Policy Research Working Paper 11029 Did Program Support for the Poorest Areas Work? Evidence from Rural Viet Nam Hai-Anh H. Dang Klaus Deininger Cuong Viet Nguyen Development Data Group & Development Research Group January 2025 Policy Research Working Paper 11029 Abstract This paper investigates the impact of a large-scale poverty significantly increases the share of nonfarm income for rural alleviation program targeted at the 62 poorest districts in households. A possible explanation for the positive effects Viet Nam. The analysis of multiple data sets spanning the on nonfarm employment is the improved access to credit past 20 years uses a regression discontinuity design with that the program provides to participating households. The district fixed effects. The findings do not reveal significant findings also show that the program increases household program effects on household welfare (as measured by per access to electricity, public transfers, educational subsidies capita income and poverty) or local economic development for students residing in the program districts, and health (as measured by nighttime light intensity and establishment care utilization, possibly through improving the availability of new firms). However, the findings show that the program of commune health care centers. facilitates a shift from farm to nonfarm employment and This paper is a product of the Development Data and the Development Research Groups, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank. org/prwp. The authors may be contacted at hdang@worldbank.org and vietcuong@vnu.edu.vn. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Did Program Support for the Poorest Areas Work? Evidence from Rural Viet Nam Hai-Anh H. Dang, Klaus Deininger, and Cuong Viet Nguyen* JEL: C15, D31, I31, O10, O57 Key words: poverty, targeting, household surveys, Viet Nam * Dang (hdang@worldbank.org; corresponding author) is a senior economist in the Living Standards Measurement Study Unit, Development Data Group, World Bank and is also affiliated with GLO, IZA, Indiana University, London School of Economics and Political Science, and University of Economics Ho Chi Minh City, Vietnam; Deininger (kdeininger@worldbank.org) is a lead economist with the Sustainability & Infrastructure Unit, Development Research Group, World Bank; Nguyen (vietcuong@vnu.edu.vn; corresponding author) is a researcher in International School, Vietnam National University, and Thang Long Institute of Mathematics and Applied Sciences (TIMAS), Thang Long University, Hanoi, Vietnam. We would like to thank Alessandro Barattieri, Paul Burke, Hal Hill, Duong Le, Hoa Nguyen, Chi Ta, Peter Warr, and participants at seminars at Australian National University and the World Bank Chief Economist’s Office for East Asia and Pacific Region for helpful feedback on earlier drafts. We are grateful for funding support from the United Kingdom Foreign Commonwealth and Development Office (FCDO)’s Data and Evidence for Tackling Extreme Poverty (DEEP) Research Program and University of Economics Ho Chi Minh City, Vietnam (UEH). 1. Introduction Reducing poverty is the first goal of the 17 Sustainable Development Goals (SDGs) in the United Nations' 2030 Agenda for Sustainable Development. Yet, due to limited assets and credit constraints, poor households usually face challenges in investing in human capital and productive activities, leading to these households being captured in a poverty trap (Balboni et al., 2022). Consistent with earlier observations that the geographic structure of living standards can be reasonably stable over time (Ravallion and Wondon, 1999; Jalan and Ravallion, 2002), recent studies offer further evidence that persistent poverty is more likely to be concentrated in remote rural regions in many countries (Beegle et al., 2011; Kraay and McKenzie, 2014; Hallegatte et al., 2016; Pritchett and Hani, 2020; McBride et al., 2022). Consequently, a common approach to poverty reduction involves implementing large-scale poverty alleviation programs that target disadvantaged, remote areas (OECD, 2009; World Bank, 2009). 1 We investigate the effects of a large-scale poverty alleviation program on household welfare in Viet Nam. Despite significant progress with poverty reduction, Viet Nam still faces substantial disparities in poverty across different ethnic groups and geographic regions (Dang, 2012; Lanjouw et al., 2017; World Bank, 2022). Recognizing the geographical dimension of poverty, since 2009 the Vietnamese government has initiated a major poverty reduction program—the 30A Program—that focuses on the country’s 62 poorest districts (Government of Vietnam, 2008). 2 This program is designed to improve income and living standards within the program districts by fostering agricultural productivity and generating nonfarm employment opportunities for the local population. The 30A Program encompasses 1 Such programs are commonly referred to in the literature as “place-based” policies (which differ from people- based policies aimed at targeting support to households or individuals). Place-based programs could generate positive externalities stemming from network effects, thus stimulating economic development in the areas (Neumark and Simpson, 2015). 2 There are 63 provinces and around 700 districts in Viet Nam. 2 four primary components: support for production and employment, infrastructure investment, policies concerning personnel in poor districts, and educational assistance for individuals and public personnel. One significant form of support for households is the program provision of microcredit and cash subsidies, which aim to enhance poor households’ agricultural productivity and facilitate their transition to non-farm employment. Overall, our analysis reveals no significant impact of the 30A Program on household income or poverty. We also do not find significant program effects on local economic development, as measured by commune-level nighttime light data, establishment of new firms or migration inflows and outflows. However, we find positive program impact on household transitions from farm self-employment to nonfarm self-employment, leading to increased nonfarm income and decreased farm income. The results are robust to various falsification tests, RDD bandwidths, model specifications, different ways of estimating the standard errors, and program spillover effects. Notably, we find substantial and positive program effects on formal borrowing and microcredit utilization, which could be a plausible reason for these positive program effects on nonfarm employment. We also find heterogeneous program effects on nonfarm employment across gender, education level, and ethnicity and larger effects in the longer term. Moreover, the program has improved household access to public services, including electricity, public transfers, health care utilization, and educational subsidies for students. We make several new contributions to the literature on evaluating poverty alleviation programs. Despite the existence of numerous poverty alleviation initiatives in low- and middle-income countries, there is still a dearth of empirical research investigating the impact of large-scale poverty reduction programs targeting impoverished areas. Recent studies have focused on other developing countries such as China, India, and countries in Latin America 3 (Meng et al., 2013; Banerjee et al., 2015; Asher and Novosad, 2020; Bahal, 2020; Millan et al., 2020; Chaurey and Le, 2022), but not much evidence is available for Viet Nam. Viet Nam presents an interesting, if not unique, case study. The country has achieved one of the fastest poverty reduction rates in the world. Following its economic reforms in 1986, its (headcount) poverty rate sharply decreased by about one-third from 58% in 1993 to 37% in 1998, and another two-thirds to 14% in 2010 and 12% in 2012 before reaching a remarkable low of 4% in 2020 (World Bank, 2022). While the country has implemented a number of poverty reduction programs based on a strong pro-poor development strategy, there is, however, limited evidence linking these interventions to its successes in poverty reduction. More worrisome, recent evidence suggests that poverty is increasingly segregated in certain provinces and there is rising within-province inequality in Viet Nam (Benjamin et al., 2017; Lanjouw et al., 2017; Dang et al., 2023). By offering a multi-component poverty reduction program supporting various aspects of the economy, ranging from employment support, investment in infrastructure, improvement with government personnel policies to education support, the country aims to sustainably reduce its poverty. Yet, the few existing studies exploring the impact of other large-scale poverty reduction programs in Viet Nam offer mixed evidence. Nguyen et al. (2015) show positive impact of Program 135 during 2007-2012, which specifically targeted ethnic minority communes, on the living standards of households within the program's targeted areas. However, Phan et al. (2016) do not find significant effects of national targeted programs for poverty reduction during 2002-2010. At the same time, the 1993-1998 period had no national targeted poverty alleviation programs but still witnessed a significant poverty decline. From 2000 onwards, Viet Nam has implemented various poverty alleviation programs, with nearly VND 560 trillion (approximately US$25 billion) invested in all national targeting programs 4 between 2010 and 2019, including the 30A poverty reduction program (World Bank, 2022). The question of whether poverty alleviation programs truly assist the poor in escaping poverty in Viet Nam is thus of importance for policy makers and various aid donors, regarding both accurate evaluation of past programs and effective inputs for future policies. Evaluating poverty alleviation programs that focus on disadvantaged areas, however, presents a challenge due to the nonrandom placement of such programs (Ravallion, 2007; Meng, 2013). In Viet Nam, the 30A Program covers the poorest districts that had a poverty rate exceeding 50% in 2006. This threshold enables us to employ a sharp Regression Discontinuity Design (RDD) method combined with the difference-in-difference method, together with district fixed effects regressions, on rich data from the Vietnam Household Living Standard Surveys spanning the period 2004-2020 to rigorously estimate the program effects on household welfare. Our study thus offers more rigorous evaluation of a large-scale poverty reduction program and covers the longest time period that such a program was in operation to date in Viet Nam. 3 Furthermore, beyond offering new evidence from Viet Nam, to our knowledge, very few existing studies on poverty reduction programs for other developing countries analyze a rich set of outcomes as we do. These outcomes are diverse, ranging from employment (including working rates and work sectors), wages (including incomes from different sources), health insurance (including numbers of health care visits), education (including school enrollment and receipt and amount of school subsidies), poverty measures to household access to credits and loan sizes, and availability of community public services and 3 Several other features distinguish our study from these existing studies. While Nguyen et al. (2015) employ a difference-in-difference model to analyze household-level outcomes, Phan et al. (2016) use a GMM model to analyze province-level outcomes. We employ a more rigorous, combined RDD and difference-in-difference model with district fixed effects and analyze a richer database consisting of household surveys, firm surveys, and nighttime lights data. The time period that we analyze is also (more than) twice longer and more updated than those in the other studies. 5 infrastructure. The outcome variables are nationally representative and measured at different levels including individuals, households, communes, and districts, which allow us to gain a deeper understanding of the program effects and the mechanisms through which the program could impact household welfare for the country. This paper has six sections. Section 2 presents an overview of the 30A Program in Viet Nam. Section 3 presents data sets and descriptive analysis. Section 4 discusses the estimation method, and Section 5 discusses the estimation results, including robustness checks, potential mechanisms, and heterogeneous program impact. Finally, Section 6 concludes. 2. The program for the poorest districts in Viet Nam Viet Nam has adopted a "growth with equity" strategy for its socio-economic development, demonstrating a commitment to broad-based development. The government has implemented various social assistance policies and programs to promote inclusive development and alleviate poverty. From 2006 to 2010, the central government allocated approximately VND 45 trillion (equivalent to approximately USD 2.8 billion at that time) towards poverty alleviation programs, which increased to around VND 75 trillion (approximately USD 3.4 billion at that time) for the 2012-2020 period (MOLISA and UNDP, 2009; Government of Vietnam, 2016). In addition, local provinces and private and international organizations also contribute to poverty alleviation programs, contributing around VND 70 trillion (approximately USD 3.2 billion) to poverty alleviation programs during the 2016-2020 period (Government of Vietnam, 2021). Between 2010 and 2019, nearly VND 560 trillion (approximately US$25 billion) was invested in all national targeting 6 programs, including the 30A poverty reduction program (World Bank, 2022). 4 Moreover, Viet Nam received substantial overseas development assistance amounting to nearly USD 50 billion during 2000- 2021 (World Bank, 2023). Since 1998, Viet Nam has implemented two major programs focused on poverty alleviation. The first is the National Targeted Program for Poverty Reduction (NPPR), aimed at assisting impoverished households and communities in developing their production capabilities, increasing incomes, accessing social services, eliminating hunger, and reducing poverty. The second program, 'Socio-economic Development for the Communes Facing Greatest Hardships in the Ethnic Minority and Mountainous Areas’, commonly known as Program 135, provides support to nearly 2,000 communes located in poor and remote areas. These communes are characterized by a significant proportion of ethnic minority households and high poverty rates (CEMA and UNDP, 2009). Since 2009, Viet Nam has implemented another major program dedicated to sustainable and rapid poverty reduction in the country’s poorest districts, which is also known as the 30A Program (because it was approved by Resolution No. 30a/2008/NQ-CP (Government of Vietnam, 2008)). The program specifically targeted 62 districts with a poverty rate exceeding 50% in 2006 (using the poverty rate estimated by the Ministry of Labor, Invalids, and Social Affairs (MOLISA)). Figure 1 presents the geographic distribution of poverty rates in those districts during 2006. The 30A Program districts are located in 20 provinces, primarily in the Northern midlands and mountain areas, North Central and Central coastal areas, and Central Highlands. These regions represent the poorest areas in the country, characterized by a high concentration of ethnic minority populations. In the 62 targeted districts, ethnic minorities constitute 90% of the population. 4 This sum roughly equals 10% of the country’s annual GDP during this period. 7 Notably, Viet Nam has combined both the 30A Program and the 135 Program into the NPPR since 2012 (Government of Vietnam, 2012 and 2016). However, the support policies and target groups remain separate between the 30A Program and the 135 Program. From 2009 to 2016, approximately VND 20 trillion (equivalent to approximately USD 1 billion at that time) was allocated and mobilized to support the poor districts under the 30A Program. Between 2016 and 2020, as part of the NPPR, the 30A Program and the 135 Program received a total support of VND 48 trillion (equivalent to approximately USD 2.1 billion at that time) (Government of Vietnam, 2021). The primary objective of the 30A Program is to increase the average incomes of households in the targeted districts to five to six times higher than the 2008 level by 2020. The government expects to achieve this goal through policies aimed at improving agricultural productivity and generating non-farm employment opportunities (Government of Vietnam, 2008). The 30A Program aims to significantly enhance the living standards of impoverished and ethnic minority communities in the targeted districts, with the goal of reaching the same living standards in other districts in the region by 2020. The program's targets include reducing the poverty rates in the targeted districts to 40% by 2010, aligning these poverty rates with the provincial averages by 2015, and matching them with the regional averages by 2020. Figure 2 provides a simple causal-chain framework that links the inputs and expected outcomes of the 30A Program. To achieve the objectives of increasing income, reducing poverty and improving public services, the program implements four main components including: (i) production and employment support, (ii) education and vocational training assistance, (iii) provision of cadres for poor districts, and (iv) infrastructure investment in villages, communes, and districts. One of the key support measures for households is the 8 provision of microcredit and cash subsidy to increase agricultural productivity and facilitate the transition to non-farm employment. Households can receive one-time support for purchasing seeds and fertilizers that encourage them to cultivate high-value crops and livestock. At the village and commune levels, the program invests in basic infrastructure, including electricity, irrigation, markets, roads, schools, and health care facilities. 3. Data and descriptive analysis 3.1. Data sources Our main data source is the Vietnam Household Living Standard Surveys (VHLSSs) spanning over 16 years from 2004 to 2020. The VHLSSs are conducted biennially since 2002 by the General Statistics Office of Vietnam (GSO) in collaboration with the World Bank. The VHLSSs cover around 45,000 households from around 3,000 enumeration areas and provide detailed socio-economic data on households and their members. One key advantage of the VHLSSs is their comprehensive coverage, including all districts in the country with the exception of a few islands. Thus, these surveys cover all the districts that participate in the 30A Program, as well as districts with a poverty rate close to the threshold of 50% in the 2006. The VHLSSs are representative at the provincial level. We focus on the rural sample, since the rural population accounted for 98% of the total population in the 30A districts in 2008. Moreover, we limit the analysis to households living in districts with a poverty rate greater than 40% in 2006, such that the control group comprises districts with a poverty rate ranging between 40% and 50% in 2006. There are 65 control districts, which is approximately equivalent to the number of treatment districts. Consequently, our final sample includes a total of 127 districts (62 program districts and 65 control districts). We also conduct various robustness checks using different bandwidths, resulting in varying numbers of districts in the analysis. 