can bureaucrats be paid like ceos? school managers incentives for anemia reduction in rural china...

Can Bureaucrats be Paid Like CEOs?School Managers Incentives for Anemia Reduction in

Rural China

Renfu Luo, Chinese Academy of SciencesGrant Miller, Stanford Medical School and NBER

Sean Sylvia, University of MarylandScott Rozelle, Freeman Spogli Institute, Stanford University

Marcos Vera-Hernández, University College London

We gratefully acknowledge support from the National Heart, Lung, and Blood Institute (5R01Hl10602302)

Public Service Delivery in Developing Countries

Poor quality of public services in developing countries Growing number of health examples

Banerjee, Deaton, and Duflo (2004) Das, Hammer, and Leonard (2008) Berendes et al. (2011) Das, Holla, Das, Mohanan, Tabak, and Chan (2012)

Two prominent explanations: 1. Lack of resources and skills 2. Weak or misaligned incentives (inc. poor

accountability)

Addressing Weak Incentives: Performance Pay

Usual approach in the private sector: provide monetary incentives

Will that work in the public sector?

Concerns about multi-tasking Some tasks are difficult to measure, they are difficult to incentive Individuals might neglect them in favor of the incentivized ones

Intrinsically motivated agents (risk of crowding out)

Bureaucratic mindset: stability, aversion to change

Addressing Weak Incentives: Performance Pay

Not the first ones… Reward teachers for improved test scores or attendance (e.g. Lavy

2002; Banerjee and Duflo 2006; Muralidharan and Sundararaman 2011; Behrman et al. 2015)

Reward health workers for condoms distributed (e.g. Ashraf et al. 2013)

Reward organizations or local government for health service indicators (e.g. Bloom et al. 2006; Basinga et al. 2011; Olken et al. 2011; de Walque 2015)

Some find that monetary incentives are ineffective Ashraf et al. (2012) and Behrman et al. (2015), for example

Some common aspects in the literature: Front-line workers Reward of inputs rather than outputs (health literature) Lack comparisons with increasing resources (Lavy 2002 and

Muralidharan and Sundararaman 2011 are exception)

Our main contributions:

Focus on higher level public sector workers (“bureaucrats”): school principals Most literature focuses on lower level ones: teachers,

nurses, etc. (exception of Rasul and Rogger 2014) Interesting because they have control over resources, and

more autonomy to implement innovative solutions

Study the role of incentives in combination with resources The literature does one or the other Natural to do this in our context given that bureaucrats

have control over the allocation of resources Are incentives and resources complements or substitutes?

Our main contributions:

Reward health output (anemia rates) rather than inputs (intake of iron supplements, intake of meat)

Natural when the incentive is given to someone with more autonomy

She can use local knowledge to better achieve the outcome So we provide them with information on how they could

reduce anemia but we are not prescriptive on how they could do it

Other contributions

Incentive size: small vs. large: Small incentives are effective: Thornton 2008; Banerjee et

al. 2010; Karlan et al. 2011 Small incentives are ineffective: Gneezy and Rustichini

(2000)

Focus on the price effect of incentives Shutting down other channels through which providing

incentives might have an effect For instance: Raise salience on the outcome variable

Health interventions in schools: school-based programs might be a cost-effective way to deliver nutritional interventions targeting school-aged children (Bundy & Guyatt 1996; Orazem et al. 2008)

Outline

1. Anemia in Rural China: Causes, Consequences and Interventions

2. Model3. Experimental Design, Data Collection,

Empirical Strategy4. Results5. Conclusions

Anemia: Causes and Consequences

Anemia: Low red blood cell concentration, too little oxygen to the body

Most common cause in China: Iron deficiency (accounting for 85-95%) Iron required for hemoglobin, the oxygen-carrying protein in blood Anemia is severe form of iron deficiency; iron deficiency 2-3 times

more prevalent

Consequences: lethargy, fatigue, physical and cognitive impairment Particularly harmful in childhood, detrimental to human capital

development Inferior educational outcomes among school-aged children (Nokes

et al. 1998, Bobonis et al. 2006)

Anemia affects nearly 25% of school-age children worldwide 20-60% anemia prevalence among children in rural areas of

western China

Interventions

How to reduce anemia:

Fortification of staples (might not be feasible in areas that grow their own food)

Iron supplements/multivitamins (might need monitoring, secondary effects…)

