1 Introduction

The rapid expansion of Chinese oversea investments since the Belt and One Road Initiative has been followed by concerns over the practice and the following impact. Critics claim that the explosion of Chinese development finance may draw recipient countries intro debt traps and hurt the economics in the long time, encourage corruption (Isaksson and Kotsadam 2018a, b) and exploit local natural resources, hire too much Chinese expatriates at the expense of native workers (Yang 2022). At the same time, macro estimates suggest that the higher level of connectivity as a result of more investment in infrastructure will lead to more trade, more FDI and stronger growth and less poverty (Bird et al. 2020; Chen and Lin 2020; de Soyres et al. 2019). However, most research are conducted at macro-level and empirically analysis of the response of households at micro-level is scared and the potential mechanism is under-explored. What is the casual effect of Chinese investment on poverty reduction? What is the potential mechanism?

By linking geographical information on Chinese investment project from Global Chinese Development Finance Dataset (Version 2.0) with survey clusters in the Afrobarometer's public opinion survey, we can identify whether respondents lives near Chinese investment project and explore the local impact of Chinese investment projects on poverty. Besides, there are concerns about the impact of Chinese investment on local institution quality, some research suggests that Chinese investment fuels local corruption (Isaksson and Kotsadam 2018a). However, how the interaction between Chinese investment and local institutions will affect the poverty effect of Chinese projects is still under-explored. Existing literature points out that infrastructure investment is a high-incidence area of corruption (Ali and Pernia 2003; Campos et al. 2021; Kenny 2007; Monteiro et al. 2020; Owusu et al. 2021). The low level of the institution quality will reduce the level of external supervision and constraints on infrastructure investment, intensify the capture of local elites, and make it difficult to achieve the policy goal of benefiting the poor. Inspired by that, we try to develop a framework that can analyze the mediating role of institution quality on the impact of Chinese investment on local poverty and empirically evaluate the poverty reduction effect of Chinese investment and the institution mechanism based on micro level data of projects and individuals living near projects invested by China.

We find that individuals living in areas near active Chinese investment projects have lower level of poverty relative to individuals who not. The poverty reduction effect of Chinese investment projects has always existed, no matter whether we use multidimensional poverty or income poverty to measure the level of individual poverty. In addition, we found that institutional improvement has a significant moderating effect on the income poverty reduction effect of Chinese investment projects, and institutional improvement can increase the improvement effect of Chinese investment projects on income poverty. In contrast, institutional improvement has no significant effect on improving multidimensional poverty in China's investment projects.

This paper expects to make these contributions. Firstly, we expected to contribute to the literature that explore the relationship between poverty and infrastructure investment from developing countries. We try to emphasize the fact that infrastructure investment does not lead to poverty reduction naturally and infrastructural can lead to more poverty if it combined with corruption and lower level of institutional quality. By empirically evaluate the causal effect of Chinese infrastructure investment on poverty reduction, which has arose much concerns and emphasize more on commercial interests rather than just concession and aid, we can provide solid empirical evidence on the relationship between poverty and infrastructure investment of developing countries. Secondly, we try to introduce institutional improvement from the institutional economics perspective as underlying mechanisms that mediate the relationship of local poverty and Chinese infrastructure investment under the Belt and Road Initiative.

1.1 Literature review

The key contend of Chinese investment is infrastructure investment under the Belt and Road Initiative. Therefore, this part reviews the impact of Chinese overseas investment and focus on the infrastructure investment related literature. A lot of literature studies the poverty reduction effect of infrastructure investment. In theoretical research, existing studies have pointed out that infrastructure investment can promote economic growth by increasing input of production factors, empowering production factors, and improving total factor productivity (Arrow and Kurz 1970), thereby driving poverty reduction. At the same time, existing studies have also pointed out that there is an optimal investment scale for infrastructure investment, and infrastructure investment exceeding the optimal level will crowd out other investments, thereby inhibiting productivity growth (Barro 1990; Futagami et al. 1993).

According to different research levels and research methods, empirical studies on the poverty reduction effect of infrastructure investment can be roughly divided into two categories. The first is macro research, which uses data at the national level or a domestic provincial level to conduct cross-country and cross-regional comparative research. The second is micro research, which considers the construction of a certain infrastructure project as a micro intervention, codes the residents of a certain country or region into the experimental group and the control group, and evaluates the impact of infrastructure project intervention on the poverty of individuals.

In macro research, there are two main approaches to measure infrastructure investment. First, use public investment data as a proxy variable for infrastructure investment, and construct the measurement of infrastructure investment from the dimension of monetary value. The basic logic here is that the main flow of government public investment is to build infrastructure, so public investment expenditure can be approximately equal to national infrastructure investment. Second, using the data of infrastructure accessibility as the proxy variable of infrastructure investment. The variables of infrastructure investment are constructed by category from the dimension of physical assets. The basic logic here is that what actually benefits the economy and the people is the availability of infrastructure services such as roads, railways, and power grids. The main limitation of the former lies in the fact that public investment and infrastructure development are not completely equal, and there are other investment flows in public investment, so the measurement error is relatively large. The main difficulty of the latter lies in the limitations of the actual accessibility data of different types of infrastructure services. Comparing the quantity and quality of infrastructure across countries and types requires data conversion and the construction of infrastructure indices.

