Introduction

Poverty is one of the main problems plaguing most countries worldwide (OECD, 2021). By 2020, China had eliminated absolute poverty. However, similar to other low- and middle-income countries (LMICs) that made significant achievements in combating absolute poverty, China still faces great challenges of relative poverty reduction (Wan et al., 2021). The relatively poor people often face problems in employment, education, medical care, pension, etc., due to resource shortages, poor environment, and lack of individual ability. Essentially, their poverty is multidimensional (Guo et al., 2022). In addition, in LMICs, some deep poverty-stricken or marginal poverty-stricken households have a high risk of returning to poverty due to poor foundations (Échevin, 2013; Wan et al., 2021). Therefore, it is urgently needed to understand the dynamics of multidimensional poverty and its determinants to reduce relative poverty.

Academically, poverty has long been a heated research topic. Thanks to Sen’s innovative contribution to the capacity approach, many poverty researchers have gradually shifted from a single monetary perspective to a multidimensional approach (Alkire and Santos, 2014). A series of approaches were developed to measure multidimensional poverty. Among them, the most widely used measurement follows the Alkire-Foster (A-F) method. Many researchers used cross-sectional data and applied this method to analyse the current situation and influencing factors of multidimensional poverty (United Nations Economic Commission for Europe, 2017).

Another shift in poverty studies is that more and more researchers have moved from static poverty analysis to dynamic poverty analysis (Jenkins, 2011). For those in poverty, it is essential to know whether they are in poverty for most of their lives or only for a brief period. Therefore, from a policy perspective, it is crucial to understand the process of falling into poverty or escaping from poverty, which gives rise to the problem of poverty dynamics. The studies on poverty dynamics mainly focus on two aspects: poverty durations and poverty transitions, including exiting from poverty, falling into poverty, persistent poverty, reentering into poverty, etc. (Addison et al., 2009).

Among the existing research on poverty dynamics, the majority were on income poverty dynamics, and only a very small part were on multidimensional poverty dynamics. Some studies on the latter topic analysed multidimensional poverty dynamics in adjacent years by constructing a Multidimensional Poverty Index (MPI) and poverty transition matrix (Madden, 2022; Guo et al., 2022). Other related studies estimated the persistence of multidimensional poverty by constructing chronic MPI (Apablaza and Yalonetzky, 2016; Alkire et al., 2017). However, the above studies have separated the analysis of durations of multidimensional poverty and its state transition, thus ignoring the cumulative effect that durations may have on poverty dynamics. Thus, there is a lack of studies systematically exploring multidimensional poverty dynamics over the whole period. Additionally, among the existing studies on multidimensional poverty, only a few centred on older adults. And the few such studies were all based on the static analysis of multidimensional poverty (Amarante and Colacce, 2022; Kamal et al., 2022). To our knowledge, no study has been conducted so far on multidimensional poverty dynamics among older adults. According to the global experience of poverty alleviation, this population is more vulnerable to poverty than the general population (OECD, 2021). This study is designed to fill the above knowledge gap. The aim of this study is to understand the dynamics of multidimensional poverty and its determinants among mid-aged and older adults using the China Health and Retirement Longitudinal Study (CHARLS) survey from 2011 to 2018 in a country in the stage of rapid population ageing.

The paper is organised into five sections. Section ‘Methodology’ introduces the methodology, including conceptual framework and hypothesis, data, measurement and variables, and statistical methods. Section ‘Results’ depicts the results. Section ‘Discussions’ discusses the main findings. Section ‘Conclusions’ concludes.

