## 2nd_richard_goode_lecture_duflo_presentation

## Source details

**Canonical URL:** [2nd_richard_goode_lecture_duflo_presentation](https://www.imf.org/-/media/files/news/seminars/2nd_richard_goode_lecture_duflo_presentation.pdf)

## Other formats

- [Markdown version](/-/media/files/news/seminars/2nd_richard_goode_lecture_duflo_presentation.pdf.md)
- [Structured JSON version](/-/media/files/news/seminars/2nd_richard_goode_lecture_duflo_presentation.pdf.json)

---

### Attribution and contributors
- Based on work by/with Abhijit Banerjee, Clement Imbert, Michael Greenstone, Santhosh Mathew, Karthik Muralidharan, Paul Niehaus, Nick Ryan, Rohini Pande, Sandip Sukhtankar, November 2016.

### The Economist as plumber — motivation and framing
- Core argument:
  - General principles (e.g., Good governance; Democracy; Don’t waste public funds; Control entitlement programs) are insufficient as a guide to policy.
  - Policymakers need precise, operational guidance on how to act on these ideas.
- Observation:
  - Most corruption advice focuses on direct incentives of civil servants but ignores how the rules of the game and administrative “plumbing” affect outcomes.

### Example 1 — Fixing the pipes: plumbing of public finances (MGNREGS)

- Program description:
  - MGNREGS is a demand-based workfare program providing upto 100 days of work per rural household in India.
  - Federally funded program which transfers funds to state on basis of projected demand. Beneficiary selection and work provision undertaken at village (Gram Panchayat).

- Known problems prior to reform:
  - Niehaus and Sukhtankar (2013) search for 1499 reported MGNREGS workers in the state of Orissa: 50% were ghost workers; those that received work typically received less than reported amount payments.
  - Historically reliance on cash-based management systems based on advances (not expenditures) created delays between transfer and justification, leakage opportunities, and idle funds that increase the budget deficit.

- Policy need:
  - Design operational fund-flow mechanisms to reduce leakage without disrupting service delivery.

### Initial reforms: audit and e‑governance in Bihar

- Federal vigilance enquiry (2010-11) into MGNREGS corruption in Orissa prompted tighter audits in several states including Bihar.
- Timeline of administrative changes:
  - June 2011: Bihar rural development department began requiring weekly audits of ongoing and completed works.
  - November 2011: Clarified that MGNREGS public database be used to sample projects and additional MGNREGS documentation be made available to teams in the field.
  - Between June 2012-13: 64% of GPs in the sample districts were audited at least once.
  - 2010: Bihar introduced Central Planning Scheme Monitoring Scheme (CPSMS) to monitor account balances; created common state pool of funds and opened Zero Balance accounts for districts.
- Operational effect:
  - More frequent auditing and electronic account monitoring set the stage for an experimental fund-flow reform.

### Funds flow: intervention design (experimental reform)

- Fiscal year context:
  - For FY 2012-13 the first tranche of funds (April 1-Sep 1) throughout Bihar (treatment and control) occurred on basis of intended expenditures.
- Randomization and timing:
  - July 2012: Randomization and infrastructure preparation starts.
  - Sept 1 2012: Launch of expenditure-based fund flow system in treatment blocks.
  - Financial reform ran from Sep 1 2012 till March 31 2013.
- Coverage and randomization:
  - Covered 12 districts in Bihar (covering 33 million rural people).
  - In each district GPs in one third of (randomly selected) blocks used reformed system to request MGNREGS funds.
  - Throughout, final stage of payments (from GP account to workers) unchanged.
- Sample notation from source:
  - 12Districts
  - 69T126CBlocks
  - 1002T2029CGP

### How the reform changed transparency and detection

- Under status quo: data entry of worker details lags work by many months.
- Under reform: realtime worker entry ⇒ audit possible sooner.
- Empirical audit evidence:
  - On average one audit per block per month.
  - For projects audited immediately after experiment, detection of malfeasance is 5 pp larger (or double) in T than in C.

### Timeline (key dates and events)

- July 2012: Randomization of blocks into treatment; infrastructure preparation.
- Sept 1st 2012: Launch of expenditure based fund flow system in treatment blocks.
- Sept 18th: State Pool runs dry.
- Dec 11th: State Pool replenished.
- Dec 15th-end Dec: Strike of GP Personnel.
- April 1st 2013: Intervention is rolled back.
- May 15th - July 15th 2013: Endline survey.

### Program take-up and measured fiscal impacts

- Decrease in Spending (CPSMS):
  - Estimated effect = Rs 230,000 per GP for total of 4.1 million USD.
  - Using nrega.nic.in: Rs. 330,000 per GP (slightly higher).
- Balance in GP Accounts:
  - Over the course of the project, the center credited USD 6.3 million less to GPs in the treatment group.
- Reported employment and activity (from MIS/nrega.nic.in):
  - Tables show declines in reported employment measures during intervention windows; sample treatment cell values include 91.88, -130.3, -404.6*, -267.8, -672.4*, -859.5 (with standard errors in parentheses).
  - Panel summaries include:
    - Panel A: Days worked (nrega.nic.in).
    - Panel B: Days per working household (nrega.nic.in).
    - Panel C: Number of working households (nrega.nic.in).
  - Note: Unit of observation is a Gram Panchayat (GP). nrega.nic.in data covers July 2011 to Sept 2013. Specifications include district fixed effects.