In the micro research, existing studies identify individuals affected by infrastructure services based on household survey data and compared outcomes between the experimental group and the control group. They pay more attention to outcomes that closely related to poverty level, such as the incidence of poverty, living standards, employment rates, wages et al. As a result, they evaluate the local causal effects of infrastructure investment projects at the micro level. Existing studies have shown that hardware infrastructure such as roads and communications can help improve regional connectivity, promote regional information flow, improve the flow and matching efficiency of labor in different regions, promote the economic development of rural non-agricultural sectors, and ultimately reduce the incidence of poverty (Jalan and Ravallion 2003; Reardon et al. 2007; Zhu and Luo 2005). At the same time, existing studies point out that sound infrastructure contributes to the development of the local market, which increase the employment of the local population in the non-agricultural sector, increase diversity of economic activities and income sources, and increases the income and risk resistance of low-income groups (Khandker and Koolwal 2007, 2010; de Walle and Mu 2007). In the Chinese scenario, infrastructure investment in roads and electricity in rural areas has greatly increased China’s agricultural productivity. When infrastructure investment is matched with education and technology investment, infrastructure can play a greater role in poverty reduction (Fan and Zhang 2004; Zhang and Fan 2004).

The above studies regard infrastructure projects as a whole intervene and pay more attention to the positive impact of infrastructure projects. However, the construction of infrastructure projects has costs. In addition to monetary costs such as financial expenditures, large-scale infrastructure projects may also be accompanied by environmental costs, which may not be evenly distributed among different groups. For example, the construction of the Three Gorges Dam created a reservoir submerged area, involving 21 districts and counties in Hubei Province and Chongqing City. The submerged area includes 2 cities and 11 counties, and 1.31 million immigrants need to be relocated. Based on this, some studies have tried to explore the differential impact of infrastructure projects on different groups. For example, Duflo and Pande (2007) took the dam as the research object, tested the impact of the dam on the upstream people and the downstream people respectively, and pointed out that the dam can reduce the agricultural production loss of the downstream people caused by the rainfall uncertainty, but the ecological and production environment of the upstream people living around became more fragile. The evaluation results showed that after the dam was built, the poverty incidence of the downstream people was significantly reduced, but the marginal effect of the reduction was not enough to offset the marginal effect of the increase in the poverty incidence of the upstream people (Duflo and Pande 2007).

From the above literature analysis, it can be seen that a batch of empirical studies examining the poverty reduction effects of infrastructure and corresponding mechanisms have been gradually born, but there are still the following deficiencies. First, few studies have explored the dynamic model of the poverty reduction effect of infrastructure investment based on the life cycle of infrastructure, and even fewer studies have included institutional improvement as an intermediate mechanism variable into the analysis framework of infrastructure investment's impact on poverty. Second, in terms of research methods, a large number of studies face endogenous problems, and there is a relative lack of micro empirical evidence on the poverty reduction effect of infrastructure investment, the dynamic effect of poverty reduction, and the mechanism of institutional improvement based on causal identification. Endogenous problems specifically include pseudo-regression caused by common time trends, reverse causality, and endogenous infrastructure investment decisions. Third, there are few studies using questionnaire data at the household level to explore the actual responses of individual residents to infrastructure, and a few studies using micro data to explore the heterogeneous welfare effects of infrastructure have made great impact (Duflo and Pande 2007). Fourth, although the Belt and Road Initiative has accumulated rich practices and has attracted widespread attention at home and abroad both in policy discussions and academic exchanges, there are few micro empirical studies try to evaluate the poverty reduction effect of infrastructure investment based on the Belt and Road Initiative. Also, few studies try to identify the poverty reduction effect of the Belt and Road Initiative and assess the potential contribution of the Belt and Road Initiative to fulfilling the promise of the Road to Poverty Reduction and promoting the goals of the United Nations’ 2030 Sustainable Development Agenda.

Therefore, this paper aims to evaluate the poverty reduction effect and the institutional improvement mechanism of China's infrastructure investment projects under the Belt and Road Initiative using project-level dataset and micro household survey.