Methodology

Conceptual framework and hypothesis

We formed a conceptual framework to systematically analyse the determinants of multidimensional poverty dynamics (Fig. 1). Based on Schultz’s (1961) human capital theory, individual characteristics such as gender, age, and marital status can affect the accumulation process of human capital, which in turn can affect multidimensional poverty dynamics. According to Rowntree and Bradshaw’s (2000) life cycle of poverty theory, poverty is closely related to significant life events in the family. Therefore, household size, labour status, and living arrangements can affect multidimensional poverty dynamics. Based on Bourdieu’s (1986) work on social capital, the lack of social capital means the lack of the possibility of obtaining development opportunities and social resources, which will lead to poverty. From the perspective of spatial poverty traps (Bird, 2019), different regions have differences in geographic capital in terms of economic development, culture, etc., affecting residents’ poverty status. Based on the literature on poverty dynamics (Devicienti and Gualtieri, 2010; You, 2011), the state of poverty usually changes over time, and poverty durations usually affect the change of its state. According to the above theories and our understanding of China, we postulated the following hypotheses:

H1: In terms of individual characteristics, women, those with higher age, and those not married are more difficult to exit from multidimensional poverty and easier to return to multidimensional poverty among mid-aged and older people.

H2: In terms of family structure and living arrangement, the household size, the number of working population within a household, and living with children have a positive effect on the probability of exiting from multidimensional poverty, but have a negative effect on the probability of returning to multidimensional poverty.

H3: In terms of social capital, economic transfer and social activities have a positive effect on the probability of escaping from multidimensional poverty, but have a negative effect on the probability of returning to multidimensional poverty.

H4: In terms of living areas, the mid-aged and older people in rural or western regions have more difficulties escaping from multidimensional poverty and are more likely to return to multidimensional poverty.

H5: With durations increase, the probability of multidimensional poverty transitions became smaller.

Data

We used data from CHARLS, which representatively collects social, demographic, economic, health, retirement, etc. data among a population aged ≥45 years in China. In 2018, CHARLS collected data on a sample size of 190,000 respondents in 124,000 households. In this study, we used all four waves of data, i.e., 2011, 2013, 2015, and 2018 rounds. We kept those who participated in all four rounds and deleted those below 45 years in 2011. In the end, we included 11,566 respondents in the analysis.

Measurement

Construction of multidimensional poverty index

We utilised the Alkire-Foster (A-F) method to construct the MPI, which sets a double cutoff in poverty identification. First, the deprivation cutoffs were utilised to determine whether the respondent was deprived in each dimension. Second, the deprivation scores of all dimensions were weighted and summed. Third, the poverty cutoffs were applied to determine whether the respondent was multidimensionally poor. Forth, the MPI (or M0) was calculated as the product of headcount ratio H and poverty intensity A (Alkire and Seth, 2015). H, A, M0 are expressed as follows:

$$H = q/n$$
$$A = \mathop {\sum}\limits_{i = 1}^q {c_i\left( k \right)/q}$$
$$M_0 = H \times A = \frac{q}{n} \times \frac{{\mathop {\sum}\nolimits_{i = 1}^q {c_i\left( k \right)} }}{q} = \frac{1}{n}\mathop {\sum}\limits_{i = 1}^q {c_i\left( k \right)}$$

where n is the sample size, q is the number of those who were multidimensionally poor, ci is the deprivation scores of individual i in all dimensions, and k is a poverty cutoff that identifies who is poor. When ci ≥ k, individual i is defined as multidimensionally poor and ci(k) = ci; otherwise ci(k) = 0. M0 has good decomposability, which can be decomposable by subpopulations or indicators (Alkire and Santos, 2014; Kamal et al., 2022).