### Household survey and assets/projects outcome

- No change in employment (Household Survey, May-July 2013).
- No change in projects built (MIS and MGNREGS Asset survey May-July 2013):
  - Example treatment table excerpt showing 0.0494 -0.2100 0.0172 0.0125 with control mean examples 13.82, 11.62, 0.85, 0.847.
- Interpretation:
  - Reported and field-verified asset construction did not decline despite reduced reported expenditure.

### Results summary (impacts and magnitudes)

- 25% Drop in GP Expenditure: 4.1 million USD in total.
- Drop in funds parked in GP account.
- Combined, during the intervention period a GP received 38% less funds on average: more than 6.3 million USD in total.
- No drop in participation in household survey.
- No decline in assets built according to official reports and found in the field.
- Direct evidence consistent with reduced leakage (via reduced expenditures and reduced dormant funds) but ambiguous interpretation of unmet demand vs. less leakage.

### Ghost-busters: matching job cards to census to measure ghost-workers

- Data sources and scale:
  - NREGA Job Cards Data: 18,513 villages across 195 blocks within 12 districts (6 Million names). Fields: Registration number, name, husband/father name, age, etc.
  - SECC Census Data: 16,480 villages across 195 blocks within 12 districts (33 Million Names). Fields: Name, father name, age, etc.
  - Goal: determine for each household in job cards whether there is a matching household in the census data (person with same name in same village).
- Matching outcomes:
  - Overall match rate: a bit over 50% of working households during the intervention period.
  - Comparison: similar to another leakage calculation (survey-estimated working households divided by NREGA count): 60%.
  - Program reduced fraction of ghost working household by 5%.
- Match-rate table excerpts:
  - Panel A: Match Rate for job cards with one member only — Treatment coefficients include 0.0187**, 0.0181**, 0.0107 with Observations 3,095; 2,868; 2,922 and Mean in Control 0.644, 0.673, 0.698.
  - Panel B: Match Rate for job cards with two members or more — Treatment coefficients include 0.0135**, 0.0126, 0.0104 with Observations 3,093; 2,836; 2,906 and Mean in Control 0.243; 0.282; 0.286.
  - Note: Unit of observation is a Panchayat. Dependent variable is fraction of job cards matched by name with SECC census households. nrega.nic.in data covers July 2011 to March 2014. Treatment is a block-level dummy. All specifications include district fixed effects.

### Impact on functionaries’ declared wealth

- Policy context:
  - Beginning in 2012, functionaries who had worked on MGNREGS were required to report their assets.
- Data examined:
  - Functionaries’ declared assets for 2012-13 (before and during) and 2013-14 (just after).
  - Caveat: data is self-reported (first and second round); caution needed.
- Result:
  - Decline in assets of block and GP officials in the middle of the distribution.
  - Kolmogorov-Smirnov test of stochastic dominance: p=0.057.

### Epilogue and policy traction

- Principal outcomes:
  - The reform reduced dormant funds and reduced leakage (through a decline in expenditure, not an increase in actual delivery).
  - Implementation occurred at large scale under difficult circumstances; initial technical problems increased delays but conditions improved after a few months.
- Political economy and discontinuation:
  - At the end of the fiscal year the system was discontinued due to concerns about increased payment delays, reduced fund flow, and complaints from district officials.
  - Difficulty for state officials to disentangle whether lowered expenditure meant more unmet demand or less leakage.
  - There was no constituency in favor: beneficiaries saw no benefits; officials perceived reduced bribes.
- Subsequent developments:
  - The reform was approved for nationwide rollout in MGNREGS combined with direct payment to beneficiaries.
  - Effort to expand the expenditure-based system to other centrally sponsored programs.
  - Government of India annual spending on Centrally Sponsored Schemes (CSS): roughly 50 billion USD.
  - June 2016: India’s Ministry of Finance issued orders to extend the use of Public Financial Management System for all CSS and emphasized the need to facilitate expenditure based (just-in-time) financing.

### Example 2 (preview) — biometric payments for benefits

- Muralidharan et al (2014) evaluated introduction of biometric smart card for payment of NREGA and pension.
- Status quo problem:
  - Payment at post office created opportunities for leakage (village official could collect a beneficiary’s entire payment and remit only part).
- New system:
  - Biometric smart card plus a network of (female) bank employees handling cash payments with mini ATMs.
- Rollout and scale:
  - Randomized roll out across 158 subdistricts (19 million people).