1.2 Hypotheses

Investment in infrastructure is believed to boost growth and development. At micro level, implementing projects strength the demand for local labour and increase employment rate. Employment channel is effective in reducing poverty when the sector of FDI is agriculture which tend to hire more unskilled and poor people (Maertens et al., 2011). MNE tend to hire more native worker since it is more economically and politically beneficial since local labour force is usually cheaper and creating job can earns more support from local government and communities. Besides, by hiring native worker, the training and employment experiences help local people accumulating human capital and increase the chance of getting higher wages. Lastly, investment and staffs for projects implementation create demand for service industry which benefit unskilled people who usually works as trader or vendor, or run a shop and do retailing. Thus, we tend to hypothesize that:

H1: Living near active projects invested by China is negatively associated with the probability of living under poverty

Institution quality is a key determinant for both economic development and individual wellbeing because it shape individual’s incentive structure and the distribution of economic gains. On the one hand, investment may fuel corruption and encourage rent seeking behaviors by increasing the expected gains from corruption. As a result, poor people do not share much gain from the growth brought by investment and corruptive elites took the whole pie. On the other hand, officials in power may make decision to maximize the social welfare and put more resources to improve the institution which can enhance the positive impact of investment on economic growth and attract more investment. In other words, investment can enhance the expected return of good institution and encourage official in power to improve institution. As a result, poor people who suffer most from bad institution can benefits most both from faster economic growth and larger share of the economic gains. In other words, Chinese investment can improve the lives of the poor both from the direct impact of the investment and the indirect impact from better institution. Thus, we tend to hypothesize that:

H2: The negative association between proximity of Chinese project and poverty is stronger in individuals who experience institution improvement.

2 Method

2.1 Data

To explore the local effect of China's overseas investment projects on individual poverty, we draw on two main sources of data. The data on Chinese investment projects are from AidData's Global Chinese Development Finance Dataset (Version 2.0) (BenYishay et al. 2017), which covers basic information about China's overseas investment projects, like locations and start dates. The data on local residents are from Afrobarometer's public opinion survey. We are limited to the Round 6 data from 2014 to 2015.

Global Chinese Development Finance Dataset (Version 2.0) is collected and geocoded by the AidData Laboratory (Dreher et al. 2019), covering relevant information on China's global development finance projects, including project location, commitment date, start date, completion date, the monetary value of the investment, investor, etc. In terms of the project information collection process, the AidData Lab uses the Tracking Underreported Financial Flows (TUFF) methodology to collect project data, which synthesizes the open source information and more technical details is in (Strange et al. 2013, 2017). Since there are no official data on Chinese overseas investment projects, this dataset is currently the most comprehensive dataset so far and is widely used in the study of China's overseas development finance projects. Also, there is a study using field experiments to test the reliability of the Tracking Underreported Financial Flows (TUFF) methodology (Muchapondwa et al. 2016). Researches on the impact of Chinese overseas investment projects using the dataset have been published in authoritative journals such as American Economic Journal: Economic Policy, Journal of Development Economics, Journal of Public Economics, and World Development (Dreher et al. 2019, 2021; Isaksson and Kotsadam 2018b; McCauley et al. 2022). It is worth noting that we used relatively less controversial and less sensitive information in the dataset with caution, including project location, project commitment date, project implementation date, and project completion date (Isaksson and Kotsadam 2018b).

The second dataset we focus on is the Afrobarometer's survey data and we are limited to Round 6 from 2014 to 2015 (Anon n.d.). Afrobarometer's survey data is a large-scale, individual-level and nationally representative microdata survey conducted in local languages. The survey collects information on the living conditions of local residents, employment and income, corruption experience, trust, and other public opinion issues. Seven rounds of questionnaire surveys have been carried out on the Afrobarometer data, which is widely used to study political and economic issues related to Africa, which published in American Economic Review (Nunn and Wantchekon 2011), Journal of Public Economic (Isaksson and Kotsadam 2018a), World Development (Isaksson and Kotsadam 2018b) and other authoritative journals. Besides, Afrometer Survey also provides precise location information of the survey cluster, which usually includes several villages or a community. By calculating the distance between survey clusters and Chinese investment projects, we can identify whether the individual lives near the Chinese investment project or not and explore the local impact of the Chinese investment project on the daily lives of local residents.

Following the practice of Guo and Jiang (2020) and Knutsen et al. (2017a), we matched Afrobarometer's survey data with AidData's Chinese global finance development project dataset. We take the survey cluster as the center, check whether there is a Chinese project within a radius of 50 km, and further divide survey clusters into three groups: (1) Active, there are Chinese investment projects within 50 km of the survey cluster, and the projects have already been implemented at the time of the survey. (2) Inactive, there will be a Chinese investment project within 50 km of the survey cluster, but the project has not been implemented yet at the time of the survey; (3) No project group, there is no Chinese investment project within 50 km of the survey unit during the entire sample period.