9 The VHLSSs provide a wealth of information at both the household and individual levels. Household-level data includes details on assets, production, income, housing conditions, and participation in government programs. Meanwhile, individual-level data encompass demographics, education, and employment information. Additionally, the VHLSSs offer insights into the basic characteristics and infrastructure of the communes and villages where the sampled households reside. In addition to the VHLSSs, we utilize two other data sources to measure local communities’ development. The first source is nighttime light data. Nighttime light intensity has been widely used as a reliable proxy for GDP (e.g., Henderson et al., 2011; Hu and Yao, 2022). One advantage of using nighttime light data is its extensive temporal and spatial coverage, making it less susceptible to sampling errors compared to household surveys. We obtain nighttime light data from the Defense Meteorological Satellite Program (DMSP)/Operational Linescan System (OLS) and the Visible Infrared Imaging Radiometer Suite (VIIRS) on the Suomi National Polar-orbiting Partnership satellite. The annual nighttime light intensity is measured at the district level during the period 2001- 2017. The second data source is firm data at the district level. We compute the number of firms and their density using the annual Vietnam Enterprise Censuses during 2000- 2017, which were conducted by Viet Nam’s GSO. This data source provides indicators of wage and nonfarm employment opportunities at the district level. 3.2. Descriptive analysis We next present descriptive statistics on the main outcome variables of households in the treatment (program) districts participating in the 30A Programs (i.e., districts with a poverty rate greater than 50% in 2006) and the control districts (i.e., districts without the 30A Programs but with a poverty rate above 40%). 10 Figure 3 shows that both the treatment and control districts experienced an upward trend in real per capita income over time. In 2020, the per capita income for the treatment group was 21.9 million VND, while for the control group it was 25.5 million VND. While the ratio of per capita income between the control group and the treatment group decreased from 1.48 in 2004 to 1.16 in 2020, Panel A of Figure 3 does not reveal a significant change in the income gap between the two groups before and after the implementation of the program. Thus, it is not clear from this figure whether the 30A Program has improved the income of the treatment districts. To ensure comparability of poverty estimates over time, we use a common income poverty line which was set by the government for rural households during the 2016-2020 period. Specifically, we use an income poverty line of VND 8.4 million per year for 2020 and adjust this poverty line for other years using the overall Consumer Price Index (provided by GSO). Panel B of Figure 3 indicates a significant decrease in poverty for both the treatment and control groups. In 2020, the poverty rate for the treatment group was 16.8%, while for the control group it was 10.7%. Notably, the rural poverty rates for the Northern midlands and mountain areas, North Central and Central coastal areas, and Central Highlands in the same year were 11.2%, 3.0%, and 4.8%, respectively. These regional poverty rates were considerably lower than the average poverty rate observed in the 62 poorest districts. Hence, it can be concluded that a key objective of the 30A Program to reduce the poverty rate of the program districts to the average poverty rate of the regions in 2020 was not achieved. In addition to income and poverty, we use a wide range of outcome variables to measure the effects of various components of the 30A Program according to the conceptual framework presented in Figure 2. Table 1 (upper half) shows that the treatment group exhibits 11 a higher rate of working individuals, mainly in farm self-employment, compared to the control group. The proportion of individuals with wage jobs is slightly higher in the control group than in the treatment group, but the difference is not statistically significant. Among wage workers, the average monthly wage in the treatment group surpasses that of the control group. Table 1 (bottom half) presents a summary of employment variables for individuals and households' per capita income from different sources in the treatment and control groups. Additionally, Table A.2 (Appendix A) provides summary statistics for other outcome variables, including health insurance and health care utilization of individuals, education of children, educational subsidies for students, and access to loans from formal sources. 4. Empirical method We use the sharp regression discontinuity design (RDD) combined with district fixed effects model for analysis. Districts were selected into the 30A Program if their poverty rate in 2006 was larger than 50%. The running variable utilized in our analysis is the MOLISA poverty rate from 2006, employing a cutoff threshold of 50%. The sharp RDD equation can be written as follows ,, = 0 + 1 + 2 � − 50� +3 � − 50�. + 2 ,, + + + ,,, (1) where ,, denotes an outcome of interest of individual (or household) i in district j in year t. The variable equals 1 for districts covered in the 30A Program and 0 otherwise, reflecting the fact that the treatment districts had a poverty rate larger than 50% in 2006. The variable represents the 2006 poverty rate measured in percentage. To facilitate analysis, we center the poverty rate at 50% since it serves as the cutoff point. The local effects of the 30A Program are estimated by 1. 12 Equation (1) includes interaction between ( − 50) and to allow the different slopes of the outcome equation of the treatment and control group. The control variables, which are denoted by ,, , include both individual-level and household-level variables. We include the district fixed effects and the year fixed effects to capture the unobservables that commonly affect districts and years as a whole. The household-level (or individual-level) error terms are denoted by ,, . We examine the impact of the program on a comprehensive set of outcome variables encompassing both individual-level and household-level variables in different areas ranging from employment, education, health and poverty to household access to credits and services. The lists of these outcome variables are shown in Table 1 and Table A.2 in the Appendix. Regarding control variables, they should not be influenced by the treatment variable under investigation (Heckman and Vytlacil, 2007; Angrist and Pischke, 2008). As the program may impact various outcomes of individuals, we control for a limited set of exogenous variables including age, gender, and ethnicity. Additionally, we conduct robustness checks by including additional control variables such as households’ educational levels and demographic characteristics. As will be shown in the empirical results section, the inclusion of additional variables does not alter the results, thereby affirming the robustness of our findings. For interpretation, we primarily rely on the results derived from the parsimonious model specifications. The RDD model relies on the crucial assumption of continuity in the running variable around the cutoff point. The 30A Program, as per Resolution 30a/2008/NQ-CP by the Government of Viet Nam, was approved in December 2008, with the 2006 poverty rate serving as the cutoff variable. Thus, districts were unlikely to manipulate the poverty rate to be selected into the program. Nonetheless, it is useful to conduct a density test on the data 13 sample. We present the density graph of districts based on the 2006 poverty rate in Figure 4, which shows no discernible spike or jump immediately after the poverty rate of 50%. To further assess the possibility of manipulation, we employ the McCrary manipulation test (McCrary, 2008). 5 The test statistic is equal to 1.52 with a corresponding p-value of 0.13. It does not show statistical evidence of systematic manipulation occurring around the threshold of 50%. The RDD model relies on the assumption that the cutoff point can serve as a surrogate for a randomized treatment, ensuring similarity among individuals around that point. However, a potential challenge in our study is that the number of treatment districts is relatively low, at 62. Consequently, we perform falsification (placebo) tests to examine the balance between the treatment and control groups around the cutoff. We utilize the VHLSS data collected before the program (specifically, VHLSSs 2004, 2006, and 2008) to estimate equation (1) and examine whether significant differences exist between the treatment and control groups in various outcomes at the cutoff. Our analysis incorporates individual-level and household-level outcomes. The results of these regressions are reported in Tables A.3 to A.5 in the Appendix. Notably, 1 is statistically significant in multiple regressions, suggesting that households in the treatment districts and those in the control districts differ significantly in these outcome variables even in the absence of the program. Thus, we employ the combined RDD and difference-in-differences (DD) model to account for the differences observed in the absence of the program. The RDD-DD equation is fully written as follows ,, = 0 + 1 . + 2 � − 50� + 3 � − 50�. 5 We implement this test using the user-written 'rddensity' Stata code (Cattaneo et al., 2018). 14 +4 � − 50�. + 5 � − 50�. . +6 ,, + + + ,, , (2) where is a dummy variable, which equals 1 for years after 2009 and 0 otherwise. The local effect of the 30A Program is estimated by 1 . We take advantage of the panel data at the district level to control for district fixed effects in equation (2). Thus, the time invariant variables (variables without subscript t in equation 2) are already controlled in the time- invariant variables, , and as a result they are dropped from equation (2). Equation (2) can be re-written as follows ,, = 0 + 1 . + 4 � − 50�. +5 � − 50�. . + 6 ,, + + + ,, . (3) The district fixed-effect RDD-DD model specified in equation (3) serves as our final model for estimating the impact of the 30A Program. We use this model specification to estimate the program impact on all the outcome variables. In terms of standard errors, it is common practice to cluster them at the level of the running variable, which in our case is the district level (Lee and Lemieux, 2010). Additionally, we cluster the standard errors at the sampling unit of VHLSSs, which is the village level. We employ the multiway clustering technique introduced by Cameron et al. (2011), allowing us to cluster standard errors simultaneously at both the district and year- village level. However, it is worth noting that recent research by Kolesár and Rothe (2018) indicates that confidence intervals based on clustering standard errors according to the running variable may have inadequate coverage properties. We conduct additional analysis by employing alternative clustering approaches for robustness checks; our estimates remain robust across various clustering methods. 15 5. Estimation results 5.1. Program impact We start first with examining the local impact of the 30A Program by plotting individual- level outcomes based on the 2006 poverty rate of the districts where people reside. In Figures 5 and 6, we present RDD graphs both before and after the program's initiation in 2009. The graphs demonstrate that the disparity in self-employed farm work between the treatment and control groups was reduced after the program. Moreover, the treatment group experienced an increasing trend in self-employed farm work as well as wage jobs, compared to the control group. In Figure 6, we find that the treatment group received higher education subsidies than the control group, particularly after the program's implementation. Using equation (3) we further estimate the effects of the 30A Program on various outcomes for households and individuals. The direct outcomes of the 30A Program are employment. We start with the program impact on employment of individuals aged 15 and older in Table 2. 6 Table 2 shows that there are no significant program effects on the likelihood of working (column 1). However, the program has significant effects on individuals’ probabilities of being self-employed in farm and nonfarm. Specifically, the program decreases the probability of self-employment in farm work by 0.11, while increasing the probability of self-employment in nonfarm work by 0.054 (columns 2 and 3). This suggests that the program encourages individuals to transition from self-employed farm work to self- employed nonfarm work. The program impact on the likelihood of having a wage job is positive, although it is not statistically significant (column 4). Similarly, the program effects on monthly wages of 6 According to Viet Nam’s Labor Code, the minimum age for working is 15. We also try to use the sample of individuals aged 15-64 (Table A.6 in the Appendix), and the program impact estimates are quite similar to those derived from the sample of individuals aged 15 and older. 16 wage workers are positive but not statistically significant at the conventional level (column 5). However, when we expand the analysis to include all individuals (including those without a wage job, who are considered to have zero wages), we observe positive program effects on monthly wages that are marginally statistically significant (column 6). In Table 3, we analyze the program impact on access to public services, including health insurance and health care utilization, and education achievement and subsidies (for children ages 6-17). The results indicate that the program does not have a significant effect on health insurance (column 1), possibly because a high proportion of individuals in the program districts already had health insurance prior to the program. In 2008, 90% of people in the program districts had health insurance, and this figure increased to 98% in 2020 (Appendix A, Table A.2). Health care utilization is measured by the number of annual visits for both inpatient and outpatient treatments. Interestingly, the program has positive effects on the annual number of health care visits (column 2), which appears primarily driven by the increased utilization of outpatient health care services (column 3) rather than inpatient health care (column 4). This finding is reasonable, as individuals with serious health conditions typically require inpatient health care services regardless of their participation in the program. Regarding education, Table 3 shows no significant program effects on the number of completed schooling years or school enrollment among children ages 6-17 (columns 5 and 6). We also estimate program impact on school enrollment separately for primary-school-age children and secondary-school-age children, but we do not find any significant program effects (not shown). However, we observe a strong and statistically significant program effect on the educational subsidy received by students (column 7). This suggests that despite the 17 increased student subsidies, the program is not successful in attracting more children to attend school. Figure 7 illustrates the 30A Program’s local impact on per capita income and per capita public transfers, both before and after its initiation in 2009. The disparity in per capita nonfarm income between the treatment and control groups at the local poverty rate of 50% decreased after the program. Conversely, the treatment group initially had a higher per capita farm income than the control group, but this trend reversed after the program. The treatment groups, however, receive more public cash transfer after the program. Providing more rigorous analysis using the RDD-DD model, Table 4 shows that the program does not have statistically significant effects on per capita income or wage income (columns 1 and 2). However, we do observe significant program effects on per capita income from different sources. Specifically, the program leads to a decrease in farm income (column 4) but an increase in self-employed nonfarm income (column 3). This finding aligns with our discussion above that the program motivates individuals to transition from self-employed farm work to self-employed nonfarm work. It is also consistent with similar positive effects of large-scale program interventions that move workers out of agriculture to non-agricultural work documented for India (Asher and Novosad, 2020; Blakeslee et al., 2022; Chaurey and Le, 2022). Table 4 also shows that the program has positive and statistically significant effects on public cash transfers and income from other sources (columns 6 and 7). According to the Government of Vietnam (2021), the average annual agricultural subsidy provided to households is approximately VND 300 million per commune (equivalent to around USD 15,000). This program component may be a significant factor contributing to the positive program impact on public cash transfers and other incomes. Regarding wage income and 18 private transfers received by households during the past 12 months, we find a positive but not statistically significant program effect (column 5). In Table 5, we examine the program impact on per capita income of poor households and lower-income households, which are in the bottom 20% and 40% of the income distribution. As discussed earlier, poverty is defined using the income poverty line in 2020. This provides evidence on whether the program can successfully achieve its objectives of increasing income and reducing poverty. Table 5 shows that the program does not have statistically significant effects on per capita income of poor and lower-income households. In Appendix Table A.7, we further estimate the program effects on the shares of income from different sources in the total household income. Consistent with the findings in Table 4, the program has negative effects on the share of farm income and positive effects on the share of nonfarm income, as well as the share of private and public transfers and income from other sources. It is important to highlight that for poor households, farm income continues to constitute a significant portion of their total income compared to self-employed nonfarm income. Despite the program's positive impact on nonfarm income and other income sources, these increases do not fully compensate for the decrease in farm income. As a result, the program does not yield any significant effects on the per capita income of households. One challenge in evaluating the impact of a poverty reduction program in Viet Nam is the potential contamination from other targeted programs that can affect both the treatment and control groups. As discussed earlier in Section 2, the National Targeted Programme on Sustainable Poverty Reduction comprises the 30A Program and the 135 Program. It is possible that we may underestimate the 30A Program effects because some households in the control group might also have received support from the 135 Program. To examine this issue, we exclude households residing in the 135 communes from the control group and re- 19 estimate the 30A Program impact. The results are presented in Tables A.8 to A.11 in Appendix A. Overall, the estimated program effects on both individual-level and household- level outcomes remain similar to those reported in Tables 2 to 4. These findings suggest that our estimates are not biased by the 135 Program effects. 5.2. Further robustness analysis We conduct a number of additional robustness checks that support our estimation results. First, we perform several falsification (placebo) tests. In the first test, we examine the placebo program effects using the pre-program data, specifically the VHLSSs from 2004 to 2008. We consider 2008 as the post-program year and 2004 and 2006 as the pre-program years. We apply equation (3) to estimate the placebo effect on both individual-level and household- level outcomes in 2008. The results, presented in Tables A.12 and A.13 in the Appendix, indicate that the variable 'Program*Year 2008' is not statistically significant in nearly all regressions. It is only statistically significant at the 10% level in the regressions of 'wage job'. 