Dimensions, indicators, and weights of MPI

There is no unified standard for selecting dimensions and indicators of MPI, but most studies chose the dimensions of education, health, and living standards (Alkire and Santos, 2014; Alkire and Seth, 2015). Some related studies using cross-sectional data pointed out that the ageing of physical function and the weakening of labour ability are the reasons that make older adults more vulnerable to poverty. Thus, social security, work, and income are essential for older people to cope with poverty risks and maintain welfare (Zhang and Yang, 2020; Amarante and Colacce, 2022; Kamal et al., 2022). Therefore, in this study, we included education, health, living standards, social security, work, and income as the five dimensions of multidimensional poverty on 14 indicators (Table 1). Notably, in the education dimension, considering that the mid-aged and older people generally had a low level of education in the CHARLS sample (Wang and Tian, 2018), the deprivation cutoff was defined as poverty if they had not completed primary school. In the health dimension, we included both physical and mental health indicators. As one indicator of physical health, chronic diseases included 14 diseases, such as hypertension, diabetes, asthma, stroke, etc. We covered social and commercial pension/health insurance in the social security dimension. In the work and income dimension, we used the annual per capita net income within a household as the indicator of income. As to weights of MPI, we chose a normative weighting structure like most prior studies did (Alkire and Santos, 2014; Alkire and Seth, 2015) and selected k = 2 as the poverty cutoff, poverty in at least two dimensions was considered to be with multidimensional poverty.

Measurement of the selected determinants

As described in conceptual framework, we selected five groups of determinants: poverty or nonpoverty durations, personal characteristics, family structure and living arrangement, social capital, and living areas (Table 2). Only two groups of determinants needed explanations. Social capital was defined as a collection of actual or potential resources, mainly embodied in some kind of social relationship network familiar or recognised by the actors with specific institutionalised but informal characteristics (Lin, 1999). Similar to previous studies (Sun et al., 2016; Cao et al., 2022), we used economic transfer with relatives and social activities to measure social capital. As to durations, poverty durations referred to the time from being identified as multidimensional poor to escaping from it. At the same time, nonpoverty durations referred to the time from exiting poverty to returning to poverty.

Analytical methods

First, we used basic descriptive statistics to outline the trend and decomposition of multidimensional poverty.

Second, we applied the poverty transition matrix to analyse the multidimensional poverty dynamics in adjacent survey years (Apablaza and Yalonetzky, 2016; Madden, 2022). The sample size in this part was the whole sample included in this study, i.e., 11,566 respondents. In adjacent survey years, the transition had four types: from poverty to poverty (persistent multidimensional poverty), from poverty to nonpoverty (exit from multidimensional poverty), from nonpoverty to poverty (fall into multidimensional poverty), from nonpoverty to nonpoverty (persistent nonpoverty). Let 1 denote poverty, 0 denote nonpoverty, and p denote the probabilities of the four poverty transition situations. Furthermore, the multidimensional poverty transition matrix from time t to time t + 1 is:

$$P_{t,t + 1} = \left( {\begin{array}{*{20}{c}} {p_{11}} & {p_{10}} \\ {p_{01}} & {p_{00}} \end{array}} \right)$$

Third, considering that poverty transitions between adjacent survey years cannot explore the overall dynamics of poverty through all survey years, we applied the Kaplan–Meier method to estimate poverty’s survival and hazard rates at each time point (Devicienti and Gualtieri, 2010; Glauben et al., 2011). The Kaplan–Meier method is a common nonparametric survival analysis. Survival analysis is becoming a popular method in social sciences since it can model the occurrence of a social event (such as marriage, poverty, and migration) over time. The Kaplan–Meier method is based on all non-left-censored spells (Kleinbaum and Klein, 2012). We included 3,155 respondents starting a poverty spell in 2011 to analyse exit rates. Moreover, we included 1,616 respondents who escaped from poverty in 2013 to analyse reentry rates. The survival rate (hazard rate) is calculated from the survival function (hazard function). Taking exit from poverty as an example, the survival function is expressed as the probability that an individual remains poor after time t:

$$S\left( t \right) = P\left( {T_i \,>\, t} \right)\mathop {\prod}\limits_{t_a \le t} {\frac{{n_a - d_a}}{{n_a}}}$$

where Ti is the duration of individual i being identified as multidimensionally poor. Ta is anytime not exceeding time t. na is the number of individuals being identified as multidimensional poor at ta. da is the number of individuals escaping from multidimensional poverty at ta. The hazard function is expressed as the probability that individual i with Ti ≥ t will escape from multidimensional poverty at time t.