Increase dietary iron intake Red meat (heme iron) Green vegetables (non-heme iron) Fruits and vegetables high in complementary Vitamin C Improves iron absorption

Outline




Model

subject to:

Model

First Order Conditions :

Model

Comparative statics:

Model

Substitution between incentives and resources:

Experimental Design

All principals uniformly provided with information about: Causes and consequences of anemia Known effective strategies to address anemia Relationship with academic performance based on peer-

reviewed studies from China

Randomly assignment using a 3×2 design:No Incentive Small Incentive Large Incentive

Small Block Grant 32 schools 20 schools 33 schools

Large Block Grant 33 schools 20 schools 32 schools

Design: to look at incentives, resources, and their interaction Keeping information and “salience” constant

Block Grants

Small block grant: 0.3 yuan/student/day (~$0.05) Sufficient to purchase multivitamins On average, 7,452 yuan/school during the study

(~$1,212)

Large block grant: 0.7 yuan/student/day (~$0.11) Sufficient to purchase 60 grams of red meat 3x per week On average, 17,388 yuan/school during the study

(~$2,829)

Principals can use grants at their discretion: Any strategy to reduce anemia Other school functions (e.g. school supplies)

No monitoring

Anemia Reduction Incentives

Small Incentives PaySmall = 12.5 yuan (~$2) x (Anemicb – Anemice) if (Anemicb –

Anemice) ≥ 0

Large Incentive PayLarge = 125 yuan (~$20) x (Anemicb – Anemice) if (Anemicb –

Anemice) ≥ 0

The large incentive was calibrated to be 2 monthly wages at the expected effect size, while the small was calibrated to be 20% of a monthly wage

Sampling & Randomization

Sampling frame: Primary schools in 25 officially designated poor counties in Gansu, Qinghai, and Shaanxi

170 schools randomly selected for inclusion 1 school per township

Study Sites

Data Collection

Baseline survey (September 2011)

Endline survey (May 2012)

Little attrition (6% overall)

Estimation

Main Specification (for child i in school j located in county c) in sample of children anemic at baseline:

Yijc=α + T΄jc β + x΄ijϒ + μc + λj + εijc

Yijc Outcome of interest at endline

Tjc Vector of treatment dummies and interactions Small incentive, Large Incentive, Large Block Grant, (Small Incentive)X(Large Grant), (Large Incentive)X(Large Grant)

xij Baseline student, household, school characteristics

μc County fixed effects

λj Randomization stratum fixed effects

Adjustments for multiple testing

• It is becoming common practice to correct p-values if multiple outcome variables are used

• If we are testing many outcome variables, it is not unlikely that one of the is statistically different from zero, although the null hypothesis is true

• But the same problem of multiplicity applies to the case of one outcome variable but several treatment variables

• We adjust the p-values to keep the probability of rejecting at least one when all are true (Family Wise Error Rate) at standard levels of 0.05 or 0.10 (Westfall and Young, 1993)

Building indices to explore inputs

• Estimate effect of the treatments on:• Vitamins intake• Food intake• Provision of information to households

• We have several variables measuring each of these inputs and from several sources (student questionnaire, household questionnaire…)• We build an index for them

Outline




Small Grant, No Incentive Coefficient (standard error) on:

NP-value: Equality

of All Groups

Mean SD Small Incentive

Large Incentive Large Grant

(Small Incentive)X

(Large Grant)

(Large Incentive)X

(Large Grant)

A: Child Characteristics1. Hemoglobin Concentration (g/L) 134.191 12.912 -0.912 -1.192 0.514 0.140 -0.021 8398 0.541(1.127) (1.009) (1.028) (1.501) (1.476)

2. Anemic (0/1) 0.233 0.423 0.024 0.017 -0.015 -0.001 0.003 8398 0.222(0.017) (0.019) (0.018) (0.024) (0.025)

3. Age (years) 10.713 1.173 -0.172 -0.041 -0.030 0.352* -0.013 8398 0.379(0.128) (0.111) (0.106) (0.185) (0.144)

4. 5th Grade (0/1) 0.531 0.499 -0.002 0.001 -0.005 0.007 0.001 8398 0.941(0.006) (0.006) (0.008) (0.011) (0.010)