2.2 Variables

2.2.1 Dependent variable

In order to explore the impact of Chinese investment projects on the poverty status of local residents, our core dependent variable is the poverty level of local residents. The measurement of poverty can be divided into two categories. The first method focuses on the monetary dimension and use income and consumption to identify the poor. For example, the World Bank proposed a poverty line of US$1.9 per person per day (PPP in 2011) and claims that individuals whose income and consumption levels are lower than the poverty line can be identified as poor. The second category emphasizes that poverty is multidimensional poverty, including but not limited to insufficient income, but also insufficient basic services, lack of ability, social or political marginalization, etc. The key is that individuals whose basic needs of survival and development are not met can be identified as poor. This method also called the basic need method, is widely used when people care about a broader dimension of poverty. For example, the United Nations has extracted ten indicators from three dimensions of health, education, and living conditions to measure the individual's poverty status. If an individual is below the poverty standard on three or more indicators, the individual is defined as multidimensionally poor. The Global Multidimensional Poverty Index (GMPI) calculated by the United Nations on this basis is a typical example of this type of poverty measurement.

This study proposes two measures of poverty: the lived poverty index and income poverty. There are five questions in the Afrobarometer survey data that are directly related to poverty: (1) Q4a: In the past year, whether, or how often, did you and your family: not have enough food to eat; (2) Q4b: In the past year, whether, or how often, you and your family: lack of sufficient food; (3) Q4c: In the past year, whether, or how often, you and your family: lack of sufficient cleaning Water for household use; (4) Q4d: In the past year, whether, or how often, did you and your family: Lack of adequate medication or medical services; (5) Q4e: In the past year, did you and your family Whether, or how often, their family members: Lack of cash income. The respondents' answers were on a five-level scale from 0 to 4. The frequency of unmet basic needs increased in turn, and the poverty level increased in turn, including: "Never have", coded as 0; "Only 1 to 2 times", code 1; "several times", code 2; "many times", code 3; "always", code 4.

Following Mattes (2008), we first calculate the living poverty index as a measure of the poverty status of local residents. The basic idea is that we use the mean of answers of Q4a-Q4e as the value of the living poverty index. The value of the living poverty index range from 0 to 4, with 0 means not poor, and the poverty level gradually increases from 1 to 4, with 4 representing extreme poverty. The living poverty index is a comprehensive measure of poverty, and the theoretical ideas behind it come from Nobel Laureate Amartya Sen. In Sen (1999), Amartya Sen proposed the perspective of the ability to understand welfare and poverty. He believed that using income to measure welfare and even determine poverty has great limitations because income is just a tool for obtaining welfare rather than a purpose. In real life, people are constrained by the environment and ability, and people with different abilities and living in different environments may get different levels of welfare from the same level of income or consumption. Therefore, rather than just focus on income, he proposed the perspective of capability to measure the state of individual welfare and proposed the definition of poverty from the perspective of capability. He believes that the core of capability is the freedom to choose different ways of life, and the evaluation of welfare should include "no worries about food, no worries about clothing, nutrition and health, guaranteed learning, longevity, political participation and freedom of choice. On this basis, he believes that the essence of poverty is the deprivation of capacity and defines poverty as a state of being unable to meet basic needs or achieve basic abilities. Under the guidance of this poverty theory, the method of identifying and measuring poverty from the dimension of basic needs was developed and widely used. The Human Development Index, the Millennium Development Goals of the United Nations, and the goals in the 2030 Agenda for Sustainable Development are all applications of this basic need method. As an indicator following the basic need method, the living poverty index belongs to the secondary category of poverty measurement. In terms of measurement validity, some studies have pointed out that the living poverty index has strong internal validity and can be used for cross-country comparisons between multiple rounds of questionnaire data (Meyer and Keyser 2016).

Considering the importance of income in the research of poverty, we construct the measure of income poverty. We use Q4e in the Afrobarometer questionnaire data, that is, "Did, or how often, in the past year, you and your family: lack of cash income.", and divide income poverty into a five-level scale from 0 to 4, where 0 means that there is no shortage of cash income and 4 means that individuals are in a state of extremely short of cash income and extreme poverty. Due to the wide application of poverty measurement indicators focusing on the income dimension, such as the incidence of poverty, in the policy formulation process of international organizations, taking income poverty as the dependent variable make our result more comparable with other researchers.

2.2.2 Independent variable

In order to assess the impact of Chinese investment projects on the poverty of local residents, following the practices of exsting literature (Isaksson and Kotsadam 2018a, 2018b; Knutsen et al. 2017), we divided individuals into three groups, and our core explanatory variable was the grouping variable \(\mathrm{Active}\) and \(\mathrm{Inactive}\).

Active

The Active variable will be coded as 1 if there is an active Chinese project within 50 km of the surveyed cluster where the individual belongs at the time of the interview, and will be coded as 0 in other cases.

Inactive

The Inactive variable will be coded as 1 if there is a Chinese invested project within 50 km of the survey cluster where the individual is located during the whole sample period, but the project has not yet started construction at the time of the interview, Inactive will be assigned a value of 1, and the Inactive variable will be coded as 0 in other cases.