### Access to payments and speed of payment (Section 2)

- Partial implementation: "50% of payment after roll out (like our program)"
- People get paid faster and with less delay.
- Table 2 notes and selected numeric values:
  - Dependent variables include average time to collect a payment (minutes) and average lag (days) between work done and payment received.
  - Treatment coefficients examples: -21 ∗∗, -21 ∗∗, -5.6, -2.8, -7.1 ∗, -10 ∗∗∗, -2.9 ∗∗∗, -4.7 ∗∗∗.
  - Standard errors examples: (9.3)(8.7)(5.3)(5.6)(3.8)(3.6)(1.1)(1.5).
  - BL GP Mean entries: .08 ∗, .22 ∗∗∗, -.027, .043 (standard errors (.041)(.069)(.09)(.054)).
  - Adj R-squared: .06, .08, .06, .11, .14, .31, .07, .17.
  - Control Mean values: 112 112 777 734 341 212.
  - N. of cases examples: 10252 1018 138143 591142797254142797254.
  - Levels reported: Indiv., Indiv., Indiv., Indiv., Indiv-Week, Indiv-Week, Indiv-Week, Indiv-Week.

### Less leakage: official vs survey reports

- Summary:
  - Less leakage detected when comparing official records to beneficiary survey reports.
- Table 3 panels and notes:
  - Panel (a) (NREGS): household-level average weekly amounts for NREGS work; "Official" = amounts in official muster records; "Survey" = payments received reported by beneficiaries; "Leakage" = difference.
  - Panel (b) (SSP): average SSP monthly amount for May, June, and July; "Official" = official disbursement records; "Survey" = payments received; "Leakage" = difference.
  - Standard errors clustered at mandal level.
- Selected numeric values:
  - Panel (a) Treatment coefficients: 9.9 7.6 35 ∗∗ 35 ∗∗ -25 ∗ -27 ∗∗ with standard errors (12)(12)(15)(15)(13)(13).
  - BL GP Mean values: .12 ∗∗∗, .11 ∗∗∗, .089 ∗∗ with standard errors (.027)(.037)(.038).
  - Adj R-squared: .03 .05 .05 .06 .03 .04.
  - Control Mean examples: 127 127 146 146 -20 -20.
  - N. of cases examples: 5179 5143 5179 5143 5179 5143.
  - Panel (b) Treatment coefficients: 4.55 12 ∗∗ 12 ∗ -7.6 ∗ -7.3 ∗ with standard errors (5.5)(5.6)(5.9)(6.2)(3.9)(4).
  - BL GP Mean entries: .16 ∗, .0081 -.019 with standard errors (.093)(.022)(.024).
  - Control Mean examples: 251 251 236 236 15 15.
  - N. of cases examples: 3354 3151 3354 3151 3354 3151.

### Channels of leakage reduction (who lost from reduced leakage)

- Table 4 overview and definitions:
  - Channels analyzed include "Ghost households", "Other overreporting", and "Bribe to collect" for NREGS; decompositions in Rs for SSP including "Other overreporting" and "Underpayment".
  - "Ghost households": households or all beneficiaries within households confirmed not to exist or permanently migrated before study start (May 28, 2012; May 31, 2010 for baseline).
  - "Other overreporting" (NREGS): jobcards with positive official payments but zero survey amounts (excluding ghosts).
  - "Bribe to collect": bribes paid to receive payments on NREGS.
  - Standard errors clustered at mandal level.
- Selected numeric values:
  - Panel (a) Treatment coefficients examples: -.011 -.011 -.082 ∗∗ -.083 ∗∗ -.0021 -.0028 with standard errors (.02)(.021)(.033)(.036)(.0088)(.0092).
  - BL GP Mean examples: -.013 .019 .014 with standard errors (.067)(.043)(.018).
  - Adj R-squared examples: .02 .02 .05 .04 .01 .01.
  - Control Mean examples: .11 .11 .26 .26 .021 .021.
  - N. of cases examples: 53145 27839 84 370 310 43710366.
  - Panel (b) Treatment coefficients (Rs): -2.7 -2.2 -2.7 -3.3 -2.2 -2.3 with standard errors (2.6)(2.7)(2.9)(3)(1.8)(1.9).
  - BL GP Mean entries: .19 .024 ∗∗∗ -.02 with standard errors (.16)(.0088)(.045).
  - Adj R-squared examples: .01 .01 .01 .01 .01 .01.
  - Control Mean examples: 11 11 1.6 1.6 2.4 2.4.
  - N. of cases examples: 3354 3151 3354 3151 3354 3151.

### Program usage and access

- Table 5 focus:
  - Columns 1-2: proportion of households doing work in 2012 endline study period (May 28-July 15).
  - Columns 3-4: indicator for whether any household member was unable to obtain work despite wanting to work during May (slack demand) or January (peak demand).
  - Columns 5-6: belief that anyone in village who wants NREGS work can get it at any time.
  - Columns 7-8: whether respondent had to pay a bribe to obtain NREGS work during endline.
  - Columns 9-10: whether respondent had to pay a bribe to get on the SSP beneficiary list in years 2011 and 2012.
  - Standard errors clustered at mandal level.
- Selected numeric values:
  - Treatment coefficients examples: .075 ∗∗, .074 ∗∗, -.025 -.031 .026 ∗ .023 -.00016 -.00038 -.046 -.055 with standard errors (.033)(.033)(.027)(.033)(.015)(.015)(.0015)(.0015)(.031)(.039).
  - BL GP Mean entries: .14 ∗∗∗, -.023 -.0056 ∗∗ .025 with standard errors (.037)(.027)(.0027)(.045).
  - Adj R-squared examples: .05 .06 .10 .10 .02 .02 .00 .00 .05 .05.
  - Control Mean examples: .42 .42 .2 .42 .035 .035 .0022 .0022 .075 .075.
  - N. of cases examples: 4978 4944 4478 3453 1479 0475 0723 2690 8587 354.