5. Female (0/1) 0.485 0.500 0.003 -0.008 -0.009 0.024 0.010 8398 0.808(0.020) (0.017) (0.019) (0.030) (0.025)B: School Characteristics

6. Number of Students 207.094 64.823 -1.276 3.623 -5.396 25.344 12.357 170 0.797(17.567) (14.959) (16.043) (25.554) (20.856)

7. Has Kitchen (0/1) 0.063 0.246 0.141 0.074 0.059 -0.075 -0.068 170 0.681(0.101) (0.075) (0.083) (0.162) (0.120)

8. Student-Teacher Ratio 16.228 4.227 2.538* 0.893 -0.286 -1.506 1.064 170 0.257(1.354) (1.210) (1.159) (1.911) (1.657)9. Time to Furthest Village Served (mins) 62.031 36.695 12.218 -2.281 3.878 -7.346 3.764 170 0.921(13.109) (11.564) (12.945) (21.467) (17.794)10. Percent Boarding Students (%) 5.327 11.404 1.511 0.106 0.610 -0.079 -1.611 170 0.991(4.112) (3.006) (3.492) (6.293) (5.179)

Baseline Balance

Result 1: Large Incentives

Our main motivation

Large incentives are effective at reducing anemia (amongst those anemic at baseline)

Large reduction of almost 14 percentage points from a base of 36%

Result 1: Large incentives

Children Anemic at Baseline Full Sample

Hemoglobin Concentration

(g/L)

Anemic at Endline


(g/L)

Anemic at Endline

1. Small Incentive -0.387 -0.012 1.054 -0.028

(1.101) (0.040) (0.987) (0.020)0.792 0.972 0.747 0.587

2. Large Incentive 2.567 -0.138* 0.918 -0.045(1.044) (0.039) (0.946) (0.022)0.285 0.064 0.767 0.373

3. Large Grant 4.205** -0.145** 2.872 -0.073**(1.123) (0.038) (0.989) (0.021)0.045 0.047 0.117 0.049

4. (Small Incentive)X(Large Grant) 1.445 -0.042 -0.857 0.027(1.541) (0.056) (1.340) (0.027)0.664 0.888 0.829 0.647

5. (Large Incentive)X(Large Grant) -4.580 0.196* -3.312 0.086(1.586) (0.058) (1.404) (0.031)0.173 0.072 0.235 0.149

6. Observations 1923 1923 7945 79457. Mean in Small Grant, No Incentive Group 129.901 0.364 136.334 0.176

Effects on inputs (using indices)

Mean in Small Grant, No Incentive

Group

Coefficient (standard error) and P-values on:

NSmall Incentiv

e

Large Incentiv

eLarge Grant

(Small Incentiv

e)X (Large Grant)

(Large Incentiv

e)X (Large Grant)

A. Children Anemic at Baseline

1. Index: Vitamin Provision-0.050

0.141 0.158 0.242* -0.272 -0.2901921(0.084) (0.081) (0.072) (0.115) (0.106)

0.241 0.241 0.0522 0.176 0.1226

2. Index: Food -0.040

-0.020 0.134 0.198* -0.117 -0.317**1923(0.050) (0.050) (0.066) (0.090) (0.090)

0.7892 0.1032 0.0685 0.4888 0.0285

3. Index: Vitamin and Food -0.040

0.072 0.145* 0.224** -0.214* -0.302**1923(0.054) (0.052) (0.048) (0.078) (0.073)

0.3034 0.0997 0.0045 0.0997 0.0098

B. Full Sample


0.192* 0.191* 0.250** -0.367** -0.261*7920(0.072) (0.075) (0.060) (0.099) (0.095)

0.0766 0.0766 0.0067 0.0156 0.0766

6. Index: Food-0.020

0.055 0.113* 0.117 -0.099 -0.1567945(0.037) (0.040) (0.049) (0.059) (0.070)

0.2781 0.0846 0.1521 0.2781 0.1669

7. Index: Vitamin and Food-0.070

0.133** 0.155** 0.193** -0.258** -0.213**7945(0.048) (0.050) (0.043) (0.066) (0.067)

0.0324 0.0324 0.0025 0.0095 0.0324

Result 1: Large incentives

What about the hypothesis that principals might engage households to prevent compensatory behaviors?