Relying on these two binary variables of Active and Inactive, we divided the individuals in the sample into three groups, namely active group, inactive group, and no project group. Compared with the traditional division of individuals into groups with projects and without projects, this division method has the following advantages. Firstly, the introduction of Inactive variables is helpful in testing whether the location selection bias is significant by analyzing the significance of the coefficient of Inactive. The significance of the coefficient of the project inactivity variable can reflect whether there is a systematic difference in poverty between the locations selected by Chinese investors and those not. If the coefficient is not significant, it means that the poverty status of local residents has little influence on investors' location decisions process, and the location selection bias is well addressed by adding Inactive as control. Secondly, we can construct and test diff-in-diff estimators \({\upbeta }_{\mathrm{Active}} - {\upbeta }_{\mathrm{Inactive}}\). The DID estimator \({\upbeta }_{\mathrm{Active}} - {\upbeta }_{\mathrm{Inactive}}\) captures the local causal effect of active Chinese investment projects on the poverty of residents nearby.

Institution

In order to explore whether the institutional quality is a mechanism through which Chinese projects affect local poverty, we introduce Institution variable to capture the dynamics of the local institutional environment. Following Acemoglu and Johnson (2005), we use corruption levels as a r Q54: In your perception, has the level of corruption in this country increasemeasurement of institutional quality. This variable is based on the answers of Afrobarometed, decreased or remained the same over the past year? If the individual's answer is "a little drop" and "a lot of drop", the Institution variable is assigned a value of 1, meaning that the institutional quality improved, and the rest are assigned a value of 0. On this basis, we aggregated the individual-level perceptions of institutional improvement in the survey cluster level. Our aggregation method is that if more than 50% of the residents in the geographic unit believe that the quality of the institution has improved and the level of corruption drops, the survey cluster level Insitution variable is coded as 1; otherwise, the survey cluster level Institution variable will be coded as 0.

Considering data availability and following existing literatures (Guo and Jiang 2020), we also controlled for the following variables that may affect whether an individual is poor: gender, race, ethnicity, age, education level, and family labor force size.

2.2.3 Data analyses

In order to explore the impact of Chinese investment projects on the poverty of residents nearby, we divide individuals in the sample into three groups according to whether there is a Chinese project within 50 km of the individual's location and whether the project has been active at the time of the interview: active group, inactive group, and no project group. Following existing literature (Isaksson and Kotsadam 2018a, 2018b; Knutsen et al. 2017), we constructs the following model to compare the differences in poverty levels among three groups of individuals to identify the poverty reduction effect of Chinese investment projects:

$${Y}_{is}={\beta }_{0}{\text{ Inactive }}_{s}+{\beta }_{1}{\text{ Active }}_{s}+{{\varvec{X}}}_{i}\gamma +{u}_{z}+{v}_{ist}$$
(1)

Among them, \(i\) represents the individual, s represents the survey cluster where the individual is located, generally a neighborhood in a neighboring village or town, and z represents the country where the individual is located. The explanatory variable \({Y}_{is}\) measures the poverty status of individual \(i\) in the survey cluster \(s\), including the living poverty index (\({LPV}_{is}\)), which comprehensively measures multidimensional poverty, and income poverty (\({IncomePoverty}_{is}\)), which measures poverty in the monetary value dimension. The core explanatory variable is the grouping variable. \({\text{ Active }}_{s}\) is a binary variable that identifies survey clusters nearby active Chinese projects. If there is a Chinese investment project within 50 km around the survey cluster \(s\), and the start time of the project is earlier than the survey time, \({\text{Active }}_{s}\) is assigned a value of 1, otherwise, it is 0. \({\text{Inactive }}_{s}\) is a binary variable that identifies survey clustes neaby Chinese investment projects but these projects have not yet been implemented at the time of survey. If the survey cluster s has a Chinese investment project nearby during the sample period, but the Chinese investment project has not been active at the time of the survey, then the value of \({\text{Inactive }}_{s}\) is 1, otherwise \({\text{Inactive }}_{s}\) will be coded as 0. The benchmark group includes survey clusters located not near any Chinese investment projects during the whole sample period. \({\beta }_{1}\) measures the difference in average poverty levels between the active group and the non-project group, and \({\beta }_{0}\) measures the difference in average poverty levels between the inactive group and the non-project group. \({\beta }_{1}-{\beta }_{0}\) captures the difference in average poverty levels between the active group and inactive group using the non-project group as benchmark, i.e., difference in difference effect. It is worth noting that the introduction of \({\text{Inactive }}_{s}\) allows us to directly test the significance of the location selection bias, and \({\beta }_{0}\) reflects the systematic difference in poverty level between locations selected for Chinese investment projects and locations not selected. It means that there is no systematic difference between the areas selected by Chinese investment projects and the areas not if \({\beta }_{0}\) is not significant, and the location selection bias is not significant.

\({{\varvec{X}}}_{i}\) contains individual-level control variables that may affect individual poverty status, including individual gender, age, race, whether or not the ethnic group is discriminated against, years of education, and the size of the family labor force. In addition, we also control for the regional fixed effect \({u}_{z}\). Standard errors are heteroscedastic standard errors and are clustered at the survey cluster level.