### Interpretation of program reform effects

- Reported interpretations:
  - More transparency in program implementation.
  - Better service at the final step.
  - Unlike the "pipes" reform, there was little push back by program officials.
  - Possible reason: reform was voluntary and partial (leaves the option to steal from the "ghosts").
  - Also, in this case there was a direct benefit to beneficiaries; in the pipes reform, benefits accrued to the government.
  - Emphasizes importance of better customer service, not just fighting corruption.

### Environmental regulation and audit reform in Gujarat

- Context and motivation:
  - Global agreement to limit CO2 emissions; India will play a major role.
  - Air quality in Delhi is so poor that the Supreme Court ordered government action.
  - Regulatory challenge: how to enforce standards effectively.
  - Quote attributed to Manmohan Singh at Delhi Sustainable Development Summit, 2011: "I must emphasise that standards are not enough. They must also be enforced which is often difficult. . . . It is also necessary to ensure that these regulatory standards do not bring back the License Permit Raj which we sought to get rid of in the wake of economic reforms of the nineties."
- Regulatory framework in India:
  - National-level command-and-control regulations set by the Water Act (1974), Air Act (1981) and Environment Protection Act (1986).
  - Main standards: maximum allowable concentrations for emissions; states can tighten but not relax.
  - Severe, criminal sanctions for violations.
  - Enforcement at state level via State Pollution Control Boards (SPCBs); responsibility increased by later Acts but staffing not increased commensurately.
  - Experiments conducted with Gujarat Pollution Control Board; ongoing work with national ministry.

### Gujarat: industrialization and pollution

- Key facts:
  - 8% annual output growth since 1991-1992.
  - Gujarat has the largest share of post-licensing reform investment of any Indian state.
  - State with most critically polluted industrial clusters (8), including 2 most polluted in the country.
  - Vapi area noted among ten most polluted places on Earth in 2007 (groundwater mercury 96 times higher than WHO).
  - 3 of India’s 5 most polluted rivers and major cities in violation of National Ambient Air Quality Standards.

### Inspections, audits, and third-party auditing design questions

- Regulatory tools and objectives:
  - Key objective: provide regulator information on plant compliance with pollution standards.
  - Inspections: staff engineers and scientists visit plants, observe and sample water and air emissions; summarized in inspection reports.
  - Audits: third-party auditors hired by firms provide annual audit reports to regulator.
- Research questions:
  - Can private third-parties substitute for low state capacity?
  - How should third-party audits be designed to limit conflict of interest?
  - Can rule-based inspection assignment improve pollution compliance?
  - What are social costs and benefits of a rule-based system.
- Third-party auditing trade-offs:
  - Potential advantages: capacity, expertise, flexibility and cost; support for ISO 14001 and carbon offsets.
  - Mixed incentives: audited company hiring the auditor may create conflict of interest; audits have been unreliable in many contexts.

### Audit reform field experiment (Gujarat)

- Scope:
  - Audit treatment piloted for 233 of 473 plants, mostly textile processing.
- Three reform elements implemented:
  1. Random assignment of auditors and fixed payment from a central pool (independence).
  2. Backcheck auditors on performance (monitoring).
  3. In year 2, auditors additionally paid for accuracy relative to backchecks (accuracy incentives).
- Data collection:
  - Final-outlet water and boiler-stack air samples collected.
  - Figures referenced for Audit Readings for Suspended Particulate Matter (SPM) and Quantile Regression Effects of Treatment on Endline Pollution.
  - Interpretation hint: "Response in Tail Reflects Likelihood of Penalty".

### Conclusion and policy implications

- Plumbing problem (implementation/infrastructure of public policy) is under-emphasized in discussions of government effectiveness and state capacity.
- There is substantial knowledge about designing effective public policy infrastructure, but it is neither shared nor put to work widely.
- There is interest among governments in getting inputs to implement policies more effectively.

*The lecture presentation, November 2016.*

### Section 1

### 2nd_richard_goode_lecture_duflo_presentation - Section 1

### Attribution and contributors
- Based on work by/with Abhijit Banerjee, Clement Imbert, Michael Greenstone, Santhosh Mathew, Karthik Muralidharan, Paul Niehaus, Nick Ryan, Rohini Pande, Sandip Sukhtankar, November 2016.

### The Economist as plumber — motivation and framing
- Core argument: general principles (e.g., Good governance; Democracy; Don’t waste public funds; Control entitlement programs) are insufficient as a guide to policy. Policymakers need precise, operational guidance on how to act on these ideas.
- Observation: Most corruption advice focuses on direct incentives of civil servants but ignores how the rules of the game and administrative “plumbing” affect outcomes.