Plausibly, a consequence of rewarding outputs rather than inputs

Result 1: Large incentives were effective

working through homes? (remember rewarding outputs not inputs)

Mean in Small Grant, No Incentive Group

Coefficient (standard error) and adjusted P-value on:

NSmall Incentive

Large Incentive

Large Grant

(Small Incentive)X

(Large Grant)

(Large Incentive)X

(Large Grant)


2. Index: Food Provision

-0.040

-0.020 0.134 0.198* -0.117 -0.317**

1923(0.050) (0.050) (0.066) (0.090) (0.090)

0.7892 0.1032 0.0685 0.4888 0.0285

Index: Food Provision Home

-0.050

0.079 0.192* 0.191 -0.163 -0.297

1923(0.062) (0.065) (0.077) (0.107) (0.111)

0.3563 0.0808 0.1245 0.3563 0.1118

Index: Food Provision School

-0.040

-0.078 0.109 0.219 -0.103 -0.361*

1923(0.064) (0.062) (0.088) (0.110) (0.110)

0.5337 0.4179 0.1991 0.5337 0.0572

B. Full Sample8. Index: Food Provision

-0.020

0.055 0.113* 0.117 -0.099* -0.156

7945(0.037) (0.040) (0.049) (0.059) (0.070)

0.2781 0.0846 0.1521 0.2781 0.1669


0.010

0.135* 0.133* 0.105 -0.105 -0.106

7945(0.049) (0.048) (0.054) (0.077) (0.077)

0.0914 0.0914 0.259 0.4441 0.4441


-0.040

0.012 0.113 0.148 -0.123 -0.223

7945(0.051) (0.056) (0.066) (0.078) (0.091)

0.8454 0.236 0.2004 0.3042 0.1626


Group

Coefficient (standard error) on:

N

Small Incentive

Large Incentive Large Grant (Small Incentive)X

(Large Grant)(Large Incentive)X

(Large Grant)

1. Number of school-wide parent meetings attended this semester 1.440 0.019 0.021 0.676*** -0.978*** -0.682** 1357(0.207) (0.198) (0.206) (0.301) (0.286)2. Number of individual meetings with teacher or principal this semester 0.870 0.110 0.503** 0.660*** -0.735** -0.855** 1345(0.185) (0.231) (0.251) (0.325) (0.376)3. School contacted household about student nutrition this semester 0.430 -0.016 0.118* 0.062 -0.062 -0.140 1455

(0.077) (0.066) (0.095) (0.124) (0.126)

4. Household told to give student foods rich in iron 0.270 0.042 0.115** 0.141** -0.085 -0.273*** 1200(0.067) (0.055) (0.071) (0.105) (0.101)

5. Parent reports knowing of anemia 0.770 0.055 -0.044 0.017 -0.050 0.037 1473(0.046) (0.043) (0.047) (0.069) (0.066)

6. Parent correctly identifies foods that can prevent anemia (iron rich foods) 1.770 -0.021 0.295 0.176 -0.018 -0.410 1516

(0.201) (0.236) (0.236) (0.317) (0.331)

7. Summary Index-0.060 0.043 0.139 0.232** -0.318** -0.354** 1377

(0.085) (0.086) (0.116) (0.152) (0.150)NOTES. Data source: authors' survey. Each row shows coefficient estimates (and robust standard errors) from a separate regression estimated using equation (1) (controling for baseline hemoglobin concentration, student age, student grade, student sex, number of students in the school, whether the school has a canteen, student teacher ratio, distance to the furthest village served, percent of boarding students, whether the school has implemented the "Free Lunch" policy, county dummy variables, and dummy variables for randomization strata). Regressions estimated using only sample of children anemic (altitude adjusted hemoglobin concentration<120 g/L) at baseline. Dependent variables are shown at left. The final row shows estimates from a regression with an index summarizing all other variables as the dependent variable. This summary index was computed using the method discussed in Anderson (2008). *, **, and *** indicate significance at 10%, 5% and 1%.

Communication with Households (Unadjusted)

More communication with households attributable to large incentives

Grant Use: no differences

The fact that the money was spent in the same way, it means that the effect of the incentive is not on how they spent the money (vitamins vs. meat) but on how efficiently they worked with the resources they had.