$$h\left( t \right) = \mathop {{\lim }}\limits_{\Delta x \to 0} \frac{{P\left( {t \le T_i < t + \Delta x\left| {T_i \ge t} \right.} \right)}}{{\Delta x}} = \frac{{S\left( {t - 1} \right) - S\left( t \right)}}{{S\left( {t - 1} \right)}}$$

Forth, in line with methods used in studies on income dynamics (Stevens, 1994; Devicienti and Gualtieri, 2010; Glauben, et al., 2011), we selected the discrete-time proportional hazards model, a type of semi-parametric survival analysis, to clarify the determinants of multidimensional poverty dynamics. After deleting the observations with massive missing values in the selected determinants associated with multidimensional poverty dynamics, the final sample size in the discrete-time proportional hazards model was 11,315. We set a complementary log-log hazard rate. Moreover, the hazard function can be expressed as follows:

$$h\left( {t\left| x \right.} \right) = 1 - \exp \left[ { - \exp \left( {\ln h_0\left( t \right) + x_1\beta _1 + x_2\beta _2 + \ldots x_m\beta _m} \right)} \right]$$

where h0(t) is the baseline hazard. Xn is the matrix of the determinants influencing multidimensional poverty dynamics. β is the parameter to be estimated.

Results

The trend, decomposition, and dynamics of multidimensional poverty

From 2011 to 2018, the MPI (M0) showed a downward trend on the whole. This trend mainly depended on reducing the headcount ratio (H). At the same time, the poverty intensity (A) remained at about 0.57 without significant reduction, indicating that the poor population’s welfare status had not significantly improved in these years. Specifically, the sharp drop in the headcount ratio (H) mainly occurred in 2011–2013 and 2015–2018, while H slightly increased during 2013–2015, indicating that poverty entry, poverty exit, and poverty return might coexist during this time (Table 3).

To further explore which dimensions or indicators were the leading causes of poverty, we decomposed the poverty index over the years. From the perspective of dimensions, education had always been the dimension with the highest contribution, reflecting that the mid-aged and older adults were generally poorly educated in the CHARLS sample. The contribution of the health dimension increased year by year, indicating that with age growth, the importance of physical function and mental health to the welfare of mid-aged and older adults gradually increased. As to the dimensions of a standard of living and social security, their contributions showed a downward trend as a whole. The contribution of work and income dimension first increased and then decreased during this period. In terms of indicators, in 2018, the top contributors to overall multidimensional poverty were schooling, employment, activities of daily living, chronic diseases, and depression. From 2011 to 2018, the indicators of running water, housing structure, medical insurance, durable goods, etc. were always kept at a low contribution, reflecting that China’s “two assurances and three guarantees” policies (i.e., free from worries about food and clothing and access to compulsory education, basic medical services, and safe housing) were basically effective. Moreover, the contribution of income and pension insurance dropped sharply from 2015 to 2018, showing that China had greatly achieved eliminating absolute poverty (Table 4).

Based on the analysis from transition matrices, the rate of persistent multidimensional poverty gradually decreased between adjacent survey years (from 13.31% (2011–2013), to 11.42% (2013–2015), to 8.01% (2015–2018)), while the proportion of persistent nonpoverty gradually increased (from 67.72% (2011–2013), to 72.16% (2013–2015), to 73.84% (2015–2018)).

In addition, when comparing the rates of exiting from and falling into poverty, although the number of people who were identified as multidimensional poor gradually decreased as a whole, there were always new people aged ≥45 years who fell into multidimensional poverty (Table 5).

