Albert Park, University of MichiganSangui Wang, Renmin University

Community-based Development and Poverty Alleviation An Evaluation of China’s Poor Village Investment Program

Presentation Outline

1. Motivation and 4 research questions

2. Survey data

3. Implementation of the program

4. Methodology

5. Results and discussion of research questions

6. Conclusions and next steps

Community-based developmentMotivation

Definition: “an umbrella term for projects that actively include beneficiaries in their design and management” (Mansuri and Rao, 2004)

Increasing popular model for development assistance: World Bank lending to community-based development

projects increased from $325 million ($2 billion) in 1996 to $3 billion ($7 billion) in 2003 (Mansuri and Rao, 2004).

By 2001, the WB had financed more than 98 social fund projects in 58 countries (Rawlings and Schady, 2002).

Community-based developmentIssues

Increasing popularity of CBD reflects growing belief that sound governance and local accountability are keys to project success (e.g., World Bank, 1999; Easterly 2002)

Many have assumed that decentralization/participatory decision-making improves targeting and impact, although empirical evidence remains limited and mixed

Key issue is whether local elites capture such processes and whether they favor rich or poor, but lack of empirical evidence (Bardhan and Mookherjee, 2005 and 2000)

Growing body of research finds that local governance, and inequality/heterogeneity affects projects chosen, and the amount and quality of public goods (Araujo et al., 2005; Foster and Rosenzweig, 2003; Khwaja, 2004; Besley, Pande, and Rao, 2005)

Community-based developmentPrevious evaluations of community-based development programs Some studies find community-based schemes effectively

target the poor: Galasso and Ravallion (2005): community targeting of

Bangladesh education subsidies were pro-poor but worse with greater land inequality or remoteness

Alderman (2002): communities in Albania targeted better than the center could do using proxy means targeting

Pradhan and Rawlings (2002): Nicaragua social fund was well-targeted to poor

But others find the opposite: Rao and Ibanez (2005): Jamaica social fund did not pick

projects favored by the poor but poor were relatively satisfied with project results

Chase (2002) and Paxson and Schady (2002) find that targeting within communities in Armenia and Peru were not well-targeted toward the poor

World Bank (2002): review of social fund projects concluded that within-community targeting of poor not very effective

No studies use panel household data or examine the effect of public investments on measurable welfare outcomes (e.g., income and consumption)

Evaluating China’s poor village investment programContributions

First quantitative evaluation of the largest community-based development program and largest targeted poverty investment program in the developing world

First evaluation of the effect of community-based development programs on household welfare using panel data

First empirical evidence on relationship between participatory programs and targeting of income and consumption benefits of public investment projects

New evidence on the relationship between governance factors and the benefits of participatory schemes

Research questions

1. To what extent did the program increase public investments in targeted villages?

2. What were the impacts of the program on household income and consumption growth and poverty?

3. How did the investment program affect the propensity of rural laborers to out-migrate?

4. To what extent did governance factors (elite capture, quality of village government, household heterogeneity) mediate program impacts?

Data

2005 World Bank-National Bureau of Statistics village-level survey Sample includes 3036 villages in all poor counties

(199) and one third of non-poor counties (187) in the NBS national rural household survey sample

2004 and 2001 NBS rural household survey data in the same villages10 households per villageNearly all of the households (97 percent) are the

same in 2001 and 2004, enabling construction of a panel dataset

Sample of designated poor villages includes 666 villages and 5500 households (w/panel data)

China’s Poor Village Integrated Development ProgramKey program features

Shift from county targeting to village targeting in 2001, 148,000 villages (21% of all villages) designated as poor

Each village designs integrated investment plan (with participatory component) to coordinate targeted poverty investments from different sub-programs managed by different agencies:

Subsidized Loans by the Agricultural Bank of China Food for Work by National Development Reform

Commission Budgetary Grants by Ministry of Finance

Village selection and plan development were coordinated by of Offices of the inter-ministerial Leading Groups for Poor Area Development at different government levels.