In order to explore the variable of institutional improvement that may affect the poverty reduction effect of Chinese investment projects, we introduce the binary variable of institutional improvement and divide the survey clusters into areas with improved institutions and areas without institutional improvement. By comparing the poverty reduction effect of Chinese investment projects in areas with institution improvement and in areas without, we explore the moderating effect of institutional improvement:

$$Y_{is}=\beta_0\;{\mathrm{Inactive}}_s+\beta_1\;{\mathrm{Active}}_s+\beta_2\;{\mathrm{Institution}}_s+\beta_3\;{\mathrm{Inactive}}_s\times{\mathrm{Institution}}_s+\beta_4\;{\mathrm{Active}}_s\times{\mathrm{Institution}}_s+X_i\gamma+u_z+v_{ist}$$
(2)

Among them, the dependent variable \({Y}_{is}\) is the same as the above regression equation, which is the poverty status of individual i in the survey cluster s, measured by two variables: living poverty index (\({LPV}_{is}\)) and income poverty (\({IncomePoverty}_{is}\)). \({\text{Inactive }}_{s}\) and \({\text{Active }}_{s}\) are the dummy indicator for the inactive group and the active group, respectively. \({\text{Institution }}_{s}\) captures whether there is an institutional improvement in the survey cluster s based on individual's perceptions of the dynamic of corruption levels. \({Inactive}_{s} \times {Institution}_{s}\) is the interaction item between the inactive group and the institutional improvement, \({\text{Active}}_{s}\times {\text{Institution}}_{s}\) is the interaction item between the active group and the instituioal improvement. \({\beta }_{0}\) measures the difference in poverty level between the inactive group and the non-project group when both group does not experience institutional improvement. \({\beta }_{1}\) measures the difference in poverty level between the active group and the non-project group when the local institutional envirionment does not improve. Besides, \({\beta }_{1}-{\beta }_{0}\) measures the poverty reduction effect of Chinese investment projects in areas without institutional improvement. \({\beta }_{2}\) measures the poverty reduction effect of institutional improvement in areas that do not have any Chinese investment projects nearby during the whole smaple period. \({\beta }_{3}\) measures the additional poverty reduction effect of inactive Chinese investment projects with the help of institutional improvement. \({\beta }_{4}\) measures the additional poverty reduction effect of active Chinese investment projects with the help of institutional improvements.

Like regression Eq. (1), \({{\varvec{X}}}_{i}\) also includes individual-level control variables that may affect individual poverty status, including individual gender, age, race, discriminated ethnicity, years of education, and the size of the household labor force. In addition, we also control for the regional fixed effect \({u}_{z}\). Standard errors are heteroscedastic standard errors and are clustered at the survey geographic unit level.

3 Results

3.1 Descriptive statistics of the sample

Table 1 presents descriptive statistics of continuous variable in the sample, which includes LivedPovertyIndex, IncomePoverty, age, education and labor size. As can be seen the average lived poverty index is 2.085, suggesting that most people in the sample experience several times of some basic needs do not met. The average of Income poverty is 0.392, suggesting that 39% of individual in the sample experience many times or always short of cash income, which is in extreme poor status. The average age is 37 years old in the sample and the minimum age is 18 years old, suggesting that interviewee in the sample is adult and most of them are in the middle age. Average education status is 3.4, suggesting the average highest education level of individuals in the sample is between primary school and high school completed. On average, the family labor size is about 4 people.

Table 1 Summary Statistic for continuous variable
Full size table

Tables 2, 3 and 4 shows the percentage for categorical variables which includes Active, Inactive, Institution Improvement, and we also present the two-way table of frequency between group dummy and institutional dummy in Tables 5, 6 and 7. As can be seen in Table 5, most respondents in the sample lived in areas without any Chinese investment project nearby and 49,802 respondents is in the no project group. And 1060 respondents lives near active Chinese investment project and are assigned into the active group, and 3073 respondents lives near inactive Chinese investment project and are assigned into the inactive group. Besides, only 7.23% of respondents in the sample lives in area with institution improvement.

Table 2 Tabulation of active
Full size table
Table 3 Tabulation of inactive
Full size table
Table 4 Tabulation of institution improvement
Full size table
Table 5 Two-way tabulation of active and inactive
Full size table
Table 6 Two-way tabulation of active and institution improvement
Full size table
Table 7 Two-way tabulation of inactive and institution improvement
Full size table

In Table 6, we can find that 1,028 respondents lives in areas near Chinese investment project but do not have institution improvement, 3,865 respondents lives in areas experiencing institutional improvement but do not have any active Chinese investment project nearby, and 32 respondents live in area near active Chinese and experiencing institutional improvement, and most respondents do not live in areas near active Chinese investment project or having institution improvement.