7 In the second test, we exclude the 30A Program districts and allocate the control districts to a placebo treatment using cutoff points derived from the 2006 MOLISA poverty rate. Subsequently, we employ the same model specifications to estimate the impact of these placebo cutoffs on individual and household outcomes. Tables A.14 and A.15 report the effects on the individual-level outcomes at placebo cutoff points at 30% and 40%. It shows that there is only a marginally statistically significant effect of the cutoff point at 40% on health care utilization at the 10% level. We also conduct similar analysis for the household- 7 We also estimate the placebo effects when using 2006 and 2008 as the post-program years and 2004 as the pre-program year. The results are very similar, demonstrating that the placebo program has no statistically significant effects on almost all the outcome variables. 20 level outcomes and do not find any statistically significant effects of the two placebo cutoff points. In the third test, we assess the program impact on exogenous variables including individuals' age, gender, ethnicity, and education completion. For the education variables, we restrict the sample to individuals ages 30 or above to ensure that their education was not affected by the program. We use the same model as described in equation (3), utilizing both the pre-program and post-program data from the VHLSSs from 2004 to 2020. We do not find significant impact of the program on these exogenous variables, suggesting that there is good balance in the exogenous variables around the cutoff point (Appendix A, Table 16). Second, we examine the sensitivity of the estimated program effects when using different bandwidths. Initially, we expand the bandwidth to include districts with the 2006 poverty rate above 35%. Subsequently, we use narrower bandwidths by limiting the sample to households in districts where the poverty rate ranged from 40% to 60% in 2006. There are no available estimators for combining nonparametric RDD with fixed effects. However, we can apply nonparametric RDD to the post-treatment data to find the optimal bandwidth. We select optimal bandwidths using data-driven mean-squared errors (Calonico et al., 2017), which yields an average optimal bandwidth of approximately 7%. 8 Using this bandwidth, we estimate the program impact and present the results in Tables A.17 to A.19 in the Appendix. Due to space constraints, we only provide robustness analysis for several key variables of individual and household outcomes. The estimated program effects on individual outcomes, using different bandwidths, closely align with the main results presented in Tables 2 and 3. As another check, Table A.20 in the Appendix illustrates the 'donut' RDD, wherein we exclude districts that are very close to the cutoff point (2006 poverty rate of 50%) to assess 8 To find these optimal bandwidths, we use the 'rdrobust' command developed by Calonico et al. (2017). 21 whether the effects are sensitive to the sample around this point (Cattaneo et al., 2019). The results remain largely consistent with those displayed in Tables 2 and 3. Third, we explore the sensitivity of the results to different model specifications. In Appendix A, Table A.21, we present the RDD-DD regression (as specified in equation 2) without accounting for district fixed effects. Table A.22 further shows the district fixed-effect regression without any covariates, and Table A.23 shows the district fixed-effect regression with additional covariates. The additional control variables include dummy variables for education levels, household size, the proportion of children in households, and the proportion of older members in households. Remarkably, all the estimates exhibit minimal changes when compared to the main specification model in Tables 2 and 3. Fourth, we incorporate region-specific time trends in the model to account for potential variations in outcome trends across different regions. The results, presented in Appendix Table A.24, closely resemble the main findings. It is important to note that we use region-specific time trends for robustness checks rather than using them as the main specifications, since controlling for these time trends might absorb the program effects and bias these effects (Wolfers, 2006; Baum-Snow and Lutz, 2011). Fifth, we assess the robustness of the results to different ways of clustering standard errors. As noted, according to Kolesár and Rothe (2018), confidence intervals based on clustering standard errors by the running variable may exhibit poor coverage properties. In Table A.25 and A.26 in the Appendix, we employ one-way clustering at the village level and the traditional Eicker-Huber-White (EHW) heteroskedasticity-robust standard error. Overall, the results closely align with the main interpretation results (reported in Tables 2 and 3). Sixth, we investigate whether there are spillover program effects on nearby districts. Spillover effects can contaminate the control group and introduce bias into the impact 22 estimate. To estimate the spillover effects, we exclude the program district and consider control districts located in provinces with program districts as the ‘treatment’ districts. Appendix Table A.27 shows that the program effects are not statistically significant across all outcomes, indicating absence of spillover effects on nearby districts' outcomes. Seventh, regarding standard errors, multiple testing issues can exist when analyzing multiple dependent variables. Traditional estimation provides p-values for each estimate, representing the rate of false positives among all the results. Alternatively, we calculate q- values, which represent the false positive rate among significant results. Figures A.1 and A.2 in the Appendix display graphs of the p-values and q-values (estimated using the Simes method (1986)) for the program effects on all the outcomes. These figures reveal that the estimated effects on self-employed non-farm work, health care utilization, education subsidy, nonfarm income, and public cash transfers maintain significance at the conventional levels. Finally, we also conduct similar robustness analyses for household-level outcomes. The corresponding results are presented in Appendix Tables A.28 to A.38, which demonstrate that the estimated program effects on household-level outcomes remain similar to those reported in Table 4. 5.3. Potential mechanisms We next explore several mechanisms through which the 30A Program could increase nonfarm employment and health care utilization. Poor households often face liquidity constraints that hinder their investment in nonfarm businesses. A key policy of the 30A Program is to provide microcredit and support in accessing loans from formal sources. The role of credit in promoting nonfarm employment and household production has been well- documented in studies such as van Rooyen et al. (2012), Augsburg et al. (2015), Ksoll et al. 23 (2016), and Tria et al. (2022). In Viet Nam, Nguyen (2008) and Lensink and Pham (2012) find evidence of microcredit's poverty-reducing effects. Table 6 presents our estimation of the program impact on households' loans from various sources. The results demonstrate program positive effects on borrowing from formal sources and microcredit. Specifically, the program increases the probability of obtaining formal loans and microcredit by 0.13 and 0.09, respectively. However, the program effects on the loan size are not statistically significant. When considering the sample of both borrowers and non-borrowers, we find positive program effects on loan size. 9 Our findings align with Thanh et al. (2019), who find positive impact of microcredit on self-employed nonfarm employment in Viet Nam. Table 7 estimates the impact of the 30A Program on infrastructure using commune- level data and the same model as in equation (3). The VHLSSs provide information on whether communes have basic infrastructure, and we select outcome variables that correspond to the list of infrastructures provided by the 30A Program in the program areas. Our findings indicate marginally statistically significant but positive program effects on the availability of all-weather roads in the villages (column 1). Specifically, an all-weather road is defined as one that remains passable throughout the year. These positive effects on rural roads could stimulate local market development (Mu and van de Walle, 2011), which might in turns serve as a mechanism through which the project promotes nonfarm employment and income. Indeed, studies on other developing countries such as Bangladesh and India also find that rural roads help local people to find more nonfarm self-employment opportunities (Khandker et al., 2009; Asher and Novosad, 2020). 9 We address zero loan values by adding 1 before taking the natural logarithm of the loan values. 24 We do not observe any significant program effects on the availability of periodic markets and irrigation systems in the villages (columns 2 and 4). However, we find positive program effects on the availability of electricity grids (column 3). With respect to schools, the program does not yield statistically significant effects on kindergartens and primary schools (columns 5 and 6), possibly because most communes already have these facilities in place, but it has positive and statistically significant effects on the availability of secondary schools in communes (column 7). Regarding health facilities, we find positive program effects on availability of commune health care centers (column 8). This finding could explain why the program has positive program effects on health care utilization, although its effects on health insurance are not significant. We also examine the program impact on migration inflows and outflows, as measured by the percentages of in-migrants and out-migrants over the commune population over the past 12 months. However, the results show that the program has no significant effects on the migration flows (columns 9 and 10). In Table 8, we investigate whether the 30A Program effectively encourages the establishment of firms. To examine this, we use data from the annual Vietnam Enterprise Censuses from 2000 to 2017 and compute the density of firms of different sizes (micro, small, medium, and large) at the district level. Table 8, however, shows that the program does not yield any significant effects on firm density (columns 1 to 5). The concentration of firms is primarily influenced by other factors, such as local infrastructure and the human capital of the population. The lack of a significant effect on firm density partially explains why the program does not have notable impact on local job opportunities and wages. Finally, we evaluate the program impact on nighttime light intensity, which serves as a proxy for local economic development, but we find no statistically significant program 25 effects (column 6). This finding aligns with our previous results, which show that there are no program effects on household income and poverty levels. 5.4. Heterogenous impact Due to variations in individual characteristics, the impact of the 30A Program may differ among people. To investigate this issue, we interact the program treatment variable (Program*Post-program period) and several individual characteristics. Table 9 focuses on the heterogeneity in the program impact on nonfarm employment. The table shows that the program has more pronounced effects on males (column 1) and individuals with higher levels of education (column 3) compared to females and those with lower education levels. Moreover, the program impact appears to be lower for ethnic minorities compared to the Kinh group (column 2). We observe a positive and statistically significant interaction between Program*Post-program period and the dummy variable indicating VHLSSs conducted since 2006 (column 4). This suggests that the program impact on nonfarm employment tends to be higher in the longer term. Lastly, our analysis does not identify any heterogeneous program impact across communes with different levels of infrastructure (columns 5 to 8). 6. Conclusions Poverty in Viet Nam exhibits a strong correlation with geographic areas, with a higher concentration of poverty observed in mountainous and midland regions where ethnic minorities are more prevalent. In an effort to alleviate poverty, the Government of Viet Nam has initiated the 30A Program since 2009, targeting the poorest districts with a poverty rate exceeding 50% in 2006. This cutoff point allows us to use a sharp Regression Discontinuity 26 Design method, combined with a difference-in-difference model with district fixed effects, to estimate the impact of the program on households’ welfare. Overall, we do not find significant effects of the 30A Program on per capita income and poverty levels among households. However, we observe a shift from farm self- employment to nonfarm self-employment as a result of the program, leading to an increase in nonfarm income and a decrease in farm income. We find heterogeneous program impact on nonfarm employment across gender, education level, and ethnicity. We also find larger program effects on nonfarm employment in the longer term compared to the shorter term. Additionally, the program enhances access to microcredit and public services, including public transfers, health care utilization, and educational subsidies for students. Our study offers several policy implications. First, the finding that the 30A Program promotes nonfarm employment and improves access to credit and public services implies that a multifaceted poverty alleviation approach remains important for the poor areas. Second, in the poorest districts, nonfarm self-employment is relatively limited compared to farm and wage incomes. However, agricultural productivity tends to be low in these areas. Therefore, increasing nonfarm employment and wage job opportunities becomes vital for income growth and poverty reduction in the poor areas. Third, our analysis also demonstrates heterogeneous impacts of the program on nonfarm employment across different population subgroups, including gender, ethnicity and education levels. Hence, poverty reduction programs and measures should be tailored to specific areas and population subgroups. Finally, evaluating the impact of poverty reduction programs in Viet Nam faces challenges due to potential contamination from other targeted initiatives. Both the central and local governments can allocate investments to the other poor districts that are not included in the 30A Program. As a result, our analysis may underestimate the impact of the 30A Program. 27 Without access to data on public finance and other investments received by districts, we are unable to assess this estimation bias. But it is a promising topic for future research. 28 References Angrist, J. D., & Pischke, J. (2009). Mostly Harmless Econometrics: An Empiricist’s Companion. New Jersey: Princeton University Press. Asher, S., & Novosad, P. (2020). Rural roads and local economic development. American Economic Review, 110(3), 797-823. Augsburg, B., De Haas, R., Harmgart, H., & Meghir, C. (2015). The impacts of microcredit: Evidence from Bosnia and Herzegovina. American Economic Journal: Applied Economics, 7(1), 183-203. Bahal, G. (2020). Estimating the impact of welfare programs on agricultural output: evidence from India. American Journal of Agricultural Economics, 102(3), 982-998. Balboni, C., Bandiera, O., Burgess, R., Ghatak, M., & Heil, A. (2022). Why do people stay poor?. Quarterly Journal of Economics, 137(2), 785-844. Bandiera, O., Burgess, R., Das, N., Gulesci, S., Rasul, I., & Sulaiman, M. (2017). Labor markets and poverty in village economies. Quarterly Journal of Economics, 132(2), 811- 870. Banerjee, A., Duflo, E., Goldberg, N., Karlan, D., Osei, R., Parienté, W., ... & Udry, C. (2015). A multifaceted program causes lasting progress for the very poor: Evidence from six countries. Science, 348(6236), 1260799. Banerjee, Abhijit, Esther Duflo, and Garima Sharma. 2021. "Long-Term Effects of the Targeting the Ultra Poor Program." American Economic Review: Insights, 3 (4): 471-86. Baum-Snow, N., & Lutz, B. F. (2011). School desegregation, school choice, and changes in residential location patterns by race. American Economic Review, 101(7), 3019-46. Benjamin, D., Brandt, L., & McCaig, B. (2017). Growth with equity: income inequality in Vietnam, 2002–14. Journal of Economic Inequality, 15(1), 25-46. Blakeslee, D., Chaurey, R., Fishman, R., & Malik, S. (2022). Land rezoning and structural transformation in rural india: Evidence from the industrial areas program. World Bank Economic Review, 36(2), 488-513. Cattaneo, M. D., Idrobo, N., & Titiunik, R. (2019). A practical introduction to regression discontinuity designs: Foundations. Cambridge University Press. CEMA and UNDP (2009), Reviewing the past responding to new challenges: Mid-term review of programme 135-Ii, 2006 – 2008, Committe For Ethnic Minority Affairs and the United Nations in Vietnam. Available at: https://www.undp.org/sites/g/files/zskgke326/files/migration/vn/21361_P135_Report_E N.pdf Chaurey, R., & Le, D. T. (2022). Infrastructure maintenance and rural economic activity: Evidence from India. Journal of Public Economics, 214, 104725. 29 Dang, H. A. (2012). “Vietnam: A widening poverty gap for ethnic minorities. Indigenous Peoples, Poverty and Development”. In Hall, G. and Patrinos, H. (Eds.) Indigenous Peoples, Poverty and Development. Cambridge University Press. Dang, H. A., Dhongde, S., Do, M., Nguyen, C. V., & Pimhidzai, O. (2023). Rapid Economic Growth but Rising Poverty Segregation: Will Vietnam Meet the SDGs for Equitable Development? IZA Discussion Paper no. 15916. Government of Vietnam (2008), Resolution on the program to support rapid and sustainable poverty reduction for 61 poor districts, The government’s resolution No. 30a/2008/NQ- CP dated on 27/12/2008. Available at: https://thuvienphapluat.vn/van-ban/Van-hoa-Xa- hoi/Nghi-quyet-30a-2008-NQ-CP-chuong-trinh-ho-tro-giam-ngheo-nhanh-va-ben- vung-doi-voi-61-huyen-ngheo-83914.aspx Government of Vietnam (2012), Decision No. 1489/QĐ-TTg of the Government of Vietnam on National Target Program on Sustainable Poverty Reduction in the 2012-2015 period, the Government of Vietnam. Available at: https://chinhphu.vn/?pageid=27160&docid=163977&tagid=4&type=1 Government of Vietnam (2013), Decision 218/2013/NĐ-CP on detailed provisions and implementation guidelines of corporate income tax, the Government of Vietnam. Available at: https://thuvienphapluat.vn/van-ban/Doanh-nghiep/Nghi-dinh-218-2013- ND-CP-huong-dan-thi-hanh-Luat-thue-thu-nhap-doanh-nghiep-217811.aspx Government of Vietnam (2016), Decision No. 1722/QĐ-TTg of the Government of Vietnam on National Target Program on Sustainable Poverty Reduction in the 2016-2020 period, the Government of Vietnam. Available at: https://chinhphu.vn/?pageid=27160&docid=186132&tagid=4&type=1 Government of Vietnam (2021), Báo cáo đánh giá kết quả thực hiện chương trình mục tiêu quốc gia giảm nghèo và các chương trình mục tiêu quốc gia giai đoạn 2016-2020 (Report on the evaluation of the implementation results of the national target program on poverty reduction and the national target programs for the period 2016-2020), Government of Vietnam. Heckman, J. J., & Vytlacil, E. J. (2007). Econometric evaluation of social programs, part I: Causal models, structural models and econometric policy evaluation. Handbook of econometrics, 6, 4779-4874. Henderson, V., Storeygard, A., & Weil, D. N. (2011). A bright idea for measuring economic growth. American Economic Review, 101(3), 194-199. Hu, Y., & Yao, J. (2022). Illuminating economic growth. Journal of Econometrics, 228(2), 359-378. Jalan, J., & Ravallion, M. (2002). Geographic poverty traps? A micro model of consumption growth in rural China. Journal of Applied Econometrics, 17(4), 329-346. 