A key goal of the new strategy was to improve targeting and concentrate resources for poverty alleviation in an integrated fashion

Spending on government poverty alleviation programs: 2001-2004…

A large amount of resources are committed to the official poverty reduction programs, accounting for 5-6% of the central government budget

(billion Yuan)

YearSubsidized

loansFood for

workBudgetary

funds Total

2001 18.5 6.0 6.0 30.52002 18.5 6.0 6.6 31.12003 18.5 6.0 7.4 31.92004 18.5 6.0 8.2 32.7

Total 74.0 24.0 28.2 126.2Source: LGOPAD and MOF

What was the process of designation of poor villages? … A Formula-based Approach

A Weighted Poverty Index (WPI) was used to rank villages based on eight indicators

In practice, the local governments were allowed to change some of the indicators and weights on indicators based on local circumstances (decided through participatory approach)

Substantial mistargeting when evaluated solely using income and consumption data

Grain production/person/year

Cash income/person/year % of poor quality houses % of households with

difficulty of access to potable water

% of natural villages with access to reliable electricity supply

% of natural villages with an all-weather road access to county town

% of women with long-term health problems

% of eligible children not attending school

Regional distribution of poor villages … A Formula-based Approach

RegionTotal no.

of villages

No. ofdesignated

poorvillages

% ofvillages

designatedpoor

Share ofpoor

villages(%)

Coastal 249723 20698 8.3 14.0

Northeast 35540 9182 25.8 6.2

Central 225964 48950 21.7 33.0

Southwest 132879 42647 32.1 28.8

Northwest 65151 26654 40.9 18.0

Total 709257 148131 20.9 100.0Source: LGOPAD

The village planning process, in principle From official guidelines provided to local governments

Principles Projects helping the poor should be favored

Participation of households and different groups (e.g. women) should be emphasized

Plans should integrate resources from different sources to maximize efficiency

Plans should be for 3-5 year time horizon and reflect local conditions and causes of poverty

Plans should follow standardized procedure set by county government for easier management and integration

The village planning process, in principle From official guidelines provided to local governments

Procedures Analyze causes of poverty and project solutions, based on

analysis of village-level data and participatory workshops with 10-20 villagers

With support of technicians, conduct SWOT (Strength, Weakness, Opportunity and Threat) and feasibility analysis to help villagers and leaders gain a better understanding of the potential gap between demand and supply as well as potential impacts

More detailed assessment of project beneficiaries, project requirements, including technical and other support, and project implementation (annual schedule, including budget and labor allocations, monitoring and evaluation)

Selection of projects by a vote of entire village

The village planning process, in principle From official guidelines provided to local governments

Basic information collectionVillage leaders & informants

Basic information of the poor village

Household classification

Poverty causes analysis

Projects identification

SWOT & feasibility analysis

Projects selection by poor household

Aggregation of project demand

Aggregation of supporting requirements

Formation of village planning implementation team

Implementation plan design

Village poverty reduction plan

Basic dataset

Graphs/charts/maps on resources, time allocation, etc.

Group discussion

Plenary discussion of all villagers

Beneficiary analysis matrix

Projects matrix

The structure of the village planning document From official guidelines provided to local governments

Section 1: Introduction Section 2: Current situation Section 3: Guidelines and objectives Section 4: Identification of projects Section 5: Total budget Section 6: Supporting system Section 7: Implementation plan Section 8: Organization and management

The village planning process, in practice From field visits and interviews with staff of Poor Area Development Offices

Most official guidelines not followed, especially with regard to participatory methods

Plans often designed by village committees, small group (hamlet) leaders, party representatives, and representatives of households, with help of a township government official (trained by county Poor Area Development Office staff)

County poor area development offices could not “treat” all villages at once because of staff and budget constraints

Plan amounts often far exceeded actual financing, because funds from some programs were not coordinated with plans. Field research found that subsidized loans rarely went toward village plans and Food-for-Work projects sometimes did

By the end of 2004, 55% of poor villages had completed village plans and 37% of poor villages had begun investments based on the plans

LGOPAD reports that a higher percentage (83%) of poor villages completed village plans, but a lower percentage (32%) of poor villages began plan investment

0

10

20

30

40

50

60

2001 2002 2003 2004

％

Poor villages completed plans

Poor villages beginning plan investments

Has village planning been fully implemented?Percent of poor villages completing plan and starting investments based on the plans

Source: World Bank-NBS Special Purpose Survey

Determinants of treatment

Coef. Std. Err.