In Tables 7, 8, 9, 10, we can find that 2,801 respondents live in areas which will be picked up by Chinese investment project in the future and do not experience institutional improvement, 3,625 respondents lives in areas with institutional improvement but not near any inactive Chinese investment project, 272 respondents lives in areas with institutional improvement and near inactive Chinese investment. And we also report the percentage for categorical variables which includes Gender, Unfairly treated ethic group and Race in Table 7, 8, 9. We can find that 50.31% of respondents are men. And we also find that 84.69% of respondents belongs to ethic groups which are treated unfairly and about 96.93% of respondents belongs to Black African, and the rest of respondents lives in areas without institutional improvement and do not have any inactive Chinese investment project nearby.

Table 8 Tabulation of gender
Full size table
Table 9 Tabulation of unfairly treated ethic group
Full size table
Table 10 Tabulation of race
Full size table

We can also see that nearly 50% of respondents are men and the gender distribution is relatively balanced. And most people (84.69%) do not think they belong to ethic group that had been treated unfairly and most respondents are black African.

3.2 Results

The baseline result of the impact of Chinese investment projects on poverty measured by the living poverty index is presented in Table 11. Column 1 is the result without any control variables, column 2 controls the country-level fixed effects. The individual-level control variables that will affect the individual's poverty status are controlled one by one and the result is shown from column 3 to column 8, followed by gender, age, education, race, family workforce size, and discrimination against ethnic groups. From columns 1 to column 8, the sign of the active group is significantly negative, as expected, suggesting that individuals living in areas near active Chinese investment projects have lower levels of poverty relative to individuals who do not. In other words, active Chinese investment projects can significantly reduce the frequency with which an individual's basic needs are not met. At the same time, the significance of the coefficient of the project inactive group gradually decreased and finally became no longer significant with more controls, and the coefficient continued to decrease and finally approached 0. This means that after controlling for national fixed effects and individual-level control variables such as gender and age that affect whether individuals are poor, there are no significant differences in poverty levels in areas picked up by Chinese investment projects and areas that were not before projects became active, i.e., the effect of location selection bias was small and insignificant.

Table 11 The impact of Cinese investment projects on living poverty index
Full size table

It is worth noting that Table 11 reports the result of diff-in-diff estimator, that is, the coefficient and significance of \({\beta }_{1}-{\beta }_{0}\). It can be seen that from column 1 to column 8, the signs of the diff-in-diff estimator are all negative, as expected. The significance of the diff-in-diff estimator rises gradually with more controls and is significant at the 10% confidence level finally. This shows that, after controlling for location selection bias, active Chinese investment projects have a significant local effect on the poverty of local residents.

In Table 12, we present the baseline evidence of the impact of Chinese investment projects on poverty measured by Income Poverty. The first column is the result without any control variables, the second column controls the regional fixed effect, and the third column to the eighth column controls the gender, age, education, race, family labor size, and whether it is the discriminated ethnicity, etc. The coefficients of the active group were all significantly negative, which was in line with the theoretical prediction. Compared with areas without Chinese investment projects nearby, active Chinese investment projects can significantly improve the income poverty of local residents and reduce the frequency of individuals experiencing lacking cash income. Besides, the significance of the coefficient of the project inactive group gradually decreased with more controls and finally became insignificant, and the coefficient gradually became smaller and approached 0.01, indicating that there is no systematic difference in poverty between residents in inactive project areas and non-project areas after controlling for the country effect and individual-level control variables. In other words, the impact of location selection bias is small and insignificant, and the impact of active Chinese investment projects on the poverty of local residents is not driven by the location preference of Chinese investment projects.

Table 12 The OLS estimates of the impact of Chinese investment projects on income poverty
Full size table

Also, the estimates of the diff-in-diff estimators are reported in the Table 12. The coefficients of the double-difference estimators are all negative from columns 1 to 8, which means that individual living near active Chinese investment projects experiences a lower level of poverty compared to individuals living near inactive Chinese investment projects. It suggests that active Chinese investment projects have poverty reduction effect after controlling location selection bias. Besides, it is noted that with the addition of control variables, the significance of the diff-in-diff estimator gradually increases, but it is not significant, which indicates that active Chinese investment projects do not significantly improve income poverty.

In order to deeply understand the potential mechanism of Chinese investment projects to reduce poverty, we further explore the impact of institutional improvement on the poverty reduction effect of Chinese investment projects.

In Table 13, we shown the mediating impact of institution improvement on the poverty reduction effect of active Chinese investment projects with the living poverty index as the dependent variable. The first column does not add any control variables but includes the core explanatory variables and the interaction terms of the core explanatory variables, the second column controls the regional fixed effect, and the third to eighth columns add Individual-level control variables like gender, age, education, for race, household labor force size, and discrimination against ethnicity.

