standard directly. Besides, in the traditional culture of rural China, when one’s friends or relatives celebrate a variety of social events, such as wedding, funeral, childbirth and so on, it is conventional to give cash-gift to express their blessing, which has occupied a large chunk of the rural household income (Chen 2014). Thus, gift expenditure also could increase the subjective poverty standard significantly. In fact, the coefficient for the medical expenditure is 0.021, greater than the other expenditures. Clearly, the medical expenditure plays the most important role in the subjective poverty standard within the category of irregular expenditures.

From what we have discussed above, we can draw a conclusion that, except for the demographic characteristics, such as age, family size, and education, subjective poverty standards are linked to wealthy levels and irregular expenditures of a family. On the one hand, the wealthy level, measured by per capita income and house value, etc. can reflect the life quality, and it is comprehensible that wealthier families are expected to have high subjective poverty lines. On the other hand, irregular expenditures are linked to imperfections of the social security system in rural China (mainly medical and education expenditure), and the traditional culture of cash gift. The government should enhance the coverage of medical insurance, increase the education expenditure, and change the culture of cash gift, to reduce insecurity and enhance life satisfaction for rural households.

We now use the average subjective poverty lines as the representative subjective poverty lines to categorize the subjective poverty status. If the per capita income of the household surpass the representative subjective poverty line, they are categorized as subjectively non-poor households (0 in the Probit model); otherwise, they are subjectively poor households (1 in the Probit model). Table 8 presents the estimation results for Eq. (2) to study the determinants of the subjective poverty status. Similarly, Model 1 does not control for regional effects, while Model 2 does. The estimation results of Model 2 show that the coefficients for head’s age, head’s gender, family size, and per capita income are negative and significant, while the coefficients for human capital, material capital, social capital, and major irregular expenditures are positive and statistically significant.

Specifically, the older the head is, the less chance the household falls into subjective poverty. Male heads seem to be beneficial to the subjective well-being of the rural household. Importantly, the coefficient for per capita income is negative and statistically significant, and it suggests that the growth of per capita income for rural households is an effective way to alleviate the subjective poverty, which is consistent with the findings of the current literature (Stevenson and Wolfers 2013; Mahmood et al. 2018).

Furthermore, rural households with higher education level, more material capital, and more social capital are more likely to fall into the status of subjective poverty. Perhaps these aspects increase their expectation for better life quality. When they did not reach it, they are more likely to feel subjectively poor.

Moreover, regarding the major irregular expenditures of the rural household, only the coefficient for medical expenditure is positive and significant, meaning that medical expenditure worsens the subjective poverty status. Mainly due to the imperfection of medical insurance, many households have to pay a large chunk of medical expenditures particularly for extremely serious diseases by themselves. The results are consistent with the findings in Sect. 5.2.

Finally, this study discusses the determinants of the depth of subjective poverty, specifically for the subsample of the subjective poor. Table 9 shows the estimation results of Eq. (3). Similarly, we report both results without and with control for regional fixed effect, respectively in Model 1 and Model 2. From the estimated result of Model 2, we can find that, for the subjective poor, the coefficient for per capita income is negative and statistically significant, and it implies that the growth of per capita income can alleviate the subjective poverty depth for the subjective poor.

Meanwhile, the depth of subjective poverty decreases as the rural households have more family members or more friends and relatives working in government organizations. Besides, the land size of the households plays a negative effect on the subjective poverty depth, and it indicates that the households with more cultivated land are more likely to alleviate the subjective poverty depth. Agricultural land is a very important asset for rural households, and it could help reduce poverty, particularly the extreme poverty.

On the other hand, the coefficients for head’s gender and the value of the consumption asset are positive and statistically significant, and it suggests that male head and accumulation of the consumption asset would worsen the subjective poverty depth, perhaps males have more expenditure on addicted goods, such as cigarettes and alcohol.

It is particularly important to point out that education and medical expenditures could exacerbate the subjective poverty specifically for the subjective poor. Once again, as we mentioned above, under the context of the imperfect social security system in rural China, education and medical expenditures are two major heavy financial burdens for the rural households, especially for the poor.

In this part, we further check the robustness of empirical results by excluding 5% of the extreme values of subjective poverty standard, and the estimation results are reported in Table 10. Specifically, model 1 is the results of Eq. (1), while model 2 and model 3 are the results for Eqs. (2) and (3), respectively. It is clear that these results are consistent with previous empirical results.