Mean yrs of schooling 0.084 0.064

Ln(‘01 tot. inv. p.c.) 0.067 0.045

Ln(‘01 gov. inv. p.c.) -0.008 0.092

Poor county 0.164 0.385

Ln(county pop.) -0.344 0.219

Ln(county ag prod.) 0.576 0.842

’01 share w/telephone **0.007 0.003

’01 share w/water -0.002 0.003

’01 share w/road -0.001 0.003

’01 share w/pave rd. -0.007 0.004

Distance to market *-0.024 0.014

Mountainous ***1.082 0.327

Rev. base area -0.375 0.291

Distance to town -0.013 0.015

Distance to county 0.009 0.012

Ln(’01 inc. p.c.) 0.312 0.325

’01 poverty rate 0.334 0.645

Province fixed effects X

N 625

Logit model of determinants of whether designated poor village has begun plan investments

Finding: relatively few statistically significant coefficients, consistent with less endogenous selection

Comparison of mean values for covariates suggests better off villages more likely to begin plan investments

Change in investment structure with plan investments?Investment shares by project type before and after program in treated villages

Electricity

Roads and bridges

Marketing and transport

Schools

Irrigation

Drinking w ater

Afforestation

Terrace/land improvement

Communication

Other

Medical clinics

Electricity

Roads andbridgesMarketing andtransportSchools

Irrigation

Drinking water

Afforestation

Terrace/landimprovementCommunication

Other

Medical clinics

Before After

Nearest neighbor matching with bias adjustment Abadie and Imbens (2006)

1ˆ ˆ( ) ( ) 0

( 1)

1m

l i lli

i

Y X X if IPmY IP

Y if IP

Matching methods rely less on distributional or parametric assumptions, e.g., linear specification, common support

Bias adjustment removes bias caused by differences in the values of matching variables between treatment and matched observations, based on regression function for the controls

Explicit formula for asymptotic variance, but not asymptotically efficient

Standard errors robust to heteroskedasticity

Desirable features of matching estimatorsSmith and Todd (2001, 2005)

Avoid geographic mismatch between matched observations

Common survey questions and survey methods for treatment and control groups

Difference-in-difference matching preferred to cross-sectional matching, especially if above problems exist

Matching methodology

Estimate average treatment effect on the treated, matching control observation to the treated sample

Matching method: weighted nearest neighbors (3 matches per treated observation), with enforcement of exact matching by province (nearly 100 percent)

Propensity scores from logit estimation are used to determine common support, trimming rules

Matching variables: time-invariant or measured prior to start of program. Weighting matrix—Mahalanobis metric, inverse of sample variance/covariance matrix of matching variables.

Common support?Distribution of propensity scores (logit model) for designated poor villages that began program investments and matched poor villages that did not begin investments

0 .2 .4 .6 .8 1Propensity Score

Untreated Treated: On supportTreated: Off support

Research questions

1. To what extent did the program increase public investments in targeted villages?

2. What were the impacts of the program on household income and consumption growth and poverty?

3. How did the investment program affect the propensity of rural laborers to out-migrate?

4. To what extent did governance factors (elite capture, quality of village government, household heterogeneity) mediate program impacts?

To what extent did the program increase public investments in targeted villages?

Sources of investment finance: Government finance

Village funds

Village corvee labor

Other

Factors affecting village investments Matching requirements

Complementary investment opportunities

Substitution

Questions: How much did the program increase government investment in projects

in light of coordination problems and potential substitution? Were government poverty investments complements or substitutes for

villages’ own investments?