### Example 1 — Fixing the pipes: plumbing of public finances (MGNREGS)
- Program description:
  - MGNREGS is a demand-based workfare program providing upto 100 days of work per rural household in India.
  - Federally funded program which transfers funds to state on basis of projected demand. Beneficiary selection and work provision undertaken at village (Gram Panchayat).
- Known problems prior to reform:
  - Niehaus and Sukhtankar (2013) search for 1499 reported MGNREGS workers in the state of Orissa: 50% were ghost workers; those that received work typically received less than reported amount payments.
  - Historically reliance on cash-based management systems based on advances (not expenditures) created delays between transfer and justification, leakage opportunities, and idle funds that increase the budget deficit.
- Policy need: design operational fund-flow mechanisms to reduce leakage without disrupting service delivery.

### Initial reforms: audit and e‑governance in Bihar
- Federal vigilance enquiry (2010-11) into MGNREGS corruption in Orissa prompted tighter audits in several states including Bihar.
- Timeline of administrative changes:
  - June 2011: Bihar rural development department began requiring weekly audits of ongoing and completed works.
  - November 2011: Clarified that MGNREGS public database be used to sample projects and additional MGNREGS documentation be made available to teams in the field.
  - Between June 2012-13: 64% of GPs in the sample districts were audited at least once.
  - 2010: Bihar introduced Central Planning Scheme Monitoring Scheme (CPSMS) to monitor account balances; created common state pool of funds and opened Zero Balance accounts for districts.
- Operational effect: more frequent auditing and electronic account monitoring set the stage for an experimental fund-flow reform.

### Funds flow: intervention design (experimental reform)
- Fiscal year context:
  - For FY 2012-13 the first tranche of funds (April 1-Sep 1) throughout Bihar (treatment and control) occurred on basis of intended expenditures.
- Randomization and timing:
  - July 2012: Randomization and infrastructure preparation starts.
  - Sept 1 2012: Launch of expenditure-based fund flow system in treatment blocks.
  - Financial reform ran from Sep 1 2012 till March 31 2013.
- Coverage and randomization:
  - Covered 12 districts in Bihar (covering 33 million rural people).
  - In each district GPs in one third of (randomly selected) blocks used reformed system to request MGNREGS funds.
  - Throughout, final stage of payments (from GP account to workers) unchanged.
- Sample notation from source:
  - 12Districts
  - 69T126CBlocks
  - 1002T2029CGP

### How the reform changed transparency and detection
- Under status quo: data entry of worker details lags work by many months.
- Under reform: realtime worker entry ⇒ audit possible sooner.
- Empirical audit evidence:
  - On average one audit per block per month.
  - For projects audited immediately after experiment, detection of malfeasance is 5 pp larger (or double) in T than in C.

### Timeline (key dates and events)
- July 2012: Randomization of blocks into treatment; infrastructure preparation.
- Sept 1st 2012: Launch of expenditure based fund flow system in treatment blocks.
- Sept 18th: State Pool runs dry.
- Dec 11th: State Pool replenished.
- Dec 15th-end Dec: Strike of GP Personnel.
- April 1st 2013: Intervention is rolled back.
- May 15th - July 15th 2013: Endline survey.

### Program take-up and measured fiscal impacts
- Decrease in Spending (CPSMS):
  - Estimated effect = Rs 230,000 per GP for total of 4.1 million USD.
  - Using nrega.nic.in: Rs. 330,000 per GP (slightly higher).
- Balance in GP Accounts:
  - Over the course of the project, the center credited USD 6.3 million less to GPs in the treatment group.
- Reported employment and activity (from MIS/nrega.nic.in):
  - Tables show declines in reported employment measures during intervention windows; sample table elements include:
    - Treatment cell values such as 91.88, -130.3, -404.6*, -267.8, -672.4*, -859.5 (with standard errors in parentheses).
  - Panel summaries:
    - Panel A: Days worked (nrega.nic.in).
    - Panel B: Days per working household (nrega.nic.in).
    - Panel C: Number of working households (nrega.nic.in).
  - Note: Unit of observation is a Gram Panchayat (GP). nrega.nic.in data covers July 2011 to Sept 2013. Specifications include district fixed effects.

### Household survey and assets/projects outcome
- No change in employment (Household Survey, May-July 2013).
- No change in projects built (MIS and MGNREGS Asset survey May-July 2013):
  - Example table excerpt showing Treatment 0.0494 -0.2100 0.0172 0.0125 (standard errors and observations reported).
  - Mean in Control example values: 13.82, 11.62, 0.85, 0.847 in various columns.
- Interpretation: reported and field-verified asset construction did not decline despite reduced reported expenditure.

### Results summary (impacts and magnitudes)
- 25% Drop in GP Expenditure: 4.1 million USD in total.
- Drop in funds parked in GP account.
- Combined, during the intervention period a GP received 38% less funds on average: more than 6.3 million USD in total.
- No drop in participation in household survey.
- No decline in assets built according to official reports and found in the field.
- Direct evidence consistent with reduced leakage (via reduced expenditures and reduced dormant funds) but ambiguous interpretation of unmet demand vs. less leakage.