Result 2: small incentive

… but not any incentive makes the job

The small incentive is ineffective

Inconsistent with some recent literature on the effect of small incentives on household decisions (Thornton 2008; Banerjee et al. 2010; Karlan et al. 2011)

In our case, we do not let small incentives to raise salience

Although all principals uniformly received information about anemia

Result 2: Small incentives



(g/L)

Anemic at Endline


(g/L)

Anemic at Endline

1. Small Incentive -0.387 -0.012 1.054 -0.028

(1.101) (0.040) (0.987) (0.020)0.792 0.972 0.747 0.587

2. Large Incentive 2.567 -0.138* 0.918 -0.045(1.044) (0.039) (0.946) (0.022)0.285 0.064 0.767 0.373

3. Large Grant 4.205** -0.145** 2.872 -0.073**(1.123) (0.038) (0.989) (0.021)0.045 0.047 0.117 0.049




Effects on inputs (using indeces)


Group


NSmall Incentiv

e

Large Incentiv

eLarge Grant

(Small Incentiv

e)X (Large Grant)

(Large Incentiv

e)X (Large Grant)



0.141 0.158 0.242* -0.272 -0.2901921(0.084) (0.081) (0.072) (0.115) (0.106)

0.241 0.241 0.0522 0.176 0.1226


-0.020 0.134 0.198* -0.117 -0.317**1923(0.050) (0.050) (0.066) (0.090) (0.090)

0.7892 0.1032 0.0685 0.4888 0.0285


0.072 0.145* 0.224** -0.214* -0.302**1923(0.054) (0.052) (0.048) (0.078) (0.073)

0.3034 0.0997 0.0045 0.0997 0.0098

B. Full Sample


0.192* 0.191* 0.250** -0.367** -0.261*7920(0.072) (0.075) (0.060) (0.099) (0.095)

0.0766 0.0766 0.0067 0.0156 0.0766


0.055 0.113* 0.117 -0.099 -0.1567945(0.037) (0.040) (0.049) (0.059) (0.070)

0.2781 0.0846 0.1521 0.2781 0.1669


0.133** 0.155** 0.193** -0.258** -0.213**7945(0.048) (0.050) (0.043) (0.066) (0.067)

0.0324 0.0324 0.0025 0.0095 0.0324



(g/L)Anemic at

Endline


(g/L)Anemic at

Endline

9. Additional Hypotheses: Adjusted p-values

Large vs. Small Incentive 0.285 0.089 0.908 0.647

Large Incentive vs. Large Grant 0.597 0.972 0.301 0.587

Large Incentive + (Large Incentive)X(Large Grant)

0.511 0.650 0.209 0.373

And the effects of the small and large incentive are statistically different

Result 2: small incentive

Although the small incentive is in average not effective

The is some interesting heterogeneity which is consistent with behavioral theory

Crowding out of intrinsic or pro-social motivation

Crowding-out of pro-social / intrinsic motivation

Dependent Vble: Anemic at follow-up

VBLEVBLEx(Small

Incentive) VBLEx(Large

Incentive)VBLEx(Large

Grant)

VBLEx(Small Incentive)X

(Large Grant)

VBLEx(Large Incentive)X

(Large Grant)VBLE=Prosocial Motivation Index 0.230 -0.003 0.021 -0.148 -0.043

(0.070) (0.047) (0.046) (0.093) (0.075)0.09 0.95 0.94 0.58 0.94

VBLE=(Prosocial Motivation Index>Median)

0.310 0.061 0.156 -0.291 -0.200(0.079) (0.093) (0.085) (0.117) (0.123)

0.04 0.63 0.4 0.25 0.4

VBLE=Intrinsic Motivation Index 0.104 -0.010 -0.014 -0.102 -0.016(0.040) (0.051) (0.042) (0.056) (0.061)

0.24 0.98 0.98 0.51 0.98

VBLE=(Intrinsic Motivation Index>Median)

0.092 -0.043 -0.092 0.093 0.032(0.083) (0.091) (0.076) (0.128) (0.119)

0.6 0.98 0.98 0.98 0.98

The higher the prosocial/intrinsic motivation of the school principal, the worst are the outcomes of the Small Incentive group

So, there is some evidence of crowding out of prosocial/intrinsic motivation

Result 3: Large grants

Larger reductions in anemia are achieved in schools with large grants vs. those with smaller grants

Result 3: Large grants



(g/L)

Anemic at Endline


(g/L)