Further explorations of multidimensional poverty dynamics and its determinants

Results from Kaplan–Meier estimates

Among those who fell into multidimensional poverty in 2011, the survival rate of maintaining multidimensional poverty gradually declined from 2011 to 2018. By 2013, only 48.78% of the 3155 individuals in poverty in 2011 remained poor, indicating that more than half of the individuals were transient multidimensionally poor. By 2018, only 17% of the 3155 individuals were still poor, and such a group had spent seven years in multidimensional poverty. The hazard rate of exit from poverty showed a downward trend, on the whole, indicating that the longer the time spent in poverty, the smaller the probability of escaping from poverty. Notably, the hazard rate of exit from poverty rebounded from 2015 to 2018. This may be because China entered a critical period of poverty alleviation after 2015, and more interventions related to poverty alleviation had been carried out. Concerning reentry into poverty, the survival rate of keeping nonpoor after exiting from poverty also gradually decreased from 2013 to 2018. Specifically, 64% and 55% of the 1616 individuals who escaped from poverty in 2013 remained nonpoor in 2015 and 2018, respectively, indicating that 36% and 9% returned to poverty in 2015 and 2018, respectively. The hazard rate of returning to poverty decreased from 2015 to 2018, indicating that as nonpoverty durations increased after exiting from poverty, the probability of returning to poverty decreased (Table 6).

Results from the discrete-time proportional hazards model

Based on the discrete-time proportional hazards model, we found that the likelihood of multidimensional poverty transitions became smaller with the rise of durations of poverty/nonpoverty. Specifically, the longer an individual remained multidimensionally poor, the smaller the probability of exit from poverty; The longer an individual remained nonpoor after escaping from poverty, the smaller the probability of returning to poverty (Table 7). These results verified the above results based on Kaplan–Meier estimates.

As to other factors, we found that men, being married, larger household size, economic exchanges with close relatives or distant relatives, participation in social activities, and living in urban areas significantly increased the likelihood of escaping from multidimensional poverty. Compared with those aged between 45 to 55 years, those aged ≥65 years had a lower risk of escaping from multidimensional poverty. Compared with those living in the eastern region, those in the middle region were less likely to exit from multidimensional poverty. Additionally, men, being married, having larger household sizes, living with children, having economic exchanges with distant relatives, and participating in social activities significantly decreased the possibility of returning to multidimensional poverty (Table 7). In sum, based on the discrete-time proportional hazards model results, H5 was supported, while H1, H2, H3, and H4 were partially supported.

Discussions

This study makes great contributions to the field of multidimensional poverty dynamics as it is the first attempt to explore multidimensional poverty dynamics among older adults globally. The following aspects need further discussion. First, we found that the incidence of multidimensional poverty among mid-aged and older adults basically decreased from 2011 to 2018, but the average poverty intensity seemed not to drop. We also showed that the contribution of health indicators to the overall poverty of mid-aged and older adults increased every year. In 2018, besides schooling and employment, which were hard to improve for mid-aged and older adults, the three health indicators were the top contributors to multidimensional poverty. These results suggest that health has become the most deciding contributor to the overall welfare of mid-aged and older adults in China. That means, to further reduce the multidimensional poverty of such a population, the government should strengthen the health monitoring system for them, provide routine health examinations for them, and pay special attention to their needs related to chronic diseases, activities of daily living, and mental health.

Second, we showed that the probability of multidimensional poverty transitions was negatively associated with poverty or nonpoverty durations. These results are consistent with many existing types of research on income poverty dynamics (Devicienti and Gualtieri, 2010; You, 2011). The government should apply these findings in the dynamic monitoring of poverty. To those who fall into multidimensional poverty for a long time, the government should give more targeted interventions and help since they are the most challenging group to leave the poverty state. For those who fall into multidimensional poverty, the government should provide interventions early when they are most likely to escape from it. For those who are just out of multidimensional poverty, the government should continuously monitor them for some time to prevent them from returning to poverty.