### Ghost-busters: matching job cards to census to measure ghost-workers
- Data sources and scale:
  - NREGA Job Cards Data: 18,513 villages across 195 blocks within 12 districts (6 Million names). Fields: Registration number, name, husband/father name, age, etc.
  - SECC Census Data: 16,480 villages across 195 blocks within 12 districts (33 Million Names). Fields: Name, father name, age, etc.
  - Goal: determine for each household in job cards whether there is a matching household in the census data (person with same name in same village).
- Matching outcomes:
  - Overall match rate: a bit over 50% of working households during the intervention period.
  - Comparisons: similar to another leakage calculation (survey-estimated working households divided by NREGA count): 60%.
  - Program reduced fraction of ghost working household by 5%.
- Match-rate table excerpt (panel breakdowns):
  - Panel A: Match Rate for job cards with one member only — Treatment coefficients e.g., 0.0187**, 0.0181**, 0.0107 with Observations 3,095; 2,868; 2,922 and Mean in Control 0.644, 0.673, 0.698.
  - Panel B: Match Rate for job cards with two members or more — Treatment coefficients e.g., 0.0135**, 0.0126, 0.0104 with Observations 3,093; 2,836; 2,906 and Mean in Control 0.243; 0.282; 0.286.
  - Note: Unit of observation is a Panchayat. Dependent variable is fraction of job cards matched by name with SECC census households. nrega.nic.in data covers July 2011 to March 2014. Treatment is a block-level dummy. All specifications include district fixed effects.

### Impact on functionaries’ declared wealth
- Policy context: Beginning in 2012, functionaries who had worked on MGNREGS were required to report their assets.
- Data examined: functionaries’ declared assets for 2012-13 (before and during) and 2013-14 (just after).
- Caveat: data is self-reported (first and second round); caution needed.
- Result: decline in assets of block and GP officials in the middle of the distribution.
  - Kolmogorov-Smirnov test of stochastic dominance: p=0.057.

### Epilogue and policy traction
- Principal outcomes:
  - The reform reduced dormant funds and reduced leakage (through a decline in expenditure, not an increase in actual delivery).
  - Implementation occurred at large scale under difficult circumstances; initial technical problems increased delays but conditions improved after a few months.
- Political economy and discontinuation:
  - At the end of the fiscal year the system was discontinued due to concerns about increased payment delays, reduced fund flow, and complaints from district officials.
  - Difficulty for state officials to disentangle whether lowered expenditure meant more unmet demand or less leakage.
  - There was no constituency in favor: beneficiaries saw no benefits; officials perceived reduced bribes.
- Subsequent developments:
  - The reform was approved for nationwide rollout in MGNREGS combined with direct payment to beneficiaries.
  - Effort to expand the expenditure-based system to other centrally sponsored programs.
  - Government of India annual spending on Centrally Sponsored Schemes (CSS): roughly 50 billion USD.
  - June 2016: India’s Ministry of Finance issued orders to extend the use of Public Financial Management System for all CSS and emphasized the need to facilitate expenditure based (just-in-time) financing.

### Example 2 (preview) — biometric payments for benefits
- Muralidharan et al (2014) evaluated introduction of biometric smart card for payment of NREGA and pension.
- Status quo problem: payment at post office created opportunities for leakage (village official could collect a beneficiary’s entire payment and remit only part).
- New system: biometric smart card plus a network of (female) bank employees handling cash payments with mini ATMs.
- Rollout and scale: Randomized roll out across 158 subdistricts (19 million people).

*Based on work by/with Abhijit Banerjee, Clement Imbert, Michael Greenstone, Santhosh Mathew, Karthik Muralidharan, Paul Niehaus, Nick Ryan, Rohini Pande, Sandip Sukhtankar, November 2016.*

### Section 2

### 2nd_richard_goode_lecture_duflo_presentation - Section 2

### Access to payments and speed of payment
- Partial implementation: "50% of payment after roll out (like our program)"
- People get paid faster and with less delay.
- Table 2 findings (dependent variables and notes):
  - The dependent variable in columns 1-4 is the average time taken to collect a payment (in minutes), including unsuccessful trips to payment sites, with observations at the beneficiary level.
  - The dependent variable in columns 5-6 is the average lag (in days) between work done and payment received on NREGS.
  - Columns 7-8 report results for absolute deviations from the median mandal-level lag.
  - Data for columns 5-8 are at the individual-week level; week fixed effects included.
  - Standard errors clustered at mandal level in parentheses.
  - Statistical significance denoted as: ∗ p <0.10, ∗∗ p <0.05, ∗∗∗ p <0.01.
- Selected reported numeric values from Table 2:
  - Treatment coefficients shown: -21 ∗∗, -21 ∗∗, -5.6, -2.8, -7.1 ∗, -10 ∗∗∗, -2.9 ∗∗∗, -4.7 ∗∗∗.
  - Standard errors in parentheses examples: (9.3)(8.7)(5.3)(5.6)(3.8)(3.6)(1.1)(1.5).
  - BL GP Mean entries: .08 ∗, .22 ∗∗∗, -.027, .043 with standard errors (.041)(.069)(.09)(.054).
  - Adj R-squared reported: .06, .08, .06, .11, .14, .31, .07, .17.
  - Control Mean values: 112 112 777 734 341 212 (as presented).
  - N. of cases examples: 10252 1018 138143 591142797254142797254.
  - Levels reported: Indiv., Indiv., Indiv., Indiv., Indiv-Week, Indiv-Week, Indiv-Week, Indiv-Week.
  - Surveys listed in table: NREGS NREGS SSP SS P NREGS NREGS NREGS NREGS (as presented).