Table 13 OLS Estimates of the mediating effect of institution on poverty impact of Chinese projects
Full size table

First, the coefficients of the active group are significantly negative from columns 1 to 8, meaning that even if the effects of institutional improvement are stripped away, the poverty level of residents living near active Chinese investment projects is still significantly lower than in areas without projects. At the same time, the coefficient of the project inactive group is not significant and approaches 0 with the addition of control variables. It means that there is no systematic difference in the poverty levels of individuals in the inactive group and individuals in the non-project group, indicating that the location selection bias is small after controlling the impact of institutional improvement. This means that the poverty reduction effect of active Chinese investment projects and institutional improvement is not driven by the location preferences of Chinese investors.

The coefficient size and significance level of the diff-in-diff estimator separated the institutional improvement effect are also reported in the table. It can be seen that from columns 1 to 8, the coefficients of the double-difference estimators are all negative, along with the control variables. The added significance level gradually increases and is significant at the 10% confidence level.

Secondly, the coefficient of institutional improvement is significantly negative from column 1 to column 8, which means that residents in no project areas will significantly improve their poverty status due to institutional improvement, which is in line with theoretical expectations.

Finally, the coefficient of the interaction term between the active project group and institutional improvement is not significant, indicating that there is no complementary effect between active projects and institutional improvement. Similarly, the coefficient of the interaction term between the inactive project group and institutional improvement is not significant.

The results of exploring the moderating effect of institutional improvement with income poverty as the dependent variable are shown in Table 14. Same as Table 13, the first column contains only core explanatory variables and interaction terms, the second column controls the regional fixed effect, and the third to eighth columns control gender, age, education, race, and family labor size and whether it is a discriminated ethnicity, respectively.

Table 14 OLS estimates of the mediating effect of institution on poverty impact of Chinese projects
Full size table

First, the coefficients of the active project group are significantly negative from the first column to the eighth column, which means that the poverty status of residents in the active group has been significantly improved compared with the residents in the non-project group after controlling the impact of institutional improvement. At the same time, it is noted that after full control, the coefficient of the project inactive group becomes no longer significant, and the coefficient approaches 0, indicating that the poverty reduction effect of the active group separated from institutional improvement is not driven by location selection bias. The magnitude and significance of the coefficients of the diff-in-diff estimator are shown at the bottom of the table. The coefficients of the diff-in-diff estimator are all negative, which is in line with the theoretical expectation; that is, compared with the inactive group, the active Chinese investment projects have a stronger poverty reduction effect after separating the effect of institutional improvement.

At the same time, it is noted that the coefficient of the interaction terms between the active group and the institutional improvement is significantly negative from the first column to the eighth column, indicating that the active project and the institutional improvement have a significant complementary effect in reducing income poverty. That is, compared with areas without institutional improvement, the income poverty reduction effect of active Chinese investment projects is larger in areas with institutional improvement. Besides, the coefficient of the interaction term between the inactive project group and the institutional improvement is not significant, indicating that the complementary effect of the active Chinese investment project and institutional improvement is not driven by the location selection bias.

4 Conclusion

With the expansion of China's overseas investment and the high-quality development of the joint construction of the "Belt and Road", the real impact of China's overseas investment on host countries has attracted widespread attention. Based on this, this paper uses the project-level Global Chinese Development Finance Dataset and the individual-level Afrobarometer's public opinion survey data to construct an approximate double difference identification method to analyze and test the poverty reduction effect of China's overseas investment projects. On this basis, this paper introduces institutional improvement as a mediator and examines the impact of institutional improvement on the poverty reduction effect of China's overseas investment projects.

We made the following discovery. First, in terms of the main effect, no matter whether income poverty or multidimensional poverty is used to measure individual poverty, China's oversea investment projects have a significant poverty reduction effect on residents living near the projects. Second, when income poverty is used as the dependent variable, institutional improvement has a significant moderating effect. Institutional improvement can significantly enhance the effect of China's foreign investment projects on income poverty.

The above findings provide us with two revelations. First, the joint construction of the "Belt and Road" has a realistic basis for building a road to poverty reduction. Empirical evidence has shown that China's foreign investment projects promoted by the joint construction of the "Belt and Road" have shown a poverty reduction effect at the micro level. The high-quality development of the joint construction of the "Belt and Road" can contribute to the cause of global poverty reduction. We should attach great importance to it, actively plan and fully tap the poverty reduction potential of the joint construction of the "Belt and Road". Second, institutional improvement can effectively enhance the role of China's foreign investment projects in reducing income poverty under the joint construction of the "Belt and Road". Therefore, in the process of promoting the joint construction of the "Belt and Road" to build a road to poverty reduction, we must pay attention to the combination of software and hardware. We must not only pursue the high-quality development of investment projects, but also pay attention to the improvement and follow-up of supporting institutions to form a joint force for poverty reduction.

4.1 Data statement

The data that support the findings of this study are available from AidData laboratoryFootnote 1 and AfrobarometerFootnote 2 but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Afrobarometer.