Third, as to personal characteristics, we found that men were more likely to escape from multidimensional poverty and were less likely to return to poverty than women. Women are disadvantaged in acquiring social resources, so they are more vulnerable than men (Admasu et al., 2022; Klasen and Lahoti, 2021; OECD, 2021). Older people (aged ≥65 years) had a significantly lower probability of escaping from multidimensional poverty than the mid-aged (aged between 45 and 55 years). This is probably because older people cannot resist risks caused by physical decline and loss of labour ability (Alkire and Yingfeng Fang, 2019 ; OECD, 2021; Olarinde et al., 2020). The married were found to be more likely to escape from poverty and to be less likely to return to poverty than those not married. Those not married tend to bear more financial pressure and spiritual loneliness, so they have higher risks of suffering from multidimensional poverty (Olarinde et al., 2020).

Fourth, we identified that household size was positively associated with the probability of exiting from poverty but negatively associated with the likelihood of returning to poverty. Moreover, living with children was negatively associated with the possibility of returning to poverty. These results to some extent, contradict other studies in LMICs, showing that large household sizes led to a higher risk of falling into multidimensional poverty (Olarinde et al., 2020; Bautista, 2018). Another study in LIMCs showed that family size had a U-shaped correlation with the MPI (Lekobane, 2022). These discrepancies could be attributed to the corresponding regions’ economic development or cultural background. In China, for mid-aged and older adults, especially older adults, the larger family size means that their care, economic, and spiritual needs can be better shared, which suggests that the overall welfare of mid-aged and older adults in China is largely dependent on family support. However, with the increasing miniaturisation of family size in China, older adults’ traditional family support model is challenging to sustain. Therefore, the government should develop diversified care models for older people, thus comprehensively enhancing their ability to resist multidimensional poverty risks.

Fifth, we uncovered that all proxies of social capital significantly increased the probability of escaping from poverty and all proxies except for economic transfer with close relatives decreased the likelihood of returning to poverty. These results verify the role of social capital in alleviating poverty (Abdul Hakim et al., 2010; Devicienti and Gualtieri, 2010; Glauben et al., 2011). The government should promote communities and social organisations to organise more social activities for older adults and encourage more older people to participate in these activities, thus enhancing their social capital.

Sixth, in terms of living area characteristics, mid-aged and older adults in cities were more accessible to escape from poverty than those in rural areas. Our results are similar to another study based in China (Alkire and Fang, 2019), but are different from one study based in Tunisia using the fuzzy nets approach, showing that urban areas were marked with significantly higher rates of extreme vulnerability to poverty than rural areas (Nasri and Belhadj, 2022). Based on our analysis of the decomposition of MPI, the welfare of mid-aged and older adults mainly depended on physical and mental health. People in cities have better access to high-quality medical services and facilities, which is more conducive to alleviating multidimensional poverty. Additionally, the probability of escaping from poverty in the central region was found to be significantly lower than that in the eastern region. In China, the eastern region has the higher level of economic development than the central region. Our findings verify the positive effect of the macroeconomic level on poverty alleviation (Wang et al., 2020). At the same time, we found that the probability of escaping from poverty in the western region was similar to that in the eastern region. This may be because in the regions with the lowest level of economic development (i.e., the western region), government interventions related to poverty alleviation are intense, which increases the probability of escaping from poverty.

Conclusions

This study mainly found: (1) from 2011 to 2018, the headcount ratio of multidimensional poverty among mid-aged and older adults showed a downward trend, but the average poverty intensity remained basically stable. (2) Health has become the most deciding contributor to multidimensional poverty of mid-aged and older adults. (3) In the whole period, most mid-aged and older adults identified as multidimensionally poor had transient poverty. And in adjacent survey years, the phenomenon of poverty exit and poverty reentry coexisted. (4) Durations of poverty or nonpoverty, individual characteristics, family structure, living arrangements, social capital, and living areas significantly affected the risks of multidimensional poverty exit and reentry. And most of these factors had opposite influences on the risks of multidimensional poverty exit and reentry. These results have far-reaching implications for the government to carry out prevention interventions and also largely enrich the studies of multidimensional poverty dynamics. Considering that we had limited rounds of data in the analysis of reentry to poverty, future studies involving more rounds of data are needed in this regard. Future studies are also needed to explore the influence of COVID on multidimensional poverty.