### Less leakage: official vs survey reports
- Summary statement: Less leakage detected when comparing official records to beneficiary survey reports.
- Table 3 panels and notes:
  - Regressions include sampled households (NREGS) and beneficiaries (SSP) who a) matched official record or b) were listed in official records but confirmed as "ghosts".
  - "Ghosts" defined as households or beneficiaries not existing or permanently migrated before study start (May 28, 2012; May 31, 2010 for baseline).
  - Panel (a) (NREGS): each observation refers to household-level average weekly amounts for NREGS work during study period (baseline in 2010 - May 31 to July 4; endline in 2012 - May 28 to July 15). "Official" = amounts in official muster records. "Survey" = payments received reported by beneficiaries. "Leakage" = difference between these amounts.
  - Panel (b) (SSP): each observation refers to average SSP monthly amount for May, June, and July. "Official" = amounts in official disbursement records. "Survey" = payments received reported by beneficiaries. "Leakage" = difference.
  - Standard errors clustered at mandal level; significance codes as above.
- Selected numeric values from Table 3:
  - Panel (a) Treatment coefficients: 9.9 7.6 35 ∗∗ 35 ∗∗ -25 ∗ -27 ∗∗ with standard errors (12)(12)(15)(15)(13)(13).
  - BL GP Mean values: .12 ∗∗∗, .11 ∗∗∗, .089 ∗∗ with standard errors (.027)(.037)(.038).
  - Adj R-squared examples: .03 .05 .05 .06 .03 .04.
  - Control Mean examples: 127 127 146 146 -20 -20.
  - N. of cases examples: 5179 5143 5179 5143 5179 5143  (as presented).
  - Panel (b) Treatment coefficients: 4.55 12 ∗∗ 12 ∗ -7.6 ∗ -7.3 ∗ with standard errors (5.5)(5.6)(5.9)(6.2)(3.9)(4).
  - BL GP Mean entries: .16 ∗, .0081 -.019 with standard errors (.093)(.022)(.024).
  - Control Mean examples: 251 251 236 236 15 15.
  - N. of cases examples: 3354 3151 3354 3151 3354 3151 (as presented).

### Channels of leakage reduction (who lost from reduced leakage)
- Table 4 overview:
  - Analyzes channels: "Ghost households", "Other overreporting", and "Bribe to collect" for NREGS (panel a); decomposes amounts (in Rupees) into channels for SSP (panel b).
  - Definitions:
    - "Ghost households": households or all beneficiaries within households confirmed not to exist or permanently migrated before study start (May 28, 2012; May 31, 2010 for baseline).
    - "Other overreporting" (NREGS): jobcards with positive official payments but zero survey amounts (excluding ghosts).
    - "Bribe to collect": bribes paid to receive payments on NREGS.
    - "Other overreporting" (SSP): difference between official reports of beneficiary receipts and what beneficiaries believe they are entitled to.
    - "Underpayment" (SSP): monthly amount paid to receive pensions in May-July 2012.
  - Standard errors clustered at mandal level; significance codes as above.
- Selected numeric values from Table 4:
  - Panel (a) Treatment coefficients examples: -.011 -.011 -.082 ∗∗ -.083 ∗∗ -.0021 -.0028 with standard errors (.02)(.021)(.033)(.036)(.0088)(.0092).
  - BL GP Mean examples: -.013 .019 .014 with standard errors (.067)(.043)(.018).
  - Adj R-squared examples: .02 .02 .05 .04 .01 .01.
  - Control Mean examples: .11 .11 .26 .26 .021 .021.
  - N. of cases examples: 53145 27839 84 370 310 43710366 (as presented).
  - Panel (b) Treatment coefficients (Rs): -2.7 -2.2 -2.7 -3.3 -2.2 -2.3 with standard errors (2.6)(2.7)(2.9)(3)(1.8)(1.9).
  - BL GP Mean entries: .19 .024 ∗∗∗ -.02 with standard errors (.16)(.0088)(.045).
  - Adj R-squared examples: .01 .01 .01 .01 .01 .01.
  - Control Mean examples: 11 11 1.6 1.6 2.4 2.4.
  - N. of cases examples: 3354 3151 3354 3151 3354 3151 (as presented).