30 Khandker, S. R., Bakht, Z., & Koolwal, G. B. (2009). The poverty impact of rural roads: Evidence from Bangladesh. Economic Development and Cultural Change, 57(4), 685- 722. Kraay, A., & McKenzie, D. (2014). Do poverty traps exist? Assessing the evidence. Journal of Economic Perspectives, 28(3), 127-148. Ksoll, C., Lilleør, H. B., Lønborg, J. H., & Rasmussen, O. D. (2016). Impact of village savings and loan associations: Evidence from a cluster randomized trial. Journal of Development Economics, 120, 70-85. Lanjouw, P., Marra, M., & Nguyen, C. (2017). Vietnam’s evolving poverty index map: Patterns and implications for policy. Social Indicators Research, 133(1), 93-118. Lensink, R. and Pham, T. T. T. 2012. The impact of microcredit on self-employment profits in Vietnam. Economics of Transition, 20 (1), 73–111. Meager, R. (2019). Understanding the average impact of microcredit expansions: A bayesian hierarchical analysis of seven randomized experiments. American Economic Journal: Applied Economics, 11(1), 57-91. Medeiros, V., Ribeiro, R. S. M., & do Amaral, P. V. M. (2021). Infrastructure and household poverty in Brazil: A regional approach using multilevel models. World Development, 137, 105118. Meng, L. (2013). Evaluating China's poverty alleviation program: a regression discontinuity approach. Journal of Public Economics, 101, 1-11. Millán, T. M., Macours, K., Maluccio, J. A., & Tejerina, L. (2020). Experimental long-term effects of early-childhood and school-age exposure to a conditional cash transfer program. Journal of Development Economics, 143, 102385. Ministry of Finance (MOF). (2015). “Nghị quyết giảm nghèo bền vững: Kết quả sau 6 năm” (Resolution to sustainably reduce poverty: Results after 6 years). Accessed on June 10, 2023. Available at: https://mof.gov.vn/webcenter/portal/btcvn/pages_r/l/tin-bo-tai- chinh?dDocName=BTC263844 Ministry of Labour, Invalids and Social Affairs of Viet Nam and United Nations Development Program. (2009). Mid-term impact evaluations of National Targeted Programs for Poverty Reduction during the period 2006–2010. Review report. MOLISA and UNDP. Mu, R., & Van de Walle, D. (2011). Rural roads and local market development in Vietnam. Journal of Development Studies, 47(5), 709-734. Neumark, D., & Simpson, H. (2015). Place-based policies. In Handbook of regional and urban economics (Vol. 5, pp. 1197-1287). Elsevier. Nguyen, C. (2008). Is a Governmental Micro-Credit Program for the Poor Really Pro-Poor? Evidence from Vietnam. Developing Economies, 46 (2), 151–187. 31 Nguyen, C. V., Tung, P. D., & Westbrook, D. (2015). Do the poorest ethnic minorities benefit from a large-scale poverty reduction program? Evidence from Vietnam. Quarterly Review of Economics and Finance, 56, 3-14. Phan, P. V., O’brien, M., Mendolia, S., & Paloyo, A. (2017). National pro-poor spending programmes and their effect on income inequality and poverty: Evidence from Vietnam. Applied Economics, 49(55), 5579-5590. Ravallion, M. (2007). Evaluating anti-poverty programs. Handbook of development economics, 4, 3787-3846. Swain, S. K. (2015). Impact of poverty alleviation programmes on socio-economic development of rural poor of Odisha-A statistical analysis. Journal of Rural Development, 34(2), 187-213. Tarozzi, A., Desai, J., & Johnson, K. (2015). The impacts of microcredit: Evidence from Ethiopia. American Economic Journal: Applied Economics, 7(1), 54-89. Thanh, P. T., Saito, K., & Duong, P. B. (2019). Impact of microcredit on rural household welfare and economic growth in Vietnam. Journal of Policy Modeling, 41(1), 120-139. Tria, D., Harun, M., & Alam, M. (2022). Microcredit as a strategy for employment creation: A systematic review of literature. Cogent Economics & Finance, 10(1), 2060552. Van Rooyen, C., Stewart, R., & De Wet, T. (2012). The impact of microfinance in sub- Saharan Africa: a systematic review of the evidence. World Development, 40(11), 2249- 2262. Wolfers, J. (2006). Did unilateral divorce laws raise divorce rates? A reconciliation and new results. American Economic Review, 96(5), 1802-1820. World Bank (2022). From the Last Mile to the Next Mile – 2022 Vietnam Poverty & Equity Assessment. Washington, DC: World Bank. World Bank (2023), World Development Indicators Data, Accessed on June 10, 2023. Available at: https://data.worldbank.org/indicator/DT.ODA.ALLD.CD?locations=VN 32 Figure 1. The geographic distribution of the treatment group Note: The treatment group consists of the 62 poorest districts with the 2006 poverty rate above 50%. In this figure these districts are presented in the brown color. Source: Authors’ preparation using the 2006 poverty rate of districts. 33 Figure 2. The 30A Program and the causal chain hypothesis The 30A Program Production and Investment in Policies on cadres Education support: employment infrastructure: for poor districts: - Improving quality support: - Rural road - Increasing and number of - Providing forestry - Schools provincial- and teachers land - Healthcare center district-level - Vocational - Agricultural - Irrigation cadres for training. subsidy - Market communes. - Capacity building - Microcredit. - Electricity grid - Attracting young for cadres and staff - Nonfarm business intellectuals. and enterprises Improved Better Improved market access governance and education and management higher capacity Higher agricultural productivity More non-farm employment. Improved public Increased income; services Reduced poverty Source: authors’ preparation 34 Figure 3. The poverty rate and per capita income of the districts with the 2006 poverty rate above 40% A. Per capita income of households in the treatment B. The poverty rate of households in the treatment and and control districts control districts (in percent) 30 90 80 Per capita income (million VND) 25 70 The poverty rate (%) 60 20 50 40 15 30 10 20 10 5 0 2004 2006 2008 2010 2012 2014 2016 2018 2020 2004 2006 2008 2010 2012 2014 2016 2018 2020 Year Year Treatment Control Treatment Control Note: This figure graphs the per capita income and poverty rate (%) of households in districts covered by the 30A Program compared to other districts during the 2004-2020 period. The per capita income values are adjusted to 2020 prices using the annual Consumer Price Index (CPI). The analysis focuses on households in districts where the poverty rate in 2006 was greater than 40%. The vertical lines indicate the year 2009, which marks the implementation of the 30A Program. 35 Figure 4. The density of the district poverty rate A. The kernel density of the MOLISA poverty rate B. RDD manipulation test using local polynomial of districts density estimation .03 .04 .03 .02 Density .02 .01 .01 0 0 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 The 2006 MOLISA poverty rate (%) The 2006 MOLISA poverty rate (%) kernel = epanechnikov, bandwidth = 4.5118 point estimate 95% C.I. Note: This graph shows density of districts by the 2006 poverty rate around the threshold of 50%. 36 Figure 5: RDD plot of employment A. The probability of having farm work before 2009 B. The probability of having farm work since 2009 1 1 The probability of having farm work The probability of having farm work .8 .8 .6 .6 .4 .4 .2 .2 0 20 30 40 50 60 70 80 20 25 30 35 40 45 50 55 60 65 70 75 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 C. The probability of having nonfarm work before 2009 D. The probability of having nonfarm work since 2009 .3 .3 The probability of having nonfarm work The probability of having nonfarm work .2 .2 .1 .1 0 0 20 30 40 50 60 70 80 20 25 30 35 40 45 50 55 60 65 70 75 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 E. The probability of having a wage job before 2009 F. The probability of having a wage job since 2009 .3 .3 The probability of having a wage job The probability of having a wage job .2 .2 .1 .1 0 0 20 30 40 50 60 70 80 20 25 30 35 40 45 50 55 60 65 70 75 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 Note: This graph shows the regression discontinuity plot of employment outcomes of individuals across the 2006 poverty rate of districts. The sample is limited to individuals living districts with the 2006 poverty rate above 40%. 37 Figure 6: RD plot of healthcare utilization and education subsidy for students A. Outpatient healthcare utilization before 2009 B. Outpatient healthcare utilization since 2009 5 5 Outpatient healthcare utilization Outpatient healthcare utilization 4 4 3 3 2 2 1 1 0 0 20 30 40 50 60 70 80 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 C. Inpatient healthcare utilization before 2009 D. Inpatient healthcare utilization since 2009 .8 .8 Inpatient healthcare utilization Inpatient healthcare utilization .6 .6 .4 .4 .2 .2 0 0 20 30 40 50 60 70 80 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 E. Log of education subsidy before 2009 F. Log of education subsidy since 2009 6 6 Log of education subsidy Log of education subsidy 4 4 2 2 0 0 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 Note: This graph shows the regression discontinuity plot of healthcare utilization of individuals and educational subsidy for students across the 2006 poverty rate of districts. The sample is limited to individuals living districts with the 2006 poverty rate above 40%. 38 Figure 7: RD plot of households’ per capita income A. Log of per capita nonfarm income before 2009 B. Log of per capita nonfarm income since 2009 10 4 Log of per capita nonfarm income Log of per capita nonfarm income 8 3 6 2 4 1 2 0 0 20 30 40 50 60 70 80 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 C. Log of per capita farm income before 2009 D. Log of per capita farm income since 2009 10 12 Log of per capita farm income Log of per capita farm income 10 8 8 6 6 4 4 2 2 0 20 30 40 50 60 70 80 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 E. Log of per capita public cash transfers before 2009 F. Log of per capita public cash transfers since 2009 .4 .4 Log of per capita public cash transfers Log of per capita public cash transfers .3 .3 .2 .2 .1 .1 0 0 20 30 40 50 60 70 80 20 30 40 50 60 70 80 The 2006 poverty rate of districts (%) The 2006 poverty rate of districts (%) Sample average within bin Polynomial fit of order 2 Sample average within bin Polynomial fit of order 2 Note: This graph shows the regression discontinuity plot of households’ per capita income from different sources across the 2006 poverty rate of districts. The sample is limited to households living districts with the 2006 poverty rate above 40%. 39 Table 1. The main outcome variables Treatment Control Variables Year Year Year Year Year Year Year Year Year Year 2004 2008 2012 2016 2020 2004 2008 2012 2016 2020 Individual employment Currently working (%) 88.98 88.96 91.19 90.29 91.54 83.47 82.50 86.73 86.89 87.37 (0.96) (0.90) (0.63) (0.63) (0.63) (0.69) (0.86) (0.82) (0.68) (0.61) Self-employed farm work (%) 81.89 80.43 65.55 62.79 48.49 68.62 66.29 54.14 53.67 42.78 (1.57) (1.55) (2.21) (2.27) (2.68) (1.58) (1.85) (1.67) (1.52) (1.88) Self-employed non-farm work 1.62 2.49 2.27 2.78 9.73 5.33 5.72 5.00 4.02 9.97 (%) (0.31) (0.57) (0.36) (0.45) (1.81) (0.70) (0.77) (0.62) (0.50) (1.22) Having wage job (%) 5.47 6.04 23.37 24.73 33.32 9.52 10.49 27.59 29.20 34.61 (0.79) (0.75) (2.14) (1.95) (2.31) (0.82) (0.89) (1.40) (1.37) (1.52) Monthly wage (thousand VND) 2.20 2.42 2.45 3.33 4.18 2.16 2.94 2.66 3.40 2.96 (0.12) (0.16) (0.14) (0.21) (0.24) (0.08) (0.11) (0.10) (0.14) (0.19) Per capita income from different sources (million VND) Per capita income 6.86 8.94 10.95 13.33 21.91 10.18 13.78 16.36 20.37 25.47 (0.28) (0.40) (0.48) (0.76) (1.28) (0.40) (0.45) (0.69) (0.80) (1.02) Per capita income from wages 1.23 1.96 3.56 4.65 10.46 1.89 2.62 5.19 6.70 9.78 (0.15) (0.20) (0.36) (0.42) (0.86) (0.19) (0.20) (0.37) (0.36) (0.51) Per capita income from nonfarm 0.39 0.56 0.73 1.26 2.42 1.19 1.62 1.74 1.71 2.97 production (0.06) (0.09) (0.15) (0.26) (0.61) (0.18) (0.22) (0.23) (0.21) (0.33) Per capita income from farm 4.69 5.42 5.73 6.09 6.83 5.60 7.05 7.89 9.51 10.06 production (0.19) (0.25) (0.22) (0.39) (0.57) (0.24) (0.28) (0.44) (0.58) (0.71) Per capita remittances from 0.22 0.44 0.33 0.66 1.46 0.68 1.17 0.79 1.48 1.89 private sources (0.02) (0.08) (0.05) (0.15) (0.30) (0.10) (0.14) (0.11) (0.17) (0.19) Per capita public cash transfers 0.08 0.26 0.28 0.33 0.30 0.13 0.29 0.43 0.43 0.38 (0.02) (0.06) (0.05) (0.07) (0.06) (0.02) (0.06) (0.06) (0.06) (0.05) Per capita income from other 0.25 0.30 0.31 0.34 0.44 0.69 1.03 0.33 0.53 0.40 sources (0.05) (0.05) (0.05) (0.04) (0.06) (0.09) (0.12) (0.05) (0.08) (0.05) The income poverty rate (%) 76.42 61.25 57.64 43.63 16.85 57.99 34.95 33.00 22.28 10.65 (2.45) (3.16) (2.57) (3.13) (2.09) (1.97) (2.16) (1.83) (1.80) (1.27) Note: The sample is limited to households residing in districts with a poverty rate exceeding 40% in 2006. Within this sample, households living in the 30A districts are referred to as the treatment group, while households residing in districts without the 30A Programs but having the 2006 poverty rate above 40% are considered the control group. The income variables are adjusted to the 2020 price using overall CPI. Standard errors of the means in parentheses. 40 Table 2. The district fixed-effect RDD-DD regressions of employment of individuals Dependent variables Currently Self- Self- Having wage Log of Log of working employed employed job (yes=1, monthly wage monthly wage Explanatory variables (yes=1, no=0) farm work non-farm no=0) (wage (all workers) (yes=1, no=0) work (yes=1, workers) no=0) (1) (2) (3) (4) (5) (6) Program * Post-program period -0.0277 -0.1144** 0.0538*** 0.0330 0.1285 0.5545* (0.0198) (0.0554) (0.0151) (0.0512) (0.2694) (0.2845) (Poverty rate – 50) * Post-program -0.0024 0.0070 -0.0056*** -0.0038 -0.0059 -0.0338 period (0.0024) (0.0057) (0.0018) (0.0043) (0.0335) (0.0269) Program * (Poverty rate – 50) * Post- 0.0047* -0.0026 0.0046** 0.0026 0.0041 -0.0002 program period (0.0025) (0.0062) (0.0018) (0.0049) (0.0348) (0.0300) Age 0.0387*** 0.0243*** 0.0040*** 0.0104*** 0.0444*** 0.1132*** (0.0014) (0.0013) (0.0005) (0.0006) (0.0060) (0.0066) Age squared -0.0005*** -0.0003*** -0.0000*** -0.0001*** -0.0006*** -0.0015*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) Male 0.0136*** -0.0653*** -0.0076** 0.0865*** 0.0910*** 0.9974*** (0.0046) (0.0072) (0.0030) (0.0051) (0.0197) (0.0567) Ethnic minorities (yes=1, Kinh=0) 0.0523*** 0.1805*** -0.0990*** -0.0292* -0.3732*** -0.2248 (0.0130) (0.0265) (0.0109) (0.0168) (0.0511) (0.1580) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.1697*** 0.2242*** 0.0402*** -0.0947*** 6.8517*** -0.6763*** (0.0330) (0.0343) (0.0101) (0.0190) (0.1321) (0.1886) Observations 128,446 128,446 128,446 128,446 27,510 128,446 R-squared 0.289 0.190 0.066 0.174 0.203 0.154 The mean value of the dependent 0.854 0.734 0.040 0.081 7.483 1.394 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 41 Table 3. The district fixed-effect RDD-DD regressions of individual’s health and children’s education Dependent variables Having The annual The annual The annual The number Currently Log of health number of number of number of of completed attending education insurance healthcare outpatient inpatient schooling school subsidy for Explanatory variables (yes=1, visits healthcare healthcare years (children students no=0) visits visits (children aged 6-17) (children aged 6-17) aged 6-17) (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0174 0.3757** 0.3814*** -0.0059 0.0422 -0.0080 1.2841*** (0.0995) (0.1483) (0.1371) (0.0371) (0.3331) (0.0448) (0.4805) (Poverty rate – 50) * Post-program 0.0162 0.0561** 0.0523** 0.0037 -0.0368 -0.0102** 0.0504 period (0.0118) (0.0229) (0.0221) (0.0040) (0.0396) (0.0051) (0.0546) Program * (Poverty rate – 50) * -0.0066 -0.0580** -0.0547** -0.0032 0.0385 0.0075 -0.0467 Post-program period (0.0111) (0.0226) (0.0218) (0.0038) (0.0381) (0.0049) (0.0517) Age -0.0020*** -0.0058 -0.0059 0.0000 1.1998*** 0.1571*** 0.3874*** (0.0007) (0.0039) (0.0038) (0.0006) (0.0410) (0.0093) (0.0473) Age squared 0.0000*** 0.0003*** 0.0002*** 0.0000*** -0.0212*** -0.0081*** -0.0197*** (0.0000) (0.0001) (0.0001) (0.0000) (0.0019) (0.0004) (0.0021) Male 0.0133*** -0.1495*** -0.1382*** -0.0111** 0.1212** 0.0370*** 0.1056*** (0.0029) (0.0263) (0.0253) (0.0054) (0.0599) (0.0105) (0.0325) Ethnic minorities (yes=1, Kinh=0) 0.0153 -0.2511** -0.2432** -0.0078 -1.1867*** -0.1293*** 0.7992*** (0.0211) (0.1012) (0.0992) (0.0114) (0.1724) (0.0258) (0.1114) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 0.2507*** 0.7249*** 0.6527*** 0.0723*** -5.4798*** 0.2662*** -1.1919*** (0.0211) (0.0801) (0.0780) (0.0122) (0.3037) (0.0530) (0.2757) Observations 187,734 86,980 86,980 86,980 50,645 50,645 50,645 R-squared 0.515 0.096 0.091 0.028 0.526 0.195 0.330 The mean value of the dependent 0.285 0.619 0.519 0.100 4.733 0.823 1.254 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of individual’s health and children’s education using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 42 Table 4. The district fixed-effect RDD-DD regressions of households’ income and poverty Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) -0.0181 0.4116 0.8039** -0.3774** 0.5492 0.8739*** 0.9973** Program * Post-program period (0.0869) (0.5071) (0.3499) (0.1852) (0.5740) (0.3109) (0.4132) (Poverty rate – 50) * Post- 0.0052 -0.0756* -0.0658* 0.0099 -0.0303 -0.0566* -0.0153 program period (0.0088) (0.0442) (0.0382) (0.0216) (0.0554) (0.0317) (0.0407) Program * (Poverty rate – 50) * -0.0092 0.0284 0.0511 -0.0052 -0.0064 0.0444 0.0215 Post-program period (0.0093) (0.0502) (0.0414) (0.0223) (0.0607) (0.0373) (0.0456) Age of household heads 0.0262*** 0.0485*** 0.0565*** 0.1330*** -0.0451*** -0.0931*** -0.0598*** (0.0021) (0.0135) (0.0081) (0.0128) (0.0090) (0.0099) (0.0105) -0.0002*** -0.0008*** -0.0007*** -0.0013*** 0.0006*** 0.0012*** 0.0007*** Age squared of household heads (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head 0.0053 -0.0124 0.2031*** 0.6791*** -0.4089*** -0.0837 -0.3800*** (male=1, female=0) (0.0181) (0.0987) (0.0626) (0.0753) (0.0502) (0.0556) (0.0663) -0.6139*** -0.2264 -1.7972*** 0.8630*** -0.5437*** 0.6164*** -0.7285*** Ethnic minorities (yes=1, Kinh=0) (0.0490) (0.1793) (0.1596) (0.1217) (0.1376) (0.1165) (0.1017) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.6202*** 3.0156*** 1.6227*** 3.7472*** 4.8958*** 1.9215*** 3.0930*** (0.0824) (0.3525) (0.2387) (0.3254) (0.2912) (0.2996) (0.2578) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.364 0.164 0.107 0.148 0.228 0.149 0.091 The mean value of the dependent 9.026 3.651 1.712 8.149 3.912 0.754 1.324 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 43 Table 5. The district fixed-effect RDD-DD regressions of income of bottom-income households and poverty Dependent variables Log of per Log of per Log of per Income poor Bottom 20% Bottom 40% capita income capita income capita income households households households (the sample of (the sample of (the sample of (using the (using the (using the the poor) the bottom 20% the bottom 40% constant bottom 20% bottom 40% Explanatory variables income income poverty line in income income households) households) 2020) threshold in threshold in 2020) 2020) (1) (2) (3) (4) (5) (6) 0.0528 0.0490 0.0347 -0.0463 -0.0528 0.0141 Program * Post-program period (0.0436) (0.0434) (0.0515) (0.0604) (0.0603) (0.0523) (Poverty rate – 50) * Post- -0.0040 -0.0036 0.0012 -0.0025 -0.0017 -0.0021 program period (0.0055) (0.0055) (0.0062) (0.0057) (0.0057) (0.0055) Program * (Poverty rate – 50) * 0.0033 0.0029 -0.0027 0.0054 0.0046 0.0057 Post-program period (0.0057) (0.0057) (0.0065) (0.0063) (0.0062) (0.0058) Age of household heads 0.0059*** 0.0059*** 0.0098*** -0.0143*** -0.0141*** -0.0148*** (0.0014) (0.0014) (0.0015) (0.0013) (0.0014) (0.0014) -0.0001*** -0.0001*** -0.0001*** 0.0001*** 0.0001*** 0.0001*** Age squared of household heads (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Gender of household head 0.0249** 0.0270*** 0.0299*** -0.0115 -0.0092 -0.0126 (male=1, female=0) (0.0096) (0.0097) (0.0087) (0.0092) (0.0085) (0.0111) -0.0853*** -0.0820*** -0.1585*** 0.2422*** 0.2403*** 0.3138*** Ethnic minorities (yes=1, Kinh=0) (0.0164) (0.0165) (0.0202) (0.0202) (0.0197) (0.0259) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 8.3926*** 8.3845*** 8.4786*** 0.8696*** 0.8577*** 0.9889*** (0.0421) (0.0415) (0.0495) (0.0442) (0.0447) (0.0468) Observations 16,857 15,389 24,891 40,468 40,468 40,468 R-squared 0.159 0.157 0.196 0.252 0.252 0.274 The mean value of the dependent 8.560 8.495 8.732 0.546 