Impact of program on change in log(investment per capita) Village matching estimates

Financing source All China West Non-west

Total investment ***2.23(0.539)

***1.54(0.345)

***3.70(1.23)

Total monetary investment ***1.38(0.285)

***1.46(0.331)

***1.37(0.548)

Govt monetary investment ***0.99(0.204)

***1.13(0.313)

***0.85(0.284)

Village monetary investment ***0.64(0.232)

0.07(0.179)

***1.43(0.510)

Corvee labor days -0.19(0.164)

**-0.66(0.320)

**0.38(0.172)

N 588 373 215

Impacts on household income and consumption growth and poverty

Main goal of the poor village investment is to reduce poverty

However, evaluation of impacts on income and consumption is likely to understate program benefits

Most “treatment” villages have not completed plans

There may be lag in program benefits

Health and education benefits not captured

Identification of effects on rich and poor: Within-village estimators of effects on rich and poor (defined

by 2001 median income per capita)

Comparison of restricted (villages with both rich and poor households) and unrestricted samples

Impacts on share of labor that migrates

Motivating concerns Concern that poor are not being included in the benefits

of China’s rapid structural change

Concern about congestion effects of high migration rates

Tension between strategies to raise local productivity or facilitate out-migration?

Ambiguity over predicted effects of different infrastructure investments

Impact on household income, consumption, and migration Village matching estimates

∆ln(inc. pc) ∆ln(cons. pc) ∆migration-share

All villages

All 0.030(0.031)

588

0.010(0.029)

588

-0.025(0.017)

588

Poor -0.039(0.062)

552

0.001(0.042)

552

-0.005(0.018)

552

Rich *0.066(0.035)

484

**0.088(0.036)

484

***-0.052(0.018)

484

Villages with both poor and rich households:

All 0.029(0.037)

448

0.054(0.039)

448

-0.031(0.019)

448

Poor -0.061(0.067)

448

0.006(0.045)

448

0.000(0.020)

448

Rich ***0.096(0.039)

448

***0.114(0.038)

448

**-0.047(0.019)

448

Discussion: why didn’t the poor benefit?

Possible explanations

Lack of capacity to take advantage of public investments

Lack of participation in village planning activities

Exclusion by elites

Governance and program impacts

Relationships of interest Did the program affect governance?

Did governance influence the magnitude or distribution of program benefits?

Governance variables (from principle components analysis)

Education of village leaders village secretary years of education

village mayor years of education

share of village committee members with middle school education or above

Quality of the village committee Number of members

Frequency of meetings

Impact of investment program on village governance in 2004 Village matching estimates

Governance outcome All China West Non-west

Education of leaders ***0.606(0.122)

***0.821(0.204)

***0.754(0.128)

Quality of village committee

0.069(0.099)

0.143(0.105)

-0.057(0.165)

N 583 371 212

Governance and program impacts on the rich and poor within villages Village matching estimates

Education of village leaders Quality of village committee

Low High Diff. Low High Diff.

mean

∆ln(inc. pc) Poor 0.073(0.059)

0.031(0.059)

-0.042 **-0.195(0.082)

0.130(0.096)

***0.325

Rich -0.048(0.072)

0.032(058)

0.080 0.052(0.053)

***0.409(0.070)

***0.357

Diff. -0.121 0.001 0.122 ***0.247 ***0.279 0.032

Mean

∆ln(con. pc) Poor 0.032(0.089)

***-0.121(0.047)

***-0.153 ***-0.159(0.058)

0.047(0.075)

***0.206

Rich -0.057(0.069)

0.023(0.038)

0.080 0.057(0.044)

***0.300(0.059)

***0.243

Diff. -0.089 ***0.144 ***0.233 ***0.216 ***0.253 0.037

Conclusions

There is no evidence that participatory village plans have helped the poor to benefit more from targeted investments or reduced rural poverty in China

Governance factors affect the amount and distribution of benefits under decentralized/participatory decision-making

The relative benefits accruing to the rich increase with the education of village leaders

Greater benefits for both rich and poor increase are associated with high quality village committees

However, this does not necessarily mean that the poor in poor villages have not benefited in comparison to those in non-poor villages…

Priority to improve the design and implementation of the program, possibly with World Bank support