### Program usage and access
- Table 5 focus: household-level access to NREGS and SSP.
  - Columns 1-2: proportion of households doing work in 2012 endline study period (May 28-July 15).
  - Columns 3-4: indicator for whether any household member was unable to obtain work despite wanting to work during May (slack demand) or January (peak demand).
  - Columns 5-6: indicator for belief that anyone in village who wants NREGS work can get it at any time.
  - Columns 7-8: indicator for whether respondent had to pay a bribe to obtain NREGS work during endline.
  - Columns 9-10: indicator for whether respondent had to pay a bribe to get on the SSP beneficiary list in years 2011 and 2012.
  - Standard errors clustered at mandal level; significance codes as above.
- Selected numeric values from Table 5:
  - Treatment coefficients examples: .075 ∗∗, .074 ∗∗, -.025 -.031 .026 ∗ .023 -.00016 -.00038 -.046 -.055 with standard errors (.033)(.033)(.027)(.033)(.015)(.015)(.0015)(.0015)(.031)(.039).
  - BL GP Mean entries: .14 ∗∗∗, -.023 -.0056 ∗∗ .025 with standard errors (.037)(.027)(.0027)(.045).
  - Adj R-squared examples: .05 .06 .10 .10 .02 .02 .00 .00 .05 .05.
  - Control Mean examples: .42 .42 .2 .42 .035 .035 .0022 .0022 .075 .075.
  - N. of cases examples: 4978 4944 4478 3453 1479 0475 0723 2690 8587 354 (as presented).

### Interpretation of program reform effects
- Reported interpretations:
  - More transparency in program implementation.
  - Better service at the final step.
  - Unlike the "pipes" reform, there was little push back by program officials.
  - Possible reason: reform was voluntary and partial (leaves the option to steal from the "ghosts").
  - Also, in this case there was a direct benefit to beneficiaries; in the pipes reform, benefits accrued to the government.
  - Emphasizes importance of better customer service, not just fighting corruption.

### Environmental regulation and audit reform in Gujarat
- Context and motivation:
  - Global agreement to limit CO2 emissions; India will play a major role.
  - Air quality in Delhi is so poor that the Supreme Court ordered government action.
  - Regulatory challenge: how to enforce standards effectively.
  - Quote attributed to Manmohan Singh at Delhi Sustainable Development Summit, 2011: "I must emphasise that standards are not enough. They must also be enforced which is often difficult. . . . It is also necessary to ensure that these regulatory standards do not bring back the License Permit Raj which we sought to get rid of in the wake of economic reforms of the nineties."
- Regulatory framework in India:
  - National-level command-and-control regulations set by the Water Act (1974), Air Act (1981) and Environment Protection Act (1986).
  - Main standards: maximum allowable concentrations for emissions; states can tighten but not relax.
  - Severe, criminal sanctions for violations.
  - Enforcement at state level via State Pollution Control Boards (SPCBs); responsibility increased by later Acts but staffing not increased commensurately.
  - Experiments conducted with Gujarat Pollution Control Board; ongoing work with national ministry.

### Gujarat: industrialization and pollution
- Key facts reported:
  - 8% annual output growth since 1991-1992.
  - Gujarat has the largest share of post-licensing reform investment of any Indian state.
  - State with most critically polluted industrial clusters (8), including 2 most polluted in the country.
  - Vapi area noted among ten most polluted places on Earth in 2007 (groundwater mercury 96 times higher than WHO).
  - 3 of India’s 5 most polluted rivers and major cities in violation of National Ambient Air Quality Standards.

### Inspections, audits, and third-party auditing design questions
- Regulatory tools and objectives:
  - Key objective: provide regulator information on plant compliance with pollution standards.
  - Inspections: staff engineers and scientists visit plants, observe and sample water and air emissions; summarized in inspection reports.
  - Audits: third-party auditors hired by firms provide annual audit reports to regulator.
  - Research questions:
    - Can private third-parties substitute for low state capacity?
    - How should third-party audits be designed to limit conflict of interest?
    - Can rule-based inspection assignment improve pollution compliance?
    - What are social costs and benefits of a rule-based system?
- Third-party auditing trade-offs:
  - Potential advantages: capacity, expertise, flexibility and cost; support for ISO 14001 and carbon offsets (citations in original).
  - Mixed incentives: audited company hiring the auditor may create conflict of interest; audits have been unreliable in many contexts (corporate accounting scandals, credit ratings during financial crisis).

### Audit reform field experiment (Gujarat)
- Audit treatment piloted for 233 of 473 plants, mostly textile processing.
- Three reform elements implemented:
  1. Random assignment of auditors and fixed payment from a central pool (independence).
  2. Backcheck auditors on performance (monitoring).
  3. In year 2, auditors additionally paid for accuracy relative to backchecks (accuracy incentives).
- Data collection referenced: final-outlet water and boiler-stack air samples; figures shown for Audit Readings for Suspended Particulate Matter (SPM) and Quantile Regression Effects of Treatment on Endline Pollution.
- Interpretation hint: "Response in Tail Reflects Likelihood of Penalty" (figure label).

### Conclusion and policy implications
- Plumbing problem (implementation/infrastructure of public policy) is under-emphasized in discussions of government effectiveness and state capacity.
- There is substantial knowledge about designing effective public policy infrastructure, but it is neither shared nor put to work widely.
- There is interest among governments in getting inputs to implement policies more effectively.

---


_Source: https://www.imf.org/-/media/files/news/seminars/2nd_richard_goode_lecture_duflo_presentation.pdf_