0.472 0.741 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 44 Table 6. Mechanism: the impact of the program on households’ access to credit Dependent variables Borrow from Borrow from Log of formal Log of Log of formal Log of formal sources microcredit loan (the microcredit loan (the microcredit (yes=1, no=0) sources sample of (the sample of sample all (the sample all Explanatory variables (yes=1, no=0) borrowing borrowing households) households) households) households) (1) (2) (3) (4) (5) (6) Program * Post-program period 0.1353** 0.0903* 0.0047 -0.1188 1.1861** 0.9366* (0.0657) (0.0509) (0.1413) (0.0966) (0.5540) (0.4879) (Poverty rate – 50) * Post-program -0.0146** -0.0126** 0.0062 0.0028 -0.1005* -0.1304** period (0.0066) (0.0057) (0.0161) (0.0121) (0.0571) (0.0545) Program * (Poverty rate – 50) * 0.0119* 0.0129** -0.0126 0.0019 0.0793 0.1346** Post-program period (0.0071) (0.0062) (0.0168) (0.0126) (0.0606) (0.0596) Age of household heads 0.0075*** 0.0029** 0.0236*** 0.0156*** 0.0697*** 0.0323*** (0.0017) (0.0012) (0.0058) (0.0050) (0.0127) (0.0121) -0.0001*** -0.0001*** -0.0002*** -0.0002*** -0.0010*** -0.0006*** Age squared of household heads (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0248** 0.0001 0.1504*** 0.0609** 0.2377** 0.0068 female=0) (0.0124) (0.0098) (0.0339) (0.0241) (0.0971) (0.0974) -0.0199 0.0662*** -0.5917*** -0.2038*** -0.4173** 0.6141*** Ethnic minorities (yes=1, Kinh=0) (0.0222) (0.0160) (0.0656) (0.0302) (0.1848) (0.1577) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.2244*** 0.0813* 9.1356*** 8.8530*** 2.1609*** 0.6479 (0.0459) (0.0411) (0.1432) (0.1182) (0.3486) (0.4011) Observations 30,556 40,468 11,978 11,382 30,556 40,468 R-squared 0.101 0.104 0.349 0.326 0.204 0.107 The mean value of the dependent 0.201 0.183 9.629 3.693 9.326 1.705 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of loans of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 45 Table 7. Mechanism: the impact of the program on commune outcomes Dependent variables Village has Commune There is Village has Commune Commune Commune Commune The The an all- has a electricity an has a has a has a has a proportion proportion weather whole and grid in the irrigation kindergarten primary secondary health of in- of out- road periodic village system school school center migrants migrants Explanatory variables (passable market (in percent) (in percent) for all the time) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Program * Post-program 0.1602* 0.0713 0.1399** -0.1320 0.1778 0.0030 0.1584** 0.1265** -0.2236 0.0811 period (0.0829) (0.0958) (0.0571) (0.1098) (0.1106) (0.0092) (0.0703) (0.0621) (0.2231) (0.2842) (Poverty rate – 50) * Post- 0.0018 -0.0049 -0.0106 0.0034 -0.0090 0.0003 -0.0141* -0.0030 0.0468 0.0310 program period (0.0115) (0.0116) (0.0080) (0.0103) (0.0143) (0.0010) (0.0072) (0.0076) (0.0367) (0.0483) Program * (Poverty rate – -0.0124 0.0024 0.0103 -0.0023 0.0095 -0.0004 0.0114 -0.0005 -0.0504 -0.0379 50) * Post-program period (0.0121) (0.0120) (0.0084) (0.0113) (0.0147) (0.0010) (0.0078) (0.0084) (0.0419) (0.0526) Communes in remote areas -0.0927*** -0.0735** -0.0286** -0.0197 -0.0242 -0.0001 -0.0060 0.0208 -0.6171 -0.7425* (0.0300) (0.0349) (0.0129) (0.0272) (0.0181) (0.0032) (0.0190) (0.0208) (0.3858) (0.3979) Log of commune area -0.0294 0.1595*** 0.0125 0.0244 0.0362** 0.0001 0.0664*** 0.0705*** -1.2451 -1.4580 (0.0248) (0.0452) (0.0114) (0.0284) (0.0180) (0.0022) (0.0174) (0.0249) (1.1062) (1.0997) Log of population density 0.0063 0.1645*** 0.0249* 0.0348 0.0447** 0.0036 0.0497*** 0.0584** -1.5236 -1.7573 of communes (0.0229) (0.0422) (0.0131) (0.0279) (0.0188) (0.0024) (0.0161) (0.0238) (1.2876) (1.2857) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.8560*** -0.9320** 0.7500*** 0.4928** 0.4406*** 0.9811*** 0.4075*** 0.5285*** 12.9126 14.9616 (0.1986) (0.3563) (0.1026) (0.2325) (0.1427) (0.0186) (0.1355) (0.1986) (10.3237) (10.2833) Observations 2,617 2,617 2,617 2,617 2,617 2,617 2,617 2,617 2,617 2,617 R-squared 0.227 0.339 0.284 0.189 0.437 0.053 0.279 0.162 0.176 0.165 The mean value of the dependent variables before 0.788 0.627 0.981 0.669 0.597 0.989 0.949 0.912 0.783 0.916 the program Note: This table reports district fixed-effect RDD-DD regressions of infrastructures of communes using commune-level observations. Robust standard errors in parentheses. Standard errors are clustered by district level. *** p<0.01, ** p<0.05, * p<0.1. 46 Table 8. Mechanism: the impact of the program on enterprise and economic activity Dependent variables The number of The number of The number of The number of The number of Log of firms per 100 micro firms (1- small firms medium firms large firms nighttime light square km 9 workers) per (10-50 (51-299 (300+ intensity Explanatory variables 100 square km workers) per workers) per workers) per 100 square km 100 square km 100 square km (1) (2) (3) (4) (5) (6) Program * Post-program period -0.7263 -0.6645 -0.4307 0.3238 0.0452 0.3178 (3.0742) (1.7593) (1.0410) (0.2868) (0.0809) (0.2501) (Poverty rate – 50) * Post- -0.6688 -0.3703 -0.1740 -0.0937 -0.0308 -0.0055 program period (0.9449) (0.5366) (0.2915) (0.0922) (0.0294) (0.0262) Program * (Poverty rate – 50) * 0.5014 0.2798 0.1230 0.0710 0.0277 0.0103 Post-program period (0.7820) (0.4410) (0.2448) (0.0762) (0.0258) (0.0290) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.4518 0.1390 0.2238 0.0707 0.0183 -2.6308*** (1.4249) (0.5715) (0.5626) (0.2691) (0.0394) (0.0786) Observations 2,625 2,625 2,625 2,625 2,625 2,279 R-squared 0.578 0.490 0.604 0.633 0.359 0.840 The mean value of the dependent 9.016 0.908 0.923 0.348 0.067 0.277 variables before the program Note: This table reports district fixed-effect RDD-DD regressions of density of firms and nighttime light data intensity of districts using district-level observations. The number and density of firms at the district level are computed using the annual Vietnam Enterprise Censuses from 2000 to 2017. Robust standard errors in parentheses. Standard errors are clustered by district level. *** p<0.01, ** p<0.05, * p<0.1. 47 Table 9. The heterogenous impact of the program on nonfarm employment of individuals Dependent variable is the dummy of having nonfarm work Explanatory variables (1) (2) (3) (4) (5) (6) (7) (8) Program * Post * Male 0.1027*** (0.0149) -0.0876** Program * Post * Ethnic minorities (0.0346) Program * Post * The number of 0.0086*** schooling years (0.0019) Program * Post * Dummy variable 0.0377* indicating VHLSSs since 2006 (0.0217) Program * Post * log of distance -0.0084 from village to the nearest town (0.0108) Program * Post * Village has all- 0.0128 weather roads (0.0257) Program * Post * log of distance -0.0167 from village to the nearest market (0.0110) Program * Post * log of distance -0.0076 from village to the nearest bank (0.0098) Program * Post-program period 0.0366 0.1603*** 0.0343 0.0677 0.1136* 0.0830 0.1338** 0.1153** (0.0512) (0.0518) (0.0566) (0.0520) (0.0603) (0.0529) (0.0601) (0.0569) (Poverty rate – 50) * Post-program -0.0095** -0.0095** -0.0099** -0.0094** -0.0094** -0.0094** -0.0094* -0.0098** period (0.0047) (0.0047) (0.0047) (0.0047) (0.0047) (0.0046) (0.0051) (0.0047) Program * (Poverty rate – 50) * 0.0072 0.0075 0.0082 0.0072 0.0073 0.0071 0.0066 0.0072 Post-program period (0.0052) (0.0052) (0.0053) (0.0052) (0.0052) (0.0052) (0.0056) (0.0052) Interacted variables Yes Yes Yes Yes Yes Yes Yes Yes Other control variables Yes Yes Yes Yes Yes Yes Yes Yes District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant -0.0460** -0.0619** -0.2006*** -0.0547** -0.0136 -0.0747*** -0.0406 -0.0218 (0.0233) (0.0238) (0.0249) (0.0235) (0.0301) (0.0235) (0.0258) (0.0273) Observations 128,446 128,446 128,446 128,446 128,446 128,446 103,971 128,446 R-squared 0.198 0.197 0.212 0.196 0.197 0.197 0.189 0.199 Note: This table reports district fixed-effect RDD-DD regressions of nonfarm employment of individuals using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 48 Appendix A: Additional Tables and Figures Figure A.1. P-value and q-value of estimates of the effects of the program on individual-level outcomes Currently attending school (aged 6-17) 0.859 0.934 Currently working (yes=1, no=0) 0.164 0.296 Having health insurance (yes=1, no=0) 0.862 0.934 Having wage job (yes=1, no=0) 0.521 0.703 Log of education subsidy for students (aged 6-17) 0.009 0.057 Log of monthly wage (all workers) 0.054 0.119 Log of monthly wage (wage workers) 0.634 0.815 Self-employed farm work (yes=1, no=0) 0.041 0.107 Self-employed non-farm work (yes=1, no=0) 0.001 0.014 The annual number of healthcare visits 0.013 0.068 The annual number of inpatient healthcare visits 0.873 0.934 The annual number of outpatient healthcare visits 0.006 0.057 The number of completed schooling years (aged 6-17) 0.899 0.934 0 .2 .4 .6 .8 1 Value P-value Q-value Note: The figure compares the p- and q-values of estimates of the program on individual-level outcomes. The p-value is estimated for the program effect reported in Tables 2 and 3, while the q-value is estimated using Simes’ (1986) method. 49 Figure A.2. P-value and q-value of estimates of the effects of the program on household-level outcomes Borrow from formal sources (yes=1, no=0) 0.041 0.107 Borrow from microcredit sources (yes=1, no=0) 0.079 0.152 Income poor households 0.445 0.633 Log of formal loan (the sample all households) 0.034 0.107 Log of formal loan (the sample of borrowing households) 0.974 0.974 Log of microcredit (the sample all households) 0.057 0.119 Log of microcredit (the sample of borrowing households) 0.221 0.373 Log of per capita income 0.835 0.934 Log of per capita income from farm production 0.044 0.107 Log of per capita income from nonfarm production 0.023 0.090 Log of per capita income from other sources 0.0170.077 Log of per capita income from wages 0.419 0.628 Log of per capita public cash transfers 0.006 0.057 Log of per capita remittances 0.340 0.541 0 .2 .4 .6 .8 1 Value P-value Q-value Note: The figure compares the p- and q-values of estimates of the program on household-level outcomes. The p-value is estimated for the program effect reported in Tables 4 and 5, while the q-value is estimated using Simes’ (1986) method. 50 Table A.1. The list of 62 poorest districts (Resolution 30a/2008/NQ-CP) Provinces The number of Name of the poorest districts poorest districts in the province Hà Giang 6 Đồng Văn, Mèo Vạc, Yên Minh, Quản Bạ, Hoàng Su Phì, Xín Mần Cao Bằng 5 Bảo Lâm, Bảo Lạc, Thông Nông, Hà Quảng, Hạ Lang Lào Cai 3 Si Ma Cai, Mường Khương, Bắc Hà Yên Bái 2 Mù Cang Chải, Trạm Tấu Bắc Kạn 2 Ba Bể, Pác Nặm Bắc Giang 1 Sơn Động Phú Thọ 1 Tân Sơn Sơn La 5 Sốp Cộp, Phù Yên, Bắc Yên, Mường La, Quỳnh Nhai Lai Châu 5 Mường Tè, Phong Thổ, Sìn Hồ, Tân Yên, Than Uyên Điện Biên 4 Điện Biên Đông, Mường Nhé, Tủa Chùa, Mường Ảng Thanh Hóa 7 Lang Chánh, Thường Xuân, Quan Hóa, Quan Sơn, Mường Lát, Như Xuân, Bá Thước Nghệ An 3 Kỳ Sơn, Tương Dương, Quế Phong Quảng Bình 1 Minh Hóa Quảng Trị 1 Đa Krông Quảng Ngãi 6 Sơn Hà, Trà Bồng, Sơn Tây, Minh Long, Tây Trà, Ba Tơ Quảng Nam 3 Nam Trà My, Tây Giang, Phước Sơn Bình Định 3 An Lão, Vĩnh Thạnh, Vân Canh Ninh Thuận 1 Bác Ái Kon Tum 2 Tu Mơ Rông, Kon Plông Lâm Đồng 1 Đam Rông Source: Authors’ preparation using information from Resolution 30a/2008/NQ-CP (The Government of Vietnam, 2008). 51 Table A.2. Outcome variables Treatment Control Variables Year Year Year Year Year Year Year Year Year Year 2004 2008 2012 2016 2020 2004 2008 2012 2016 2020 Individual-level outcomes 25.81 89.57 94.13 97.19 97.77 23.73 73.29 82.59 85.97 94.34 Percentage of having health insurance (3.47) (2.79) (0.94) (0.53) (0.67) (1.95) (2.28) (1.68) (1.63) (0.78) The annual number of healthcare 0.43 0.53 0.50 0.69 0.44 0.72 0.75 0.90 0.82 0.70 visits (0.03) (0.05) (0.04) (0.07) (0.04) (0.05) (0.05) (0.08) (0.07) (0.07) The annual number of outpatient 0.33 0.44 0.40 0.56 0.33 0.62 0.66 0.75 0.65 0.58 healthcare visits (0.03) (0.05) (0.04) (0.06) (0.04) (0.05) (0.05) (0.08) (0.07) (0.07) The annual number of inpatient 0.09 0.10 0.10 0.13 0.12 0.11 0.09 0.15 0.17 0.11 healthcare visits (0.01) (0.01) (0.01) (0.02) (0.02) (0.01) (0.01) (0.02) (0.02) (0.01) The number of completed schooling 4.28 4.84 4.48 4.66 5.70 4.94 5.38 4.97 5.07 5.31 years (children aged 6-17) (0.17) (0.18) (0.14) (0.12) (0.28) (0.13) (0.16) (0.12) (0.09) (0.21) Percentage of attending school 80.27 77.88 78.52 82.89 84.06 83.91 80.53 81.64 85.23 84.65 (children aged 6-17) (1.60) (2.00) (1.44) (1.60) (2.08) (1.22) (1.44) (1.28) (1.33) (1.53) Percentage of receiving subsidy for 41.68 45.44 57.76 55.18 44.72 19.33 25.82 31.09 22.24 19.19 students (children aged 6-17) (3.97) (3.67) (2.63) (3.02) (3.82) (2.32) (2.63) (2.64) (2.75) (2.46) Education subsidy for students 0.28 0.40 1.26 2.56 3.26 0.26 0.48 0.95 2.41 3.35 (children aged 6-17, million VND) (0.04) (0.06) (0.10) (0.18) (0.23) (0.03) (0.08) (0.09) (0.34) (0.42) Household-level outcomes Percentage of borrowing from the 13.80 13.89 n.a. 7.95 10.08 24.27 28.76 n.a. 18.11 15.37 formal sources (0.00) (1.83) n.a. (1.14) (0.00) (0.00) (0.00) n.a. (1.38) (0.00) Percentage of borrowing from the 14.85 25.31 39.38 31.18 29.91 12.28 20.08 24.59 28.64 24.01 micro-credit sources (1.76) (2.09) (2.69) (2.41) (2.43) (1.15) (1.42) (1.60) (1.73) (1.77) Formal loan size (sample of 3.26 5.52 n.a. 15.06 25.44 5.59 7.53 n.a. 18.53 47.94 borrowing households, million VND) (0.00) (0.00) n.a. (0.00) (0.00) (0.00) (0.00) n.a. (0.00) (0.00) Microcredit size (sample of 1.96 2.79 4.47 6.41 10.72 2.34 2.99 5.63 8.75 11.59 borrowing households, million VND) (0.12) (0.14) (0.18) (0.29) (0.61) (0.11) (0.09) (0.34) (0.45) (0.58) Formal loan size (sample all 0.45 0.77 n.a. 1.20 2.57 1.36 2.17 n.a. 3.36 7.37 households, million VND) (0.00) (0.16) n.a. (0.19) (0.00) (0.00) (0.00) n.a. (0.38) (0.00) Microcredit size (sample all 0.29 0.71 1.76 2.00 3.21 0.29 0.60 1.39 2.51 2.78 households, million VND) (0.03) (0.07) (0.14) (0.18) (0.31) (0.03) (0.05) (0.12) (0.19) (0.25) Note: The sample is limited to households residing in districts with a poverty rate exceeding 40% in 2006. Within this sample, households living in the 30A districts are referred to as the treatment group, while households residing in districts without the 30A Programs but having the 2006 poverty rate above 40% are considered the control group. There are no data on loans from formal sources in VHLSSs 2010 and 2012 Standard errors of the means in parentheses. 52 Table A.3. RDD regressions of employment of individuals using VHLSSs 2004 to 2008 Dependent variables Currently Self- Self- Having wage Log of Log of working employed employed job (yes=1, monthly wage monthly wage Explanatory variables (yes=1, no=0) farm work non-farm no=0) (wage (all workers) (yes=1, no=0) work (yes=1, workers) no=0) (1) (2) (3) (4) (5) (6) Program 0.0461*** 0.1172*** -0.0340*** -0.0371* 0.1427* -0.6372** (0.0160) (0.0314) (0.0106) (0.0201) (0.0814) (0.2508) (Poverty rate – 50) 0.0008 -0.0081** 0.0035** 0.0054** -0.0054 0.0548** (0.0019) (0.0038) (0.0015) (0.0022) (0.0069) (0.0254) Program * (Poverty rate – 50) -0.0016 0.0082** -0.0035** -0.0063*** 0.0004 -0.0445 (0.0021) (0.0040) (0.0015) (0.0023) (0.0077) (0.0281) Age 0.0392*** 0.0274*** 0.0039*** 0.0078*** 0.0482*** 0.1250*** (0.0015) (0.0013) (0.0005) (0.0007) (0.0075) (0.0078) Age squared -0.0005*** -0.0003*** -0.0000*** -0.0001*** -0.0006*** -0.0016*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) Male 0.0111** -0.0288*** -0.0071* 0.0470*** 0.0605** 0.9283*** (0.0054) (0.0069) (0.0037) (0.0046) (0.0264) (0.0580) Ethnic minorities (yes=1, Kinh=0) 0.0480*** 0.2035*** -0.0761*** -0.0794*** -0.1935*** -0.5107*** (0.0106) (0.0243) (0.0106) (0.0149) (0.0433) (0.1390) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.1671*** 0.0978*** 0.0496*** 0.0196 6.5815*** -0.3888* (0.0330) (0.0362) (0.0118) (0.0194) (0.1378) (0.2099) Observations 41,780 41,780 41,780 41,780 7,773 41,780 R-squared 0.266 0.154 0.048 0.050 0.073 0.078 Note: This table reports district fixed-effect RDD regressions of employment of individuals using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1 53 Table A.4. RDD regressions of individual’s health and children’s education using VHLSSs 2004 to 2008 Dependent variables Having The annual The annual The annual The number Currently Log of health number of number of number of of completed attending education insurance healthcare outpatient inpatient schooling school subsidy for Explanatory variables (yes=1, visits healthcare healthcare years (children students no=0) visits visits (children aged 6-17) aged 6-17) (1) (2) (3) (4) (5) (6) (7) Program 0.1991*** -0.2849*** -0.2718*** -0.0130 0.0840 0.0230 0.6334** (0.0651) (0.0877) (0.0845) (0.0174) (0.2655) (0.0386) (0.2507) (Poverty rate – 50) -0.0046 0.0132 0.0109 0.0022 -0.0492* -0.0089** -0.0098 (0.0064) (0.0119) (0.0114) (0.0018) (0.0276) (0.0039) (0.0227) Program * (Poverty rate – 50) -0.0060 -0.0094 -0.0071 -0.0022 0.0452 0.0096** 0.0161 (0.0070) (0.0123) (0.0118) (0.0020) (0.0298) (0.0043) (0.0270) Age -0.0012* -0.0063* -0.0066** 0.0003 1.2068*** 0.1701*** 0.3738*** (0.0007) (0.0032) (0.0032) (0.0007) (0.0535) (0.0116) (0.0469) Age squared 0.0000** 0.0003*** 0.0002*** 0.0000*** -0.0226*** -0.0084*** -0.0183*** (0.0000) (0.0001) (0.0001) (0.0000) (0.0024) (0.0005) (0.0020) Male 0.0219*** -0.1499*** -0.1384*** -0.0114* 0.1320** 0.0498*** 0.1180*** (0.0044) (0.0213) (0.0197) (0.0060) (0.0569) (0.0102) (0.0347) Ethnic minorities (yes=1, Kinh=0) -0.0423 -0.3841*** -0.3644*** -0.0196* -1.1113*** -0.0862*** 0.9549*** (0.0312) (0.0715) (0.0704) (0.0103) (0.1298) (0.0185) (0.1138) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 0.2386*** 0.9632*** 0.8723*** 0.0907*** -5.5569*** 0.0739 -1.6620*** (0.0331) (0.0936) (0.0917) (0.0158) (0.3042) (0.0708) (0.2945) Observations 62,232 62,232 62,232 62,232 19,133 19,133 19,133 R-squared 0.115 0.053 0.047 0.014 0.581 0.114 0.142 Note: This table reports district fixed-effect RDD regressions of individual’s health and children’s education using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 54 Table A.5. RDD regressions of households’ income and poverty using VHLSSs 2004 to 2008 Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per Poor capita income capita income capita income capita income capita capita public capita income households from wages from nonfarm from farm remittances cash transfers from other (using Explanatory variables production production sources constant poverty line) (1) (2) (3) (4) (5) (6) (7) (8) Program -0.1970** -0.9491 -0.6676 0.1892 -0.4467 -0.5750** -0.2603 0.1224* (0.0931) (0.5946) (0.4126) (0.2526) (0.4421) (0.2485) (0.2403) (0.0644) (Poverty rate – 50) 0.0050 0.0938* 0.0168 -0.0284 -0.0102 0.0509* 0.0047 -0.0012 (0.0101) (0.0558) (0.0498) (0.0337) (0.0494) (0.0264) (0.0267) (0.0065) Program * (Poverty rate – 50) -0.0083 -0.0709 0.0123 0.0249 0.0300 -0.0369 -0.0156 0.0039 (0.0108) (0.0626) (0.0525) (0.0341) (0.0535) (0.0295) (0.0283) (0.0072) Age of household heads 0.0210*** 0.0927*** 0.0549** 0.1233*** -0.0396*** -0.0263** -0.0260 -0.0140*** (0.0041) (0.0239) (0.0223) (0.0191) (0.0150) (0.0126) (0.0207) (0.0026) -0.0002*** -0.0013*** -0.0007*** -0.0013*** 0.0006*** 0.0005*** 0.0007*** 0.0001*** Age squared of household heads (0.0000) (0.0002) (0.0002) (0.0002) (0.0002) (0.0001) (0.0002) (0.0000) Gender of household head -0.0382 -0.4097*** 0.3645** 0.5771*** -0.4978*** -0.1663* -0.0721 0.0338* (male=1, female=0) (0.0260) (0.1360) (0.1721) (0.1710) (0.0909) (0.0942) (0.1566) (0.0173) Ethnic minorities (yes=1, -0.3843*** -0.5678* -1.1928*** 0.7370*** -0.8648*** 0.2361** -1.0454*** 0.2432*** Kinh=0) (0.0561) (0.3085) (0.2192) (0.1521) (0.1921) (0.1195) (0.1267) (0.0360) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 8.7636*** 3.3031*** 1.3617** 4.2311*** 5.2798*** 1.1064*** 1.8050*** 0.7455*** (0.1263) (0.8451) (0.5453) (0.5251) (0.4835) (0.3660) (0.4543) (0.0861) Observations 12,154 12,154 12,154 12,154 12,154 12,154 12,154 12,154 R-squared 0.178 0.032 0.039 0.102 0.059 0.039 0.111 0.132 Note: This