Anemic at Endline

1. Small Incentive -0.387 -0.012 1.054 -0.028

(1.101) (0.040) (0.987) (0.020)0.792 0.972 0.747 0.587

2. Large Incentive 2.567 -0.138* 0.918 -0.045(1.044) (0.039) (0.946) (0.022)0.285 0.064 0.767 0.373

3. Large Grant 4.205** -0.145** 2.872 -0.073**(1.123) (0.038) (0.989) (0.021)0.045 0.047 0.117 0.049




Effects on inputs (using indeces)


Group


NSmall Incentiv

e

Large Incentiv

eLarge Grant

(Small Incentiv

e)X (Large Grant)

(Large Incentiv

e)X (Large Grant)



0.141 0.158 0.242* -0.272 -0.2901921(0.084) (0.081) (0.072) (0.115) (0.106)

0.241 0.241 0.0522 0.176 0.1226


-0.020 0.134 0.198* -0.117 -0.317**1923(0.050) (0.050) (0.066) (0.090) (0.090)

0.7892 0.1032 0.0685 0.4888 0.0285


0.072 0.145* 0.224** -0.214* -0.302**1923(0.054) (0.052) (0.048) (0.078) (0.073)

0.3034 0.0997 0.0045 0.0997 0.0098

B. Full Sample


0.192* 0.191* 0.250** -0.367** -0.261*7920(0.072) (0.075) (0.060) (0.099) (0.095)

0.0766 0.0766 0.0067 0.0156 0.0766


0.055 0.113* 0.117 -0.099 -0.1567945(0.037) (0.040) (0.049) (0.059) (0.070)

0.2781 0.0846 0.1521 0.2781 0.1669


0.133** 0.155** 0.193** -0.258** -0.213**7945(0.048) (0.050) (0.043) (0.066) (0.067)

0.0324 0.0324 0.0025 0.0095 0.0324

Result 3: Large grant

Some tendency towards working more through the school than the home


Coefficient (standard error) and adjusted P-value on:

NSmall Incentive

Large Incentive

Large Grant

(Small Incentive)X

(Large Grant)

(Large Incentive)X

(Large Grant)


2. Index: Food Provision

-0.040

-0.020 0.134 0.198* -0.117 -0.317**

1923(0.050) (0.050) (0.066) (0.090) (0.090)

0.7892 0.1032 0.0685 0.4888 0.0285


-0.050

0.079 0.192* 0.191 -0.163 -0.297

1923(0.062) (0.065) (0.077) (0.107) (0.111)

0.3563 0.0808 0.1245 0.3563 0.1118


-0.040

-0.078 0.109 0.219 -0.103 -0.361*

1923(0.064) (0.062) (0.088) (0.110) (0.110)

0.5337 0.4179 0.1991 0.5337 0.0572

B. Full Sample8. Index: Food Provision

-0.020

0.055 0.113* 0.117 -0.099* -0.156

7945(0.037) (0.040) (0.049) (0.059) (0.070)

0.2781 0.0846 0.1521 0.2781 0.1669


0.010

0.135* 0.133* 0.105 -0.105 -0.106

7945(0.049) (0.048) (0.054) (0.077) (0.077)

0.0914 0.0914 0.259 0.4441 0.4441


-0.040

0.012 0.113 0.148 -0.123 -0.223

7945(0.051) (0.056) (0.066) (0.078) (0.091)

0.8454 0.236 0.2004 0.3042 0.1626

Grant Use: more money spent on everything

Result 4: Substitution

Are incentives and budgets complement or substitutes?

Unless resources would allow you to access much more efficient technologies, some substitution is easily expected: See model If you have more resources, you need less effort to achieve the same outcome

We find: no interaction if incentive is small But very large substitution when the incentive is large

Both in outputs and inputs We find that if the grant is large, the incentive is no longer

effective!