table reports district fixed-effect RDD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 55 Table A.6. The district fixed-effect RDD-DD regressions of employment of individuals aged 15-64 Dependent variables Currently Self-employed Self-employed Having wage Log of Log of working farm work non-farm job (yes=1, monthly wage monthly wage Explanatory variables (yes=1, no=0) (yes=1, no=0) work (yes=1, no=0) (wage (all workers) no=0) workers) (1) (2) (3) (4) (5) (6) Program * Post-program period -0.0189 -0.1021* 0.0561*** 0.0270 0.1280 0.5345* (0.0209) (0.0569) (0.0158) (0.0546) (0.2707) (0.3071) (Poverty rate – 50) * Post-program -0.0028 0.0066 -0.0060*** -0.0034 -0.0054 -0.0319 period (0.0027) (0.0059) (0.0019) (0.0044) (0.0336) (0.0294) Program * (Poverty rate – 50) * Post- 0.0048* -0.0023 0.0049** 0.0021 0.0035 -0.0039 program period (0.0028) (0.0064) (0.0019) (0.0051) (0.0349) (0.0325) Age 0.0417*** 0.0114*** 0.0062*** 0.0240*** 0.0435*** 0.2651*** (0.0023) (0.0021) (0.0010) (0.0012) (0.0064) (0.0127) Age squared -0.0005*** -0.0001*** -0.0001*** -0.0003*** -0.0006*** -0.0037*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0001) (0.0002) Male 0.0045 -0.0822*** -0.0076** 0.0942*** 0.0886*** 1.0887*** (0.0047) (0.0075) (0.0032) (0.0057) (0.0197) (0.0628) Ethnic minorities (yes=1, Kinh=0) 0.0590*** 0.1939*** -0.1021*** -0.0328* -0.3696*** -0.2361 (0.0132) (0.0283) (0.0113) (0.0180) (0.0514) (0.1705) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.1246*** 0.4184*** 0.0093 -0.3031*** 6.8637*** -2.9807*** (0.0463) (0.0455) (0.0168) (0.0275) (0.1376) (0.2782) Observations 118,369 118,369 118,369 118,369 27,361 118,369 R-squared 0.174 0.191 0.070 0.187 0.202 0.167 Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 56 Table A.7. The impact of the program on households’ income and poverty Dependent variables Share of Share of Share of Share of Share of Share of income from income from income from private public cash income from wages nonfarm farm remittances transfers other sources Explanatory variables (in percent) production (in production (in (in percent) (in percent) (in percent) percent) percent) (1) (2) (3) (4) (5) (6) Program * Post-program period 2.4046 4.1918** -15.2273*** 2.6543** 1.4392* 4.5374*** (3.6972) (1.7090) (4.4861) (1.0337) (0.7440) (1.3819) (Poverty rate – 50) * Post- -0.2635 -0.3696* 1.0891** -0.1838 -0.0554 -0.2167 program period (0.3709) (0.2036) (0.4849) (0.1581) (0.0634) (0.1631) Program * (Poverty rate – 50) * -0.0163 0.3133 -0.6639 0.1126 -0.0141 0.2683 Post-program period (0.4001) (0.2077) (0.5139) (0.1629) (0.0974) (0.1726) Age of household heads -0.1432 0.1256*** 1.1810*** -0.4116*** -0.4833*** -0.2686*** (0.1243) (0.0379) (0.1429) (0.0533) (0.0495) (0.0426) -0.0010 -0.0022*** -0.0120*** 0.0055*** 0.0063*** 0.0034*** Age squared of household heads (0.0012) (0.0004) (0.0014) (0.0006) (0.0006) (0.0005) Gender of household head -3.6036*** 0.1971 8.6076*** -3.6608*** -0.3850* -1.1553*** (male=1, female=0) (0.7196) (0.3688) (0.7596) (0.5531) (0.2236) (0.3021) Ethnic minorities (yes=1, -3.0134** -11.8797*** 15.4604*** -0.7899 1.0040*** -0.7815*** Kinh=0) (1.4817) (0.9528) (1.7760) (0.4945) (0.2947) (0.2687) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 28.6026*** 13.1056*** 24.7434*** 14.5162*** 8.9215*** 10.1106*** (3.0573) (1.2043) (3.6762) (1.6011) (1.2172) (0.9337) Observations 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.191 0.092 0.296 0.111 0.127 0.059 Note: This table reports district fixed-effect RDD-DD regressions of share of income from different sources of households using household- level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 57 Table A.8. The district fixed-effect RDD-DD regressions of employment of individuals using the sample without the program 135 communes Dependent variables Currently Self- Self- Having wage Log of Log of working employed employed job (yes=1, monthly wage monthly wage Explanatory variables (yes=1, no=0) farm work non-farm no=0) (wage (all workers) (yes=1, no=0) work (yes=1, workers) no=0) (1) (2) (3) (4) (5) (6) Program * Post-program period -0.0356 -0.1896*** 0.0623*** 0.0917* 0.1535 0.7408** (0.0292) (0.0623) (0.0225) (0.0501) (0.2958) (0.3287) (Poverty rate – 50) * Post-program 0.0020 -0.0140* 0.0050 0.0109** 0.0276 0.0326 period (0.0040) (0.0084) (0.0034) (0.0047) (0.0398) (0.0337) Program * (Poverty rate – 50) * Post- 0.0002 0.0184** -0.0061* -0.0121*** -0.0297 -0.0665** program period (0.0040) (0.0080) (0.0033) (0.0041) (0.0388) (0.0310) Age 0.0378*** 0.0229*** 0.0042*** 0.0107*** 0.0422*** 0.1122*** (0.0017) (0.0014) (0.0006) (0.0008) (0.0063) (0.0077) Age squared -0.0005*** -0.0003*** -0.0000*** -0.0001*** -0.0006*** -0.0015*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) Male 0.0130** -0.0656*** -0.0086** 0.0872*** 0.0968*** 1.0201*** (0.0058) (0.0085) (0.0034) (0.0056) (0.0248) (0.0655) Ethnic minorities (yes=1, Kinh=0) 0.0497*** 0.1700*** -0.0961*** -0.0242 -0.3798*** -0.2234 (0.0129) (0.0283) (0.0121) (0.0169) (0.0595) (0.1549) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.1920*** 0.2662*** 0.0321** -0.1063*** 6.9410*** -0.7118*** (0.0379) (0.0366) (0.0127) (0.0212) (0.1305) (0.2077) Observations 91,916 91,916 91,916 91,916 19,209 91,916 R-squared 0.292 0.204 0.068 0.183 0.189 0.162 Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 58 Table A.9. The district fixed-effect RDD-DD regressions of individual’s health and children’s education using the sample without the program 135 communes Dependent variables Having The annual The annual The annual The number Currently Log of health number of number of number of of completed attending education insurance healthcare outpatient inpatient schooling school subsidy for Explanatory variables (yes=1, visits healthcare healthcare years (children students no=0) visits visits (children aged 6-17) (children aged 6-17) aged 6-17) (1) (2) (3) (4) (5) (6) (7) Program * Post-program period 0.1206 0.4203** 0.4249** -0.0047 0.4343 0.0071 -0.0142 (0.1006) (0.2041) (0.1831) (0.0530) (0.3342) (0.0497) (0.0283) (Poverty rate – 50) * Post-program 0.0285** 0.0723** 0.0684** 0.0038 -0.0134 -0.0097 0.0007 period (0.0111) (0.0312) (0.0300) (0.0064) (0.0377) (0.0062) (0.0027) Program * (Poverty rate – 50) * -0.0188* -0.0743** -0.0709** -0.0033 0.0150 0.0070 0.0009 Post-program period (0.0104) (0.0309) (0.0298) (0.0063) (0.0359) (0.0060) (0.0022) Age -0.0029*** -0.0049 -0.0046 -0.0002 1.1898*** 0.1488*** -0.0095*** (0.0009) (0.0044) (0.0044) (0.0007) (0.0484) (0.0114) (0.0008) Age squared 0.0000*** 0.0003*** 0.0002*** 0.0000*** -0.0203*** -0.0078*** 0.0001*** (0.0000) (0.0001) (0.0001) (0.0000) (0.0023) (0.0005) (0.0000) Male 0.0158*** -0.1239*** -0.1213*** -0.0025 0.1608** 0.0421*** 0.0081** (0.0032) (0.0234) (0.0209) (0.0068) (0.0622) (0.0120) (0.0034) Ethnic minorities (yes=1, Kinh=0) 0.0365 -0.2617*** -0.2596*** -0.0021 -0.8599*** -0.0915*** 0.0484*** (0.0263) (0.0905) (0.0926) (0.0109) (0.1712) (0.0296) (0.0128) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 0.2514*** 0.6532*** 0.5861*** 0.0670*** -5.7179*** 0.2867*** 0.2305*** (0.0245) (0.0717) (0.0701) (0.0124) (0.3146) (0.0685) (0.0187) Observations 134,049 61,118 61,118 61,118 35,930 35,930 134,049 R-squared 0.532 0.097 0.092 0.030 0.520 0.196 0.182 Note: This table reports district fixed-effect RDD-DD regressions of individual’s health and children’s education using individual-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 59 Table A.10. The district fixed-effect RDD-DD regressions of households’ income and poverty using the sample without the program 135 communes Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per Income poor capita income capita income capita income capita income capita capita public capita income households Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program -0.0517 0.4750 0.9983** -0.4430* 0.5304 0.9442*** 1.1550** -0.0503 period (0.0919) (0.5635) (0.4380) (0.2309) (0.5760) (0.2807) (0.4489) (0.0598) (Poverty rate – 50) * Post- 0.0044 -0.0997* -0.1156** 0.0247 -0.0059 -0.0651** 0.0079 -0.0022 program period (0.0111) (0.0520) (0.0560) (0.0305) (0.0663) (0.0277) (0.0531) (0.0058) Program * (Poverty rate – 50) * -0.0085 0.0527 0.1009* -0.0203 -0.0306 0.0529 -0.0019 0.0053 Post-program period (0.0115) (0.0570) (0.0582) (0.0311) (0.0709) (0.0342) (0.0569) (0.0063) Age of household heads 0.0268*** 0.0528*** 0.0657*** 0.1398*** -0.0501*** -0.1080*** -0.0588*** -0.0142*** (0.0026) (0.0155) (0.0099) (0.0150) (0.0099) (0.0115) (0.0126) (0.0015) Age squared of household -0.0003*** -0.0008*** -0.0008*** -0.0014*** 0.0007*** 0.0014*** 0.0007*** 0.0001*** heads (0.0000) (0.0002) (0.0001) (0.0002) (0.0001) (0.0001) (0.0001) (0.0000) Gender of household head 0.0166 0.0184 0.2169*** 0.7117*** -0.3702*** -0.0586 -0.3139*** -0.0139 (male=1, female=0) (0.0203) (0.0978) (0.0711) (0.0886) (0.0617) (0.0631) (0.0792) (0.0100) Ethnic minorities (yes=1, -0.5658*** -0.3437* -1.7880*** 0.9697*** -0.4657*** 0.4397*** -0.5626*** 0.2113*** Kinh=0) (0.0573) (0.2002) (0.1854) (0.1433) (0.1536) (0.1129) (0.1200) (0.0242) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 8.5712*** 2.9181*** 1.3450*** 3.5091*** 4.9270*** 2.1595*** 2.9019*** 0.8883*** (0.0961) (0.4433) (0.2898) (0.3731) (0.3201) (0.3182) (0.3225) (0.0496) Observations 28,920 28,920 28,920 28,920 28,920 28,920 28,920 28,920 R-squared 0.389 0.180 0.117 0.160 0.241 0.164 0.101 0.275 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 60 Table A.11. The district fixed-effect RDD-DD regressions of households’ access to credit using the sample without the program 135 communes Dependent variables Borrow from Borrow from Log of formal Log of Log of formal Log of formal sources microcredit loan (the microcredit loan (the microcredit (yes=1, no=0) sources sample of (the sample of sample all (the sample all Explanatory variables (yes=1, no=0) borrowing borrowing households) households) households) households) (1) (2) (3) (4) (5) (6) Program * Post-program period 0.2078*** 0.1529*** -0.1693 -0.1595 1.7207*** 1.5428*** (0.0717) (0.0547) (0.1472) (0.1138) (0.5882) (0.5202) (Poverty rate – 50) * Post-program -0.0229*** -0.0171** 0.0124 -0.0023 -0.1665** -0.1732** period (0.0085) (0.0074) (0.0194) (0.0144) (0.0715) (0.0700) Program * (Poverty rate – 50) * 0.0202** 0.0174** -0.0189 0.0069 0.1452* 0.1775** Post-program period (0.0089) (0.0077) (0.0199) (0.0148) (0.0745) (0.0741) Age of household heads 0.0080*** 0.0033** 0.0207*** 0.0147*** 0.0711*** 0.0354** (0.0018) (0.0014) (0.0067) (0.0048) (0.0144) (0.0139) -0.0001*** -0.0001*** -0.0002*** -0.0001*** -0.0010*** -0.0006*** Age squared of household heads (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0245* 0.0027 0.1278*** 0.0623** 0.2379** 0.0333 female=0) (0.0129) (0.0091) (0.0374) (0.0276) (0.1026) (0.0914) -0.0129 0.0503** -0.5654*** -0.2406*** -0.3353* 0.4499** Ethnic minorities (yes=1, Kinh=0) (0.0228) (0.0198) (0.0754) (0.0356) (0.1794) (0.1958) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 0.1996*** 0.0740* 9.1621*** 8.8840*** 1.9772*** 0.5824 (0.0506) (0.0441) (0.1536) (0.1129) (0.3941) (0.4371) Observations 21,737 28,920 8,215 8,085 28,920 28,920 R-squared 0.116 0.123 0.384 0.335 0.207 0.126 Note: This table reports district fixed-effect RDD-DD regressions of loans of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 61 Table A.12. The placebo test of the program effect on individuals’ outcomes Dependent variables Self- Self- Having Log of Log of The annual The annual Log of employed employed wage job monthly monthly number of number of education farm work non-farm (yes=1, wage (all wage (wage healthcare outpatient subsidy for Explanatory variables (yes=1, work no=0) workers) workers) visits healthcare students no=0) (yes=1, visits (children no=0) aged 6-17) (1) (2) (3) (4) (5) (6) (7) (8) Program * Year 2008 0.0276 -0.0009 -0.0262* 0.1003 -0.1977* 0.0259 0.0187 -0.4615 (0.0346) (0.0149) (0.0156) (0.2917) (0.1086) (0.1326) (0.1283) (0.3139) (Poverty rate – 50) * Year 2008 -0.0044 0.0001 0.0013 0.0110 0.0123 -0.0081 -0.0061 0.0023 (0.0038) (0.0019) (0.0022) (0.0251) (0.0110) (0.0150) (0.0141) (0.0285) Program * (Poverty rate – 50) * Year 0.0051 -0.0000 -0.0007 -0.0181 -0.0093 0.0146 0.0098 0.0363 2008 (0.0039) (0.0019) (0.0023) (0.0287) (0.0122) (0.0159) (0.0152) (0.0332) Age 0.0277*** 0.0039*** 0.0077*** 0.1213*** 0.0420*** -0.0050 -0.0054 0.3567*** (0.0017) (0.0007) (0.0006) (0.0087) (0.0086) (0.0043) (0.0042) (0.0581) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** -0.0005*** 0.0003*** 0.0002*** -0.0171*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0025) Male -0.0314*** -0.0062 0.0476*** 0.9325*** 0.0348 -0.1440*** -0.1333*** 0.1443*** (0.0082) (0.0038) (0.0055) (0.0706) (0.0294) (0.0276) (0.0265) (0.0386) Ethnic minorities (yes=1, Kinh=0) 0.2366*** -0.1104*** -0.0764*** -0.3756* -0.2764*** -0.2440** -0.2373** 0.8376*** (0.0419) (0.0175) (0.0242) (0.2245) (0.0455) (0.1173) (0.1150) (0.1686) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.1410*** 0.0497*** -0.0162 -0.7904*** 6.7972*** 0.7029*** 0.6362*** -1.2519*** (0.0428) (0.0140) (0.0205) (0.2558) (0.1504) (0.0908) (0.0885) (0.3508) Observations 41,780 41,780 41,780 41,780 7,773 62,232 62,232 19,133 R-squared 0.220 0.083 0.102 0.161 0.201 0.099 0.094 0.417 Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. The sample used in these regressions includes VHLSS 2004, 2006 and 2008. The VHLSS 2008 is used as the survey after the placebo program. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 62 Table A.13. The placebo test of the program effect on individual’s health and children’s education Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per Borrow Borrow Log of Log of capita capita capita capita capita capita capita from from formal loan microcredit income income income income remittances public cash income formal micro- (the sample (the sample from from from farm transfers from other sources credit of borrowing of Explanatory variables wages nonfarm production sources (yes=1, sources households) borrowing production no=0) (yes=1, households) no=0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Program * Year 2008 -0.0349 0.0295 -0.0273 0.3211 0.0359 0.2058 -0.1799 0.0386 0.0268 0.1063 0.0532 (0.0592) (0.5273) (0.3235) (0.2259) (0.4219) (0.2911) (0.3593) (0.0614) (0.0487) (0.0900) (0.1331) (Poverty rate – 50) * -0.0013 0.0285 -0.0383 -0.0532* -0.0216 -0.0368 0.0214 0.0010 0.0034 0.0069 0.0257** Year 2008 (0.0086) (0.0440) (0.0372) (0.0305) (0.0397) (0.0300) (0.0389) (0.0088) (0.0050) (0.0115) (0.0122) Program * (Poverty rate 0.0001 -0.0256 0.0418 0.0518* 0.0097 0.0585 -0.0233 -0.0028 -0.0029 -0.0081 -0.0289** – 50) * Year 2008 (0.0091) (0.0517) (0.0404) (0.0309) (0.0436) (0.0371) (0.0411) (0.0091) (0.0059) (0.0123) (0.0139) Age of household heads 0.0173*** 0.0925*** 0.0527** 0.1209*** -0.0456*** -0.0213* -0.0266 0.0137*** 0.0063*** 0.0236*** 0.0041 (0.0039) (0.0205) (0.0205) (0.0194) (0.0129) (0.0126) (0.0205) (0.0029) (0.0021) (0.0070) (0.0060) Age squared of -0.0001*** -0.0013*** -0.0006*** -0.0012*** 0.0006*** 0.0005*** 0.0007*** -0.0002*** -0.0001*** -0.0002*** -0.0000 household heads (0.0000) (0.0002) (0.0002) (0.0002) (0.0001) (0.0001) (0.0002) (0.0000) (0.0000) (0.0001) (0.0001) Gender of hh. head 0.0006 -0.1157 0.5530*** 0.4849*** -0.4818*** -0.1337 0.0139 0.0496*** 0.0202 0.0798 0.0819* (male=1, female=0) (0.0274) (0.1297) (0.1250) (0.1357) (0.0787) (0.0825) (0.1498) (0.0170) (0.0122) (0.0483) (0.0418) Ethnic minorities -0.5658*** -0.3507 -1.8163*** 0.6421*** -0.4542*** 0.4998*** -1.1709*** -0.0606** 0.0430** -0.4543*** -0.0261 (yes=1, Kinh=0) (0.0597) (0.2528) (0.3140) (0.1440) (0.1494) (0.1835) (0.2166) (0.0277) (0.0212) (0.0572) (0.0545) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 8.8094*** 2.3786*** 1.4655*** 4.4929*** 5.0903*** 0.5021 1.6602*** 0.0642 -0.0288 9.1221*** 8.9677*** (0.1221) (0.4986) (0.4972) (0.4803) (0.3533) (0.3749) (0.4398) (0.0743) (0.0601) (0.1666) (0.1519) Observations 12,154 12,154 12,154 12,154 12,154 12,154 12,154 12,154 12,154 4,536 2,346 R-squared 0.298 0.159 0.149 0.214 0.238 0.120 0.159 0.130 0.105 0.192 0.213 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 63 Table A.14. The placebo test of the program effect on individuals’ outcomes using the threshold of the poverty rate at 40% Dependent variables Self- Self- Having Log of Log of The annual The annual Log of employed employed wage job monthly monthly number of number of education farm work non-farm (yes=1, wage (all wage (wage healthcare outpatient subsidy for Explanatory variables (yes=1, work no=0) workers) workers) visits healthcare students no=0) (yes=1, visits (children no=0) aged 6-17) (1) (2) (3) (4) (5) (6) (7) (8) Districts with the 2006 poverty above 0.0453 0.0070 -0.0186 0.0430 0.1368 0.1083 -0.0519* 0.2058 40% * Year 2008 (0.0389) (0.0155) (0.0300) (0.2323) (0.1777) (0.1742) (0.0268) (0.3198) (Poverty rate – 40) * Year 2008 -0.0034* 0.0001 0.0031** -0.0147 -0.0067 -0.0057 0.0001 0.0339*** (0.0018) (0.0010) (0.0013) (0.0109) (0.0082) (0.0079) (0.0009) (0.0111) Districts with the 2006 poverty above 0.0098 -0.0056*** -0.0066 -0.0174 -0.0532** -0.0508** 0.0075 -0.0805 40% * (Poverty rate – 40) * Year 2008 (0.0060) (0.0021) (0.0045) (0.0294) (0.0241) (0.0232) (0.0050) (0.0525) Age 0.0320*** 0.0084*** 0.0094*** 0.1108*** -0.0090*** -0.0084*** 0.1544*** 0.1415*** (0.0008) (0.0004) (0.0004) (0.0039) (0.0024) (0.0022) (0.0055) (0.0187) Age squared -0.0003*** -0.0001*** -0.0001*** -0.0015*** 0.0004*** 0.0003*** -0.0079*** -0.0071*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0003) (0.0009) Male -0.0737*** -0.0208*** 0.1152*** 1.2509*** -0.1637*** -0.1555*** 0.0130*** -0.0085 (0.0086) (0.0039) (0.0048) (0.0421) (0.0170) (0.0156) (0.0047) (0.0129) Ethnic minorities (yes=1, Kinh=0) 0.1133*** -0.0679*** 0.0083 0.1569* -0.1082** -0.1096** -0.1046*** 0.7840*** (0.0122) (0.0055) (0.0089) (0.0829) (0.0523) (0.0502) (0.0101) (0.1201) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant -0.0105 -0.0526*** -0.0442*** -0.4257*** 0.6811*** 0.6157*** 0.2591*** -0.4258*** (0.0182) (0.0072) (0.0092) (0.0872) (0.0297) (0.0283) (0.0292) (0.1131) Observations 322,913 322,913 322,913 322,913 216,298 216,298 106,531 106,531 R-squared 0.174 0.058 0.137 0.133 0.114 0.113 0.196 0.301 Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. The placebo treatment includes districts with the 2006 MOLISA poverty above 40%. The sample consists of individuals living in districts with the 2006 MOLISA poverty rate between 30% and 50%. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 64 Table A.15. The placebo test of the program effect on individuals’ outcomes using the threshold of the poverty rate at 30% Dependent variables Self- Self- Having Log of Log of The annual The annual Log of employed employed wage job monthly monthly number of number of education farm work non-farm (yes=1, wage (all wage (wage healthcare outpatient subsidy for Explanatory variables (yes=1, work no=0) workers) workers) visits healthcare students no=0) (yes=1, visits (children no=0) aged 6-17) (1) (2) (3) (4) (5) (6) (7) (8) Districts with the 2006 poverty above -0.0083 -0.0027 -0.0028 -0.1095 0.2970 0.3128 0.0201 0.0088 30% * Year 2008 (0.0370) (0.0178) (0.0273) (0.2217) (0.2287) (0.2325) (0.0170) (0.1262) (Poverty rate – 30) * Year 2008 0.0045 -0.0003 -0.0033 -0.0292 -0.0084 -0.0075 -0.0011 0.0239 (0.0046) (0.0023) (0.0026) (0.0192) (0.0196) (0.0187) (0.0021) (0.0168) Districts with the 2006 poverty above -0.0151** 0.0011 0.0136*** 0.0466 -0.0420 -0.0443 -0.0007 0.0193 40% * (Poverty rate – 30) * Year 2008 (0.0069) (0.0036) (0.0050) (0.0413) (0.0363) (0.0363) (0.0033) (0.0330) Age 0.0333*** 0.0091*** 0.0091*** 0.1083*** -0.0091*** -0.0084*** 0.1542*** 0.1161*** (0.0009) (0.0005) (0.0005) (0.0044) (0.0026) (0.0024) (0.0063) (0.0194) Age squared -0.0003*** -0.0001*** -0.0001*** -0.0015*** 0.0004*** 0.0003*** -0.0079*** -0.0058*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0003) (0.0009) Male -0.0779*** -0.0233*** 0.1222*** 1.3178*** -0.1633*** -0.1571*** 0.0129*** -0.0144 (0.0103) (0.0047) (0.0054) (0.0473) (0.0187) (0.0169) (0.0048) (0.0138) Ethnic minorities (yes=1, Kinh=0) 0.0921*** -0.0585*** 0.0180* 0.2516*** -0.0552 -0.0600 -0.0869*** 0.7475*** (0.0122) (0.0056) (0.0095) (0.0865) (0.0565) (0.0537) (0.0097) (0.1506) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant -0.0476** -0.0664*** -0.0327*** -0.3447*** 0.6606*** 0.5970*** 0.2556*** -0.3239*** (0.0197) (0.0082) (0.0105) (0.0982) (0.0321) (0.0302) (0.0333) (0.1140) Observations 251,539 251,539 251,539 251,539 167,838 167,838 80,003 80,003 R-squared 0.171 0.057 0.134 0.132 0.118 0.117 0.196 0.305 Note: This table reports district fixed-effect RDD-DD regressions of employment of individuals using individual-level observations. The placebo treatment includes districts with the 2006 MOLISA poverty above 30%. The sample consists of individuals living in districts with the 2006 MOLISA poverty rate between 20% and 40%. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 65 Table A.16. The placebo test of the program effect on exogenous variables of individuals Dependent variables Age of Gender of Ethnic The completed Individual Individual individuals individuals minorities education completed completed Explanatory variables (male=1, (yes=1, level (aged high school college and female=0) Kinh=0) from 30) and above above (aged (aged from 30) from 30) (1) (2) (3) (4) (5) (6) Program * Post-program period -1.0608 0.0024 -0.0278 -0.0161 0.0007 -0.0090 (1.0445) (0.0131) (0.0689) (0.1153) (0.0270) (0.0075) (Poverty rate – 50) * Post-program period 0.0610 -0.0013 0.0008 -0.0008 -0.0008 0.0005 (0.1287) (0.0014) (0.0073) (0.0118) (0.0033) (0.0006) Program * (Poverty rate – 50) * Post- -0.1254 0.0023 -0.0007 0.0028 0.0011 -0.0005 program period (0.1341) (0.0014) (0.0076) (0.0128) (0.0034) (0.0007) District fixed effects Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Constant 26.4921*** 0.4950*** 0.7302*** 1.9141*** 0.0964*** 0.0093*** (0.1128) (0.0013) (0.0078) (0.0135) (0.0032) (0.0009) Observations 187,734 187,737 187,734 76,879 76,882 76,882 R-squared 0.028 0.002 0.559 0.197 0.051 0.022 Note: This table reports district fixed-effect RDD-DD regressions of individual-level characteristics. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 66 Table A.17. The district fixed-effect RDD-DD regressions of individual-level outcomes using districts with the 2006 poverty rate from 35% Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1021* 0.0561*** 0.0270 0.5345* 0.3477** 0.3535*** -0.0080 1.2841*** (0.0569) (0.0158) (0.0546) (0.3071) (0.1461) (0.1340) (0.0448) (0.4805) (Poverty rate – 50) * Post-program 0.0066 -0.0060*** -0.0034 -0.0319 -0.0566*** -0.0535*** 0.0075 -0.0467 period (0.0059) (0.0019) (0.0044) (0.0294) (0.0210) (0.0201) (0.0049) (0.0517) Program * (Poverty rate – 50) * Post- -0.0023 0.0049** 0.0021 -0.0039 0.0558** 0.0519** -0.0102** 0.0504 program period (0.0064) (0.0019) (0.0051) (0.0325) (0.0214) (0.0205) (0.0051) (0.0546) Age 0.0114*** 0.0062*** 0.0240*** 0.2651*** -0.0143*** -0.0126*** 0.1571*** 0.3874*** (0.0021) (0.0010) (0.0012) (0.0127) (0.0033) (0.0031) (0.0093) (0.0473) Age squared -0.0001*** -0.0001*** -0.0003*** -0.0037*** 0.0004*** 0.0004*** -0.0081*** -0.0197*** (0.0000) (0.0000) (0.0000) (0.0002) (0.0001) (0.0001) (0.0004) (0.0021) Male -0.0822*** -0.0076** 0.0942*** 1.0887*** -0.1543*** -0.1370*** 0.0370*** 0.1056*** (0.0075) (0.0032) (0.0057) (0.0628) (0.0233) (0.0220) (0.0105) (0.0325) Ethnic minorities (yes=1, Kinh=0) 0.1939*** -0.1021*** -0.0328* -0.2361 -0.2387** -0.2305** -0.1293*** 0.7992*** (0.0283) (0.0113) (0.0180) (0.1705) (0.1059) (0.1048) (0.0258) (0.1114) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.4184*** 0.0093 -0.3031*** -2.9807*** 0.7855*** 0.6951*** 0.2662*** -1.1919*** (0.0455) (0.0168) (0.0275) (0.2782) (0.0797) (0.0782) (0.0530) (0.2757) Observations 118,369 118,369 118,369 118,369 82,402 82,402 50,645 50,645 R-squared 0.191 0.070 0.187 0.167 0.084 0.083 0.195 0.330 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 67 Table A.18. The district fixed-effect RDD-DD regressions of individual-level outcomes using a poverty rate bandwidth of 10% Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1299** 0.0480*** 0.0419 0.5341* 0.2424* 0.2608** 0.0122 1.1550** (0.0522) (0.0142) (0.0493) (0.2805) (0.1397) (0.1300) (0.0400) (0.4450) (Poverty rate – 50) * Post-program 0.0113*** -0.0040*** -0.0062** -0.0304* -0.0250* -0.0248** 0.0020 -0.0119 period (0.0033) (0.0014) (0.0024) (0.0158) (0.0128) (0.0121) (0.0025) (0.0282) Program * (Poverty rate – 50) * Post- -0.0067 0.0029* 0.0050 -0.0040 0.0232* 0.0225* -0.0047 0.0153 program period (0.0041) (0.0015) (0.0034) (0.0207) (0.0134) (0.0127) (0.0029) (0.0333) Age 0.0262*** 0.0043*** 0.0099*** 0.1107*** -0.0056* -0.0050* 0.1609*** 0.3477*** (0.0012) (0.0004) (0.0005) (0.0054) (0.0030) (0.0030) (0.0076) (0.0393) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0084*** -0.0177*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0003) (0.0018) Male -0.0728*** -0.0072*** 0.0914*** 1.1042*** -0.1599*** -0.1471*** 0.0328*** 0.0889*** (0.0064) (0.0027) (0.0047) (0.0553) (0.0211) (0.0200) (0.0081) (0.0262) Ethnic minorities (yes=1, Kinh=0) 0.1575*** -0.0863*** -0.0237** -0.1006 -0.2153*** -0.2013*** -0.1091*** 0.7050*** (0.0191) (0.0088) (0.0114) (0.1125) (0.0675) (0.0657) (0.0184) (0.0889) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.2074*** 0.0229*** -0.0880*** -0.6223*** 0.6915*** 0.6096*** 0.2490*** -0.9925*** (0.0268) (0.0077) (0.0132) (0.1374) (0.0545) (0.0527) (0.0417) (0.2224) Observations 174,774 174,774 174,774 174,774 117,124 117,124 65,879 65,879 R-squared 0.187 0.063 0.172 0.155 0.099 0.093 0.206 0.337 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 68 Table A.19. The district fixed-effect RDD-DD regressions of individual-level outcomes using a poverty rate bandwidth of 7% Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1561** 0.0575*** 0.0419 0.3801 0.3777* 0.3388* 0.0697 1.5792** (0.0733) (0.0171) (0.0775) (0.4918) (0.1922) (0.1796) (0.0619) (0.7210) (Poverty rate – 50) * Post-program 0.0069 -0.0057*** -0.0038 -0.0332 -0.0582** -0.0549** 0.0076 -0.0473 period (0.0057) (0.0018) (0.0043) (0.0269) (0.0226) (0.0219) (0.0049) (0.0517) Program * (Poverty rate – 50) * Post- 0.0040 0.0046 0.0007 0.0239 0.0474 0.0577 -0.0297*** 0.0351 program period (0.0127) (0.0037) (0.0126) (0.0752) (0.0393) (0.0352) (0.0092) (0.1234) Age 0.0254*** 0.0045*** 0.0103*** 0.1130*** -0.0054 -0.0055 0.1608*** 0.3268*** (0.0015) (0.0006) (0.0007) (0.0079) (0.0049) (0.0048) (0.0103) (0.0533) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0082*** -0.0162*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0005) (0.0023) Male -0.0578*** -0.0084** 0.0826*** 0.9369*** -0.1580*** -0.1447*** 0.0295** 0.0840** (0.0084) (0.0037) (0.0061) (0.0602) (0.0341) (0.0328) (0.0131) (0.0388) Ethnic minorities (yes=1, Kinh=0) 0.1941*** -0.1031*** -0.0340* -0.2291 -0.2750** -0.2706** -0.1525*** 0.8975*** (0.0286) (0.0120) (0.0193) (0.1848) (0.1154) (0.1124) (0.0256) (0.1226) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.1668*** 0.0354*** -0.0770*** -0.5728*** 0.7621*** 0.6897*** 0.2459*** -1.2109*** (0.0390) (0.0113) (0.0214) (0.2116) (0.0892) (0.0867) (0.0574) (0.3148) Observations 90,314 90,314 90,314 90,314 61,659 61,659 34,203 34,203 R-squared 0.190 0.067 0.174 0.153 0.097 0.091 0.197 0.295 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 69 Table A.20. The district fixed-effect RDD-DD regressions of individual-level outcomes using a ‘donut’ sample Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1479* 0.0597*** 0.0331 0.1833 0.2307 0.1966 0.0656 1.5132* (0.0780) (0.0194) (0.0796) (0.5181) (0.2178) (0.2087) (0.0689) (0.7682) (Poverty rate – 50) * Post-program 0.0063 -0.0044 -0.0040 0.0379 0.0039 0.0048 0.0112 -0.0234 period (0.0118) (0.0048) (0.0079) (0.0468) (0.0487) (0.0509) (0.0104) (0.0981) Program * (Poverty rate – 50) * Post- 0.0022 0.0007 0.0052 -0.0387 -0.0160 -0.0035 -0.0365** 0.0025 program period (0.0186) (0.0058) (0.0166) (0.0937) (0.0605) (0.0586) (0.0138) (0.1728) Age 0.0246*** 0.0053*** 0.0107*** 0.1213*** -0.0056 -0.0048 0.1612*** 0.3261*** (0.0017) (0.0008) (0.0009) (0.0095) (0.0041) (0.0040) (0.0134) (0.0619) Age squared -0.0003*** -0.0001*** -0.0001*** -0.0016*** 0.0003*** 0.0002*** -0.0082*** -0.0160*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0006) (0.0027) Male -0.0535*** -0.0108** 0.0845*** 0.9882*** -0.1671*** -0.1556*** 0.0382** 0.0873* (0.0105) (0.0051) (0.0071) (0.0740) (0.0455) (0.0442) (0.0155) (0.0440) Ethnic minorities (yes=1, Kinh=0) 0.2283*** -0.1146*** -0.0401* -0.1407 -0.3247** -0.3183** -0.1617*** 0.8786*** (0.0278) (0.0116) (0.0223) (0.2000) (0.1396) (0.1368) (0.0318) (0.1500) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.1385*** 0.0352** -0.0681** -0.6624*** 0.8494*** 0.7637*** 0.2270*** -1.2148*** (0.0402) (0.0137) (0.0261) (0.2409) (0.0941) (0.0934) (0.0753) (0.3581) Observations 62,432 62,432 62,432 62,432 43,227 43,227 23,967 23,967 R-squared 0.198 0.073 0.173 0.165 0.095 0.090 0.202 0.298 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 70 Table A.21. The DID regressions of individual-level outcomes Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program 0.1239*** -0.0352*** -0.0445 -0.6460* -0.2874** -0.2743** 0.0274 0.5869* (0.0441) (0.0133) (0.0272) (0.3508) (0.1449) (0.1395) (0.0524) (0.3346) (Poverty rate – 50) -0.0086 0.0036* 0.0060* 0.0550 0.0134 0.0112 -0.0093* -0.0058 (0.0057) (0.0019) (0.0032) (0.0357) (0.0198) (0.0192) (0.0052) (0.0341) Program * (Poverty rate – 50) 0.0088 -0.0036* -0.0070** -0.0446 -0.0096 -0.0073 0.0100* 0.0114 (0.0059) (0.0019) (0.0033) (0.0392) (0.0203) (0.0196) (0.0056) (0.0395) Program * Post-program period -0.1399*** 0.0532*** 0.0542 0.8080*** 0.4419*** 0.4473*** 0.0126 0.9449** (0.0537) (0.0143) (0.0513) (0.2988) (0.1557) (0.1481) (0.0450) (0.4628) (Poverty rate – 50) * Post-program 0.0086 -0.0048*** -0.0063 -0.0592** -0.0559** -0.0532** 0.0092* -0.0248 period (0.0060) (0.0017) (0.0049) (0.0299) (0.0232) (0.0225) (0.0049) (0.0493) Program * (Poverty rate – 50) * Post- -0.0039 0.0039** 0.0047 0.0168 0.0495** 0.0471** -0.0131** 0.0303 program period (0.0066) (0.0017) (0.0054) (0.0335) (0.0238) (0.0230) (0.0052) (0.0527) Age 0.0241*** 0.0040*** 0.0105*** 0.1161*** -0.0070* -0.0069* 0.1586*** 0.3991*** (0.0013) (0.0005) (0.0006) (0.0066) (0.0039) (0.0038) (0.0094) (0.0486) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0082*** -0.0205*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0004) (0.0022) Male -0.0641*** -0.0080*** 0.0861*** 0.9938*** -0.1551*** -0.1432*** 0.0348*** 0.0900*** (0.0075) (0.0030) (0.0050) (0.0569) (0.0276) (0.0266) (0.0103) (0.0344) Ethnic minorities (yes=1, Kinh=0) 0.1911*** -0.0729*** -0.0626*** -0.5110*** -0.3782*** -0.3585*** -0.0961*** 1.0711*** (0.0273) (0.0093) (0.0174) (0.1605) (0.0769) (0.0764) (0.0193) (0.1274) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.1470*** 0.0470*** -0.0341 -0.1693 0.9745*** 0.8798*** 0.1858*** -1.6689*** (0.0503) (0.0144) (0.0287) (0.2986) (0.1553) (0.1520) (0.0641) (0.3338) Observations 128,446 128,446 128,446 128,446 86,980 86,980 50,645 50,645 R-squared 0.141 0.044 0.126 0.085 0.052 0.046 0.137 0.204 Note: This table reports DID regression of individual-level outcomes. The impact of the program is measured by variable ‘Program * Post-program period’. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 71 Table A.22. The district fixed-effect RDD-DD regressions of individual-level outcomes without control variables Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1087* 0.0571*** 0.0401 0.6282** 0.3414** 0.3553** 0.0027 1.2549** (0.0567) (0.0171) (0.0515) (0.2876) (0.1556) (0.1419) (0.0437) (0.4825) (Poverty rate – 50) * Post-program 0.0066 -0.0056*** -0.0045 -0.0407 -0.0554** -0.0527** 0.0074 -0.0458 period (0.0057) (0.0020) (0.0044) (0.0275) (0.0228) (0.0218) (0.0049) (0.0529) Program * (Poverty rate – 50) * Post- -0.0024 0.0044** 0.0033 0.0066 0.0529** 0.0498** -0.0104** 0.0492 program period (0.0062) (0.0021) (0.0049) (0.0306) (0.0232) (0.0221) (0.0051) (0.0558) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.7368*** 0.0378*** 0.0819*** 1.3837*** 0.6064*** 0.5057*** 0.8226*** 1.0583*** (0.0059) (0.0018) (0.0045) (0.0329) (0.0073) (0.0069) (0.0035) (0.0418) Observations 128,449 128,449 128,449 128,449 86,983 86,983 50,645 50,645 R-squared 0.125 0.039 0.132 0.088 0.062 0.064 0.061 0.304 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 72 Table A.23. The district fixed-effect RDD-DD regressions of individual-level outcomes using a large-specification model Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1112** 0.0521*** 0.0296 0.5192* 0.3505** 0.3591*** -0.0217 1.2441*** (0.0553) (0.0153) (0.0522) (0.2789) (0.1483) (0.1368) (0.0355) (0.4751) (Poverty rate – 50) * Post-program 0.0070 -0.0056*** -0.0038 -0.0337 -0.0572** -0.0540** 0.0060 -0.0475 period (0.0056) (0.0018) (0.0043) (0.0262) (0.0223) (0.0215) (0.0039) (0.0506) Program * (Poverty rate – 50) * Post- -0.0027 0.0046** 0.0030 0.0027 0.0567** 0.0528** -0.0082** 0.0523 program period (0.0061) (0.0019) (0.0049) (0.0294) (0.0227) (0.0219) (0.0041) (0.0536) Age 0.0268*** 0.0039*** 0.0089*** 0.0962*** -0.0027 -0.0034 0.1373*** 0.3337*** (0.0013) (0.0005) (0.0005) (0.0058) (0.0038) (0.0037) (0.0109) (0.0466) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0012*** 0.0002*** 0.0002*** -0.0088*** -0.0198*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0005) (0.0022) Male -0.0589*** -0.0096*** 0.0803*** 0.9564*** -0.1448*** -0.1342*** 0.0290*** 0.1097*** (0.0072) (0.0031) (0.0054) (0.0595) (0.0254) (0.0245) (0.0087) (0.0299) Ethnic minorities (yes=1, Kinh=0) 0.1224*** -0.0879*** 0.0062 0.0210 -0.2176** -0.2159** -0.0344* 0.8635*** (0.0227) (0.0101) (0.0162) (0.1591) (0.0961) (0.0943) (0.0180) (0.1237) Additional control variables -0.1112** 0.0521*** 0.0296 0.5192* 0.3505** 0.3591*** -0.0217 1.2441*** District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.2852*** 0.0129 -0.1131*** -0.7239*** 0.8918*** 0.8090*** 0.4349*** -0.9731*** (0.0391) (0.0123) (0.0228) (0.2336) (0.0900) (0.0866) (0.0628) (0.2742) Observations 128,446 128,446 128,446 128,446 86,980 86,980 50,645 50,645 R-squared 0.232 0.072 0.215 0.193 0.100 0.095 0.284 0.336 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. The additional control variables include education levels of individuals (dummy variables), education levels (dummy variables), household size, the proportion of children in households, the proportion of older members in households. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 73 Table A.24. The district fixed-effect RDD-DD regressions of individual-level outcomes using region-specific time trend Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1531*** 0.0510*** 0.0601 0.5362** 0.2655* 0.2872** -0.0087 0.7291 (0.0570) (0.0168) (0.0493) (0.2695) (0.1533) (0.1435) (0.0469) (0.5125) (Poverty rate – 50) * Post-program 0.0097 -0.0054*** -0.0060 -0.0456* -0.0510** -0.0483** 0.0068 -0.0195 period (0.0059) (0.0018) (0.0047) (0.0274) (0.0228) (0.0221) (0.0050) (0.0512) Program * (Poverty rate – 50) * Post- -0.0047 0.0043** 0.0045 0.0129 0.0501** 0.0467** -0.0095* 0.0293 program period (0.0064) (0.0018) (0.0052) (0.0298) (0.0229) (0.0221) (0.0052) (0.0540) Age 0.0243*** 0.0040*** 0.0104*** 0.1130*** -0.0058 -0.0059 0.1570*** 0.3932*** (0.0013) (0.0005) (0.0006) (0.0065) (0.0039) (0.0038) (0.0093) (0.0473) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0081*** -0.0199*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0004) (0.0021) Male -0.0655*** -0.0076** 0.0867*** 0.9970*** -0.1497*** -0.1383*** 0.0371*** 0.1042*** (0.0072) (0.0030) (0.0051) (0.0567) (0.0263) (0.0253) (0.0105) (0.0322) Ethnic minorities (yes=1, Kinh=0) 0.1854*** -0.0987*** -0.0333** -0.2281 -0.2455** -0.2388** -0.1294*** 0.8573*** (0.0260) (0.0110) (0.0163) (0.1575) (0.0994) (0.0976) (0.0259) (0.1159) Region-specific time trend Yes Yes Yes Yes Yes Yes Yes Yes District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.2182*** 0.0400*** -0.0979*** -0.6343*** 0.6536*** 0.5768*** 0.2783*** -1.3268*** (0.0359) (0.0104) (0.0188) (0.1919) (0.0829) (0.0804) (0.0537) (0.2875) Observations 128,446 128,446 128,446 128,446 86,980 86,980 50,645 50,645 R-squared 0.193 0.066 0.178 0.156 0.097 0.092 0.196 0.337 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. The additional control variables include education levels of individuals (dummy variables), education levels (dummy variables), household size, the proportion of children in households, the proportion of older members in households. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 74 Table A.25. The district fixed-effect RDD-DD regressions of individual-level outcomes with clustering the standard error at the village level Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1144*** 0.0538*** 0.0330 0.5545*** 0.3757*** 0.3814*** -0.0080 1.2841*** (0.0292) (0.0145) (0.0221) (0.1777) (0.0808) (0.0779) (0.0290) (0.2386) (Poverty rate – 50) * Post-program 0.0070** -0.0056*** -0.0038* -0.0338 -0.0580*** -0.0547*** 0.0075** -0.0467* period (0.0034) (0.0018) (0.0022) (0.0206) (0.0117) (0.0114) (0.0031) (0.0248) Program * (Poverty rate – 50) * Post- -0.0026 0.0046** 0.0026 -0.0002 0.0561*** 0.0523*** -0.0102*** 0.0504* program period (0.0036) (0.0018) (0.0024) (0.0219) (0.0119) (0.0116) (0.0035) (0.0269) Age 0.0243*** 0.0040*** 0.0104*** 0.1132*** -0.0058** -0.0059** 0.1571*** 0.3874*** (0.0009) (0.0004) (0.0005) (0.0051) (0.0029) (0.0029) (0.0082) (0.0345) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0081*** -0.0197*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0000) (0.0000) (0.0004) (0.0015) Male -0.0653*** -0.0076*** 0.0865*** 0.9974*** -0.1495*** -0.1382*** 0.0370*** 0.1056*** (0.0050) (0.0025) (0.0038) (0.0398) (0.0186) (0.0172) (0.0073) (0.0245) Ethnic minorities (yes=1, Kinh=0) 0.1805*** -0.0990*** -0.0292** -0.2248** -0.2511*** -0.2432*** -0.1293*** 0.7992*** (0.0194) (0.0096) (0.0125) (0.1120) (0.0781) (0.0771) (0.0195) (0.0953) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.2242*** 0.0402*** -0.0947*** -0.6763*** 0.7249*** 0.6527*** 0.2662*** -1.1919*** (0.0247) (0.0083) (0.0129) (0.1290) (0.0629) (0.0614) (0.0473) (0.2023) Observations 128,446 128,446 128,446 128,446 86,980 86,980 50,645 50,645 R-squared 0.190 0.066 0.174 0.154 0.096 0.091 0.195 0.330 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 75 Table A.26. The district fixed-effect RDD-DD regressions of individual-level outcomes with heteroscedasticity-consistent standard errors Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Program * Post-program period -0.1144*** 0.0538*** 0.0330*** 0.5545*** 0.3757*** 0.3814*** -0.0080 1.2841*** (0.0141) (0.0066) (0.0104) (0.0949) (0.0600) (0.0545) (0.0179) (0.0890) (Poverty rate – 50) * Post-program -0.0026 0.0046*** 0.0026** -0.0002 0.0561*** 0.0523*** -0.0102*** 0.0504*** period (0.0018) (0.0009) (0.0013) (0.0118) (0.0086) (0.0079) (0.0022) (0.0103) Program * (Poverty rate – 50) * Post- 0.0070*** -0.0056*** -0.0038*** -0.0338*** -0.0580*** -0.0547*** 0.0075*** -0.0467*** program period (0.0017) (0.0008) (0.0012) (0.0111) (0.0084) (0.0077) (0.0020) (0.0094) Age 0.0243*** 0.0040*** 0.0104*** 0.1132*** -0.0058** -0.0059** 0.1571*** 0.3874*** (0.0006) (0.0002) (0.0004) (0.0039) (0.0027) (0.0026) (0.0070) (0.0271) Age squared -0.0003*** -0.0000*** -0.0001*** -0.0015*** 0.0003*** 0.0002*** -0.0081*** -0.0197*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0003) (0.0012) Male -0.0653*** -0.0076*** 0.0865*** 0.9974*** -0.1495*** -0.1382*** 0.0370*** 0.1056*** (0.0044) (0.0021) (0.0030) (0.0298) (0.0182) (0.0170) (0.0057) (0.0239) Ethnic minorities (yes=1, Kinh=0) 0.1805*** -0.0990*** -0.0292*** -0.2248*** -0.2511*** -0.2432*** -0.1293*** 0.7992*** (0.0082) (0.0045) (0.0061) (0.0575) (0.0426) (0.0408) (0.0098) (0.0418) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.2242*** 0.0402*** -0.0947*** -0.6763*** 0.7249*** 0.6527*** 0.2662*** -1.1919*** (0.0140) (0.0054) (0.0088) (0.0874) (0.0418) (0.0396) (0.0394) (0.1543) Observations 128,446 128,446 128,446 128,446 86,980 86,980 50,645 50,645 R-squared 0.190 0.066 0.174 0.154 0.096 0.091 0.195 0.330 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 76 Table A.27. The spillover effect on individual-level outcomes Dependent variables Self- Self- Having Log of The annual The annual Currently Log of employed employed wage job monthly number of number of attending education farm work non-farm (yes=1, wage (all healthcare outpatient school subsidy for Explanatory variables (yes=1, work no=0) workers) visits healthcare (children students no=0) (yes=1, visits aged 6-17) no=0) (1) (2) (3) (4) (5) (6) (7) (8) Districts in provinces with the -0.0113 0.0013 0.0128 0.1267 0.0050 0.0015 0.0098 0.1724 program (0.0252) (0.0105) (0.0197) (0.1430) (0.1154) (0.1149) (0.0147) (0.1849) Age 0.0303*** 0.0063*** 0.0095*** 0.1107*** -0.0121*** -0.0108*** 0.1588*** 0.1932*** (0.0012) (0.0005) (0.0005) (0.0053) (0.0038) (0.0036) (0.0069) (0.0315) Age squared -0.0003*** -0.0001*** -0.0001*** -0.0015*** 0.0004*** 0.0004*** -0.0082*** -0.0100*** (0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0001) (0.0003) (0.0015) Male -0.0614*** -0.0157*** 0.0954*** 1.1164*** -0.1978*** -0.1817*** 0.0170** 0.0220 (0.0085) (0.0040) (0.0049) (0.0510) (0.0252) (0.0239) (0.0072) (0.0217) Ethnic minorities (yes=1, Kinh=0) 0.1322*** -0.0723*** -0.0071 0.0997 -0.1579** -0.1482** -0.1172*** 0.6665*** (0.0180) (0.0080) (0.0114) (0.1105) (0.0671) (0.0622) (0.0146) (0.0900) District fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Constant 0.0714*** -0.0178** -0.0560*** -0.4341*** 0.8067*** 0.7216*** 0.2612*** -0.5382*** (0.0259) (0.0085) (0.0116) (0.1177) (0.0444) (0.0424) (0.0366) (0.1772) Observations 167,748 167,748 167,748 167,748 110,518 110,518 57,720 57,720 R-squared 0.179 0.055 0.149 0.136 0.122 0.118 0.209 0.287 Note: This table reports district fixed-effect RDD-DD regressions of individual-level outcomes. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 77 Table A.28. The district fixed-effect RDD-DD regressions of households’ income using districts with the poverty rate from 35% Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period 0.0283 0.3115 0.7693** -0.5190*** 0.5714 0.8281*** 1.0078*** (0.0821) (0.4804) (0.3213) (0.1741) (0.5429) (0.2908) (0.3853) (Poverty rate – 50) * Post-program -0.0070 -0.0471 -0.0620*** 0.0457*** -0.0487* -0.0443** -0.0175 period (0.0057) (0.0288) (0.0234) (0.0145) (0.0287) (0.0191) (0.0215) Program * (Poverty rate – 50) * Post- 0.0029 -0.0005 0.0472* -0.0409*** 0.0121 0.0322 0.0238 program period (0.0066) (0.0374) (0.0282) (0.0156) (0.0380) (0.0274) (0.0297) Age of household heads 0.0300*** 0.0569*** 0.0624*** 0.1478*** -0.0475*** -0.1009*** -0.0696*** (0.0021) (0.0131) (0.0074) (0.0115) (0.0085) (0.0089) (0.0093) Age squared of household heads -0.0003*** -0.0009*** -0.0007*** -0.0014*** 0.0007*** 0.0013*** 0.0009*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0108 0.1242 0.2039*** 0.7015*** -0.4569*** -0.1650*** -0.4123*** female=0) (0.0138) (0.1177) (0.0584) (0.0614) (0.0628) (0.0551) (0.0533) Ethnic minorities (yes=1, Kinh=0) -0.5423*** -0.1136 -1.8118*** 0.7284*** -0.4677*** 0.4781*** -0.5759*** (0.0369) (0.1356) (0.1192) (0.1035) (0.0974) (0.0842) (0.1036) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.4763*** 2.8778*** 1.5021*** 3.5720*** 4.9592*** 2.1890*** 3.2333*** (0.0742) (0.3380) (0.2094) (0.2810) (0.2691) (0.2528) (0.2262) Observations 55,334 55,334 55,334 55,334 55,334 55,334 55,334 R-squared 0.369 0.163 0.104 0.143 0.246 0.140 0.092 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 78 Table A.29. The district fixed-effect RDD-DD regressions of households’ income using a poverty rate bandwidth of 10% Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0833 -0.3695 1.0748*** -0.4384* 1.0473 0.8605** 1.4858** (0.1215) (0.7388) (0.3877) (0.2360) (0.9295) (0.3744) (0.6418) (Poverty rate – 50) * Post-program 0.0056 -0.0741* -0.0660* 0.0104 -0.0305 -0.0571* -0.0159 period (0.0087) (0.0442) (0.0383) (0.0215) (0.0553) (0.0318) (0.0408) Program * (Poverty rate – 50) * Post- 0.0001 0.2270* -0.0024 0.0005 -0.2262 0.0734 -0.0446 program period (0.0190) (0.1198) (0.0901) (0.0374) (0.1743) (0.0781) (0.1175) Age of household heads 0.0303*** 0.0409** 0.0501*** 0.1466*** -0.0465*** -0.0853*** -0.0584*** (0.0022) (0.0160) (0.0091) (0.0151) (0.0112) (0.0118) (0.0117) Age squared of household heads -0.0003*** -0.0007*** -0.0006*** -0.0014*** 0.0007*** 0.0012*** 0.0008*** (0.0000) (0.0002) (0.0001) (0.0002) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0119 -0.0816 0.1754** 0.7573*** -0.4447*** -0.0625 -0.4033*** female=0) (0.0205) (0.1189) (0.0713) (0.0857) (0.0577) (0.0657) (0.0771) Ethnic minorities (yes=1, Kinh=0) -0.6126*** -0.1217 -1.7998*** 0.7210*** -0.5323*** 0.6820*** -0.7981*** (0.0535) (0.1973) (0.1868) (0.1232) (0.1620) (0.1358) (0.1115) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.5328*** 3.2504*** 1.7557*** 3.4376*** 4.9524*** 1.7241*** 3.1431*** (0.0800) (0.3780) (0.2682) (0.3926) (0.3479) (0.3670) (0.2875) Observations 28,796 28,796 28,796 28,796 28,796 28,796 28,796 R-squared 0.338 0.152 0.095 0.153 0.204 0.142 0.092 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 79 Table A.30. The district fixed-effect RDD-DD regressions of households’ income using a poverty rate bandwidth of 7% Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0914 -0.5461 1.1514*** -0.4597* 1.1423 0.7776** 1.5803** (0.1264) (0.7667) (0.3930) (0.2445) (0.9663) (0.3817) (0.6662) (Poverty rate – 50) * Post-program 0.0086 -0.0587 -0.0836** 0.0089 0.0027 -0.0482 -0.0029 period (0.0093) (0.0469) (0.0380) (0.0231) (0.0566) (0.0334) (0.0430) Program * (Poverty rate – 50) * Post- -0.0026 0.2659** 0.0081 0.0099 -0.3381 0.0885 -0.1123 program period (0.0216) (0.1325) (0.0996) (0.0436) (0.2108) (0.0904) (0.1306) Age of household heads 0.0311*** 0.0417** 0.0473*** 0.1476*** -0.0449*** -0.0808*** -0.0596*** (0.0022) (0.0166) (0.0094) (0.0155) (0.0117) (0.0122) (0.0119) Age squared of household heads -0.0003*** -0.0007*** -0.0006*** -0.0014*** 0.0007*** 0.0011*** 0.0008*** (0.0000) (0.0002) (0.0001) (0.0002) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0243 -0.0345 0.1911** 0.7423*** -0.4253*** -0.0480 -0.4027*** female=0) (0.0203) (0.1203) (0.0726) (0.0880) (0.0598) (0.0674) (0.0793) Ethnic minorities (yes=1, Kinh=0) -0.5951*** -0.0827 -1.7646*** 0.6101*** -0.5391*** 0.6751*** -0.7962*** (0.0533) (0.1938) (0.1894) (0.1033) (0.1663) (0.1386) (0.1133) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.4848*** 3.2073*** 1.7711*** 3.5179*** 4.9014*** 1.6246*** 3.1781*** (0.0778) (0.3741) (0.2726) (0.3957) (0.3622) (0.3822) (0.2919) Observations 26,803 26,803 26,803 26,803 26,803 26,803 26,803 R-squared 0.336 0.150 0.092 0.148 0.206 0.140 0.093 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 80 Table A.31. The district fixed-effect RDD-DD regressions of households’ income using a ‘donut’ sample Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period 0.2085 1.4672 1.0765* -0.4884* 1.2916* 0.8844* 0.7303 (0.1604) (0.9095) (0.6333) (0.2526) (0.7531) (0.4953) (0.4981) (Poverty rate – 50) * Post-program -0.0179 -0.1060 -0.1307** 0.0534** -0.1292** -0.0556* -0.0698* period (0.0171) (0.0762) (0.0596) (0.0265) (0.0589) (0.0313) (0.0372) Program * (Poverty rate – 50) * Post- 0.0107 0.0322 0.1311** -0.0574** 0.0887 0.0433 0.1075** program period (0.0176) (0.0819) (0.0645) (0.0271) (0.0651) (0.0415) (0.0442) Age of household heads 0.0288*** 0.0540*** 0.0673*** 0.1387*** -0.0421*** -0.0973*** -0.0608*** (0.0029) (0.0163) (0.0095) (0.0144) (0.0091) (0.0115) (0.0117) Age squared of household heads -0.0003*** -0.0008*** -0.0008*** -0.0014*** 0.0006*** 0.0012*** 0.0007*** (0.0000) (0.0002) (0.0001) (0.0002) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0102 0.1407 0.2359*** 0.6502*** -0.4488*** -0.1180* -0.3801*** female=0) (0.0179) (0.1559) (0.0755) (0.0742) (0.0793) (0.0628) (0.0580) Ethnic minorities (yes=1, Kinh=0) -0.5679*** -0.1670 -1.7312*** 0.8275*** -0.5734*** 0.5718*** -0.4833*** (0.0524) (0.1819) (0.1645) (0.1361) (0.1275) (0.1055) (0.1428) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.5472*** 2.8292*** 1.3281*** 3.7045*** 4.8671*** 1.9666*** 3.1140*** (0.1037) (0.4317) (0.2560) (0.3473) (0.2964) (0.3361) (0.2909) Observations 36,184 36,184 36,184 36,184 36,184 36,184 36,184 R-squared 0.373 0.155 0.106 0.139 0.273 0.139 0.093 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 81 Table A.32. The DID regressions of households’ income Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program -0.1312 -0.8133 -0.6152 0.1825 -0.4320 -0.6369*** -0.4612** (0.1048) (0.6220) (0.4070) (0.2518) (0.4477) (0.2458) (0.2297) (Poverty rate – 50) 0.0004 0.0841 0.0132 -0.0277 -0.0112 0.0549** 0.0181 (0.0121) (0.0600) (0.0505) (0.0332) (0.0502) (0.0270) (0.0243) Program * (Poverty rate – 50) -0.0033 -0.0613 0.0162 0.0239 0.0309 -0.0420 -0.0292 (0.0127) (0.0663) (0.0533) (0.0336) (0.0543) (0.0302) (0.0259) Program * Post-program period 0.0269 0.8250 0.8851** -0.4883*** 0.5302 0.8165*** 0.6916* (0.0908) (0.5590) (0.3517) (0.1814) (0.5551) (0.3137) (0.3934) (Poverty rate – 50) * Post- 0.0040 -0.1024** -0.0708* 0.0203 -0.0255 -0.0523 -0.0027 program period (0.0092) (0.0448) (0.0386) (0.0235) (0.0528) (0.0338) (0.0395) Program * (Poverty rate – 50) * -0.0090 0.0390 0.0481 -0.0115 -0.0156 0.0420 0.0146 Post-program period (0.0098) (0.0526) (0.0423) (0.0240) (0.0578) (0.0385) (0.0432) Age of household heads 0.0300*** 0.0604*** 0.0649*** 0.1311*** -0.0434*** -0.1005*** -0.0721*** (0.0024) (0.0165) (0.0088) (0.0127) (0.0123) (0.0103) (0.0104) Age squared of household heads -0.0003*** -0.0008*** -0.0008*** -0.0013*** 0.0006*** 0.0013*** 0.0008*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, -0.0163 -0.2734** 0.1499* 0.7678*** -0.4285*** -0.1297** -0.4458*** female=0) (0.0186) (0.1150) (0.0802) (0.0931) (0.0590) (0.0657) (0.0708) Ethnic minorities (yes=1, Kinh=0) -0.5401*** -0.7838*** -1.2738*** 0.8158*** -0.8673*** 0.4293*** -0.8206*** (0.0454) (0.2310) (0.1344) (0.1023) (0.1930) (0.1062) (0.0961) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.5794*** 3.7036*** 1.3208*** 3.6383*** 5.2056*** 2.6051*** 3.8583*** (0.1080) (0.6313) (0.3732) (0.3524) (0.4262) (0.3300) (0.2658) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.297 0.053 0.042 0.090 0.059 0.066 0.045 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 82 Table A.33. The district fixed-effect RDD-DD regressions of households’ income without control variables Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0209 0.4515 0.8195** -0.3693* 0.5160 0.8267*** 0.9735** (0.0968) (0.5037) (0.3844) (0.1956) (0.5730) (0.3156) (0.4203) (Poverty rate – 50) * Post-program 0.0062 -0.0778* -0.0639 0.0063 -0.0270 -0.0542* -0.0120 period (0.0101) (0.0438) (0.0418) (0.0234) (0.0552) (0.0325) (0.0420) Program * (Poverty rate – 50) * Post- -0.0107 0.0323 0.0488 -0.0008 -0.0115 0.0403 0.0167 program period (0.0107) (0.0498) (0.0453) (0.0242) (0.0605) (0.0384) (0.0468) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.8557*** 3.3849*** 1.6412*** 8.0901*** 3.5404*** 0.7814*** 1.1349*** (0.0246) (0.1341) (0.1007) (0.0648) (0.1239) (0.0873) (0.0998) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.287 0.155 0.069 0.087 0.209 0.104 0.075 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 83 Table A.34. The district fixed-effect RDD-DD regressions of households’ income using districts with a large model specification Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0326 0.4782 0.8274** -0.3462* 0.4848 0.8810*** 0.9782** (0.0826) (0.4940) (0.3510) (0.1818) (0.5703) (0.3076) (0.4105) (Poverty rate – 50) * Post-program 0.0073 -0.0766* -0.0664* 0.0079 -0.0268 -0.0568* -0.0143 period (0.0088) (0.0438) (0.0392) (0.0220) (0.0560) (0.0316) (0.0411) Program * (Poverty rate – 50) * Post- -0.0120 0.0230 0.0493 -0.0035 -0.0077 0.0453 0.0205 program period (0.0093) (0.0500) (0.0421) (0.0226) (0.0610) (0.0373) (0.0457) Age of household heads 0.0195*** -0.0609*** 0.0255*** 0.0986*** 0.0406*** -0.0696*** -0.0095 (0.0022) (0.0159) (0.0092) (0.0117) (0.0091) (0.0095) (0.0120) Age squared of household heads -0.0002*** 0.0005*** -0.0003*** -0.0009*** -0.0003*** 0.0008*** 0.0002 (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0363*** -0.3626*** 0.0360 0.6278*** -0.1916*** -0.0918* -0.3783*** female=0) (0.0123) (0.0914) (0.0628) (0.0733) (0.0483) (0.0535) (0.0654) Ethnic minorities (yes=1, Kinh=0) -0.4101*** 0.1067 -1.6502*** 0.7840*** -0.3530*** 0.5370*** -0.6138*** (0.0341) (0.1648) (0.1667) (0.1076) (0.1221) (0.1166) (0.0990) Additional control variables Yes Yes Yes Yes Yes Yes Yes District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.9691*** 4.1633*** 1.6549*** 4.7474*** 3.4436*** 1.6957*** 1.4939*** (0.0642) (0.4084) (0.2953) (0.3386) (0.2956) (0.2891) (0.3439) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.496 0.214 0.119 0.187 0.252 0.159 0.101 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. The additional control variables include education levels of individuals (dummy variables), education levels (dummy variables), household size, the proportion of children in households, the proportion of older members in households. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 84 Table A.35. The district fixed-effect RDD-DD regressions of households’ income using districts with region-specific time trends Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0211 0.1414 0.5139 -0.3346* 0.3496 0.6362* 0.9971** (0.0836) (0.4732) (0.3495) (0.1950) (0.5840) (0.3254) (0.3931) (Poverty rate – 50) * Post-program 0.0028 -0.0747* -0.0494 0.0079 -0.0121 -0.0430 -0.0085 period (0.0081) (0.0408) (0.0378) (0.0230) (0.0588) (0.0311) (0.0387) Program * (Poverty rate – 50) * Post- -0.0065 0.0329 0.0389 -0.0037 -0.0227 0.0343 0.0139 program period (0.0086) (0.0465) (0.0398) (0.0237) (0.0632) (0.0371) (0.0435) Age of household heads 0.0263*** 0.0479*** 0.0553*** 0.1331*** -0.0460*** -0.0943*** -0.0600*** (0.0021) (0.0133) (0.0080) (0.0127) (0.0090) (0.0100) (0.0106) Age squared of household heads -0.0002*** -0.0007*** -0.0006*** -0.0013*** 0.0007*** 0.0012*** 0.0007*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head (male=1, 0.0038 -0.0244 0.1982*** 0.6801*** -0.4093*** -0.0864 -0.3757*** female=0) (0.0181) (0.0980) (0.0625) (0.0753) (0.0505) (0.0558) (0.0669) Ethnic minorities (yes=1, Kinh=0) -0.6194*** -0.2434 -1.7851*** 0.8629*** -0.5281*** 0.6340*** -0.7120*** (0.0484) (0.1838) (0.1596) (0.1221) (0.1361) (0.1185) (0.1015) Region-specific time trend Yes Yes Yes Yes Yes Yes Yes District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.6160*** 2.9543*** 1.5280*** 3.7686*** 4.7509*** 1.9300*** 3.1246*** (0.0838) (0.3548) (0.2404) (0.3276) (0.3100) (0.3053) (0.2559) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.367 0.167 0.109 0.149 0.229 0.152 0.093 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 85 Table A.36. The district fixed-effect RDD-DD regressions of households’ income with clustering the standard error at the village level Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0181 0.4116 0.8039*** -0.3774*** 0.5492** 0.8739*** 0.9973*** (0.0517) (0.3108) (0.2190) (0.1403) (0.2655) (0.1978) (0.2159) (Poverty rate – 50) * Post- 0.0052 -0.0756** -0.0658** 0.0099 -0.0303 -0.0566*** -0.0153 program period (0.0057) (0.0330) (0.0279) (0.0188) (0.0253) (0.0199) (0.0251) Program * (Poverty rate – 50) * -0.0092 0.0284 0.0511* -0.0052 -0.0064 0.0444* 0.0215 Post-program period (0.0061) (0.0360) (0.0294) (0.0192) (0.0280) (0.0238) (0.0268) Age of household heads 0.0262*** 0.0485*** 0.0565*** 0.1330*** -0.0451*** -0.0931*** -0.0598*** (0.0016) (0.0099) (0.0064) (0.0065) (0.0073) (0.0083) (0.0083) Age squared of household heads -0.0002*** -0.0008*** -0.0007*** -0.0013*** 0.0006*** 0.0012*** 0.0007*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head 0.0053 -0.0124 0.2031*** 0.6791*** -0.4089*** -0.0837* -0.3800*** (male=1, female=0) (0.0123) (0.0721) (0.0504) (0.0426) (0.0461) (0.0439) (0.0547) Ethnic minorities (yes=1, Kinh=0) -0.6139*** -0.2264** -1.7972*** 0.8630*** -0.5437*** 0.6164*** -0.7285*** (0.0232) (0.1125) (0.0980) (0.0755) (0.0848) (0.0733) (0.0792) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.6202*** 3.0156*** 1.6227*** 3.7472*** 4.8958*** 1.9215*** 3.0930*** (0.0492) (0.2853) (0.2004) (0.1945) (0.2174) (0.2131) (0.2141) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.364 0.164 0.107 0.148 0.228 0.149 0.091 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 86 Table A.37. The district fixed-effect RDD-DD regressions of households’ income with heteroscedasticity-consistent standard errors Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Program * Post-program period -0.0181 0.4116** 0.8039*** -0.3774*** 0.5492*** 0.8739*** 0.9973*** (0.0260) (0.1738) (0.1290) (0.0752) (0.1210) (0.1051) (0.1231) (Poverty rate – 50) * Post- 0.0052 -0.0756*** -0.0658*** 0.0099 -0.0303** -0.0566*** -0.0153 program period (0.0032) (0.0211) (0.0170) (0.0097) (0.0140) (0.0123) (0.0155) Program * (Poverty rate – 50) * -0.0092*** 0.0284 0.0511*** -0.0052 -0.0064 0.0444*** 0.0215 Post-program period (0.0034) (0.0223) (0.0176) (0.0100) (0.0148) (0.0133) (0.0162) Age of household heads 0.0262*** 0.0485*** 0.0565*** 0.1330*** -0.0451*** -0.0931*** -0.0598*** (0.0015) (0.0094) (0.0061) (0.0056) (0.0067) (0.0070) (0.0071) Age squared of household heads -0.0002*** -0.0008*** -0.0007*** -0.0013*** 0.0006*** 0.0012*** 0.0007*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head 0.0053 -0.0124 0.2031*** 0.6791*** -0.4089*** -0.0837* -0.3800*** (male=1, female=0) (0.0111) (0.0682) (0.0491) (0.0392) (0.0461) (0.0427) (0.0510) Ethnic minorities (yes=1, Kinh=0) -0.6139*** -0.2264*** -1.7972*** 0.8630*** -0.5437*** 0.6164*** -0.7285*** (0.0121) (0.0752) (0.0639) (0.0455) (0.0453) (0.0406) (0.0516) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.6202*** 3.0156*** 1.6227*** 3.7472*** 4.8958*** 1.9215*** 3.0930*** (0.0390) (0.2471) (0.1687) (0.1423) (0.1733) (0.1701) (0.1796) Observations 40,468 40,468 40,468 40,468 40,468 40,468 40,468 R-squared 0.364 0.164 0.107 0.148 0.228 0.149 0.091 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 87 Table A.38. The spillover program effects on households’ income Dependent variables Log of per Log of per Log of per Log of per Log of per Log of per Log of per capita income capita income capita income capita income capita capita public capita income Explanatory variables from wages from nonfarm from farm remittances cash transfers from other production production sources (1) (2) (3) (4) (5) (6) (7) Districts in provinces with the 0.0496 -0.2784 0.0602 0.0823 0.3874 0.0189 0.5494** program (0.0764) (0.3376) (0.2964) (0.1448) (0.2804) (0.1566) (0.2394) Age of household heads 0.0378*** 0.0738*** 0.0587*** 0.1936*** -0.0313*** -0.1124*** -0.0666*** (0.0020) (0.0143) (0.0088) (0.0127) (0.0091) (0.0104) (0.0090) Age squared of household heads -0.0003*** -0.0011*** -0.0008*** -0.0019*** 0.0006*** 0.0014*** 0.0009*** (0.0000) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) Gender of household head 0.0236* 0.0657 0.1049* 0.8948*** -0.4331*** -0.2015*** -0.3998*** (male=1, female=0) (0.0128) (0.1114) (0.0610) (0.0602) (0.0610) (0.0545) (0.0525) Ethnic minorities (yes=1, Kinh=0) -0.4812*** 0.1094 -1.5872*** 0.5220*** -0.4287*** 0.3832*** -0.5147*** (0.0327) (0.1153) (0.1204) (0.0956) (0.0812) (0.0836) (0.1099) District fixed effects Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 8.2226*** 3.0168*** 1.6135*** 2.3383*** 4.5253*** 2.5174*** 2.8994*** (0.0626) (0.3755) (0.2625) (0.3108) (0.3011) (0.2823) (0.2736) Observations 54,870 54,870 54,870 54,870 54,870 54,870 54,870 R-squared 0.352 0.118 0.081 0.155 0.246 0.132 0.098 Note: This table reports district fixed-effect RDD-DD regressions of income and poverty of households using household-level observations. Robust standard errors in parentheses. Standard errors are clustered by district and village-year levels. *** p<0.01, ** p<0.05, * p<0.1. 88