Overall, no evidence of complementarities

Result 4: Substitution



(g/L)

Anemic at Endline


(g/L)

Anemic at Endline

1. Small Incentive -0.387 -0.012 1.054 -0.028

(1.101) (0.040) (0.987) (0.020)0.792 0.972 0.747 0.587

2. Large Incentive 2.567 -0.138* 0.918 -0.045(1.044) (0.039) (0.946) (0.022)0.285 0.064 0.767 0.373

3. Large Grant 4.205** -0.145** 2.872 -0.073**(1.123) (0.038) (0.989) (0.021)0.045 0.047 0.117 0.049




Mean in Small Grant, No

Incentive Group


Small Incentive

Large Incentive

Large Grant

(Small Incentive)X

(Large Grant)

(Large Incentive)X

(Large Grant)

A. Children

Anemic at

Baselin

e

1. Index: Vitamin Provision -0.050

0.141 0.158 0.242* -0.272 -0.290(0.084) (0.081) (0.072) (0.115) (0.106)0.241 0.241 0.0522 0.176 0.1226


-0.020 0.134 0.198* -0.117 -0.317**(0.050) (0.050) (0.066) (0.090) (0.090)0.7892 0.1032 0.0685 0.4888 0.0285


0.072 0.145* 0.224** -0.214* -0.302**(0.054) (0.052) (0.048) (0.078) (0.073)0.3034 0.0997 0.0045 0.0997 0.0098

B. Full Sample

5. Index: Vitamin Provision -0.120

0.192* 0.191* 0.250** -0.367** -0.261*(0.072) (0.075) (0.060) (0.099) (0.095)0.0766 0.0766 0.0067 0.0156 0.0766


0.055 0.113* 0.117 -0.099 -0.156(0.037) (0.040) (0.049) (0.059) (0.070)0.2781 0.0846 0.1521 0.2781 0.1669


0.133** 0.155** 0.193** -0.258** -0.213**(0.048) (0.050) (0.043) (0.066) (0.067)0.0324 0.0324 0.0025 0.0095 0.0324

Grant Use: more money spent on non-anemia

Higher incentives, higher effort, can dedicate resources to other activities

Comparing large grant and large incentive

No Incentive Small Incentive Large Incentive

Small Block Grant 36% 34.8% 22.2%

Large Block Grant 21.5% 16% 27.3%

Regression adjusted anemia rates at endline

They both give the same results… which is cheaper?

Computing the cost of averting one anemia case

Comparing large grant and large incentive

No Incentive Large Incentive

Small Block Grant 0 55

Large Block Grant 104 325

Incremental programatic cost of averting one anemia case (US$)

Incremental social cost of averting one anemia case (US$)No Incentive Large Incentive

Small Block Grant 0 230

Large Block Grant 333 646

Social costs include cost of public funds and costs to households but exclude incentive payments to school principals

Outline




Conclusions We motivated the paper saying that incentives is

an obvious solution to poor performance but afraid that they might not work in a public sector context with bureaucrats Well… no much evidence of that… In fact, they even went the extra-mile and involved

households in their problem (rewarding outputs)

And more resources (cash) also works but they are less cost-effective (same as Lavy 2002 JPE) Obviously, we can only say it for the combinations

specified here.

No detrimental effect on test scores (not reported here)

Going back to one of the first slides

Usual approach in the private sector: provide monetary incentives

Will that work in the public sector?

Concerns about multi-tasking Not reported in this paper, but no detrimental effects on

test scores

Intrinsic motivation crowding outTrue but only relevant for small incentives

Bureaucratic mindset: stability, aversion to changeNot so important as a priori thought. They even engaged

households.

Regarding policy impacts

If the policy question is:

Should I use schools to reduce anemia?

We do not say much about its value (no pure control group) but possibly give a lower bound

This was a very conscious decision to have more sample to answer:

How should I go about reducing anemia in schools?


• How should I go about reducing anemia in schools?• Our study is much more informative about this

question as• we compare incentives, resources, and their interaction• And we vary the size of the incentive

• But answering this question usually requires larger samples because you need to distinguish between treatments

• Also, we wanted shut down other channels (“salience”) through which providing incentives might have an effect: so everyone got information and a budget


• We also document that crowding out of intrinsic motivation is only an issue if the incentive is small enough

• Multitasking: no effects on test scores (paper that is being written)

ADDITIONAL SLIDES

Did contracts convey additional information?

Beliefs: Incentives and resources signal task difficulty (Benabou and Tirole 2003)

No evidence of this based on subjective expectations

Principal Subjective Expectations (Unadj)


Coefficient (standard error) on:

NSmall Incentive Large

Incentive Large Grant (Small Incentive)X (LargeGrant)

(Large Incentive)X

(Large Grant)

A. Immediately after learning of contract1. Mean of Ex-post Subjective Distribution 24.710

-8.042** -3.492 -2.929 9.992** 1.884170

(3.549) (3.409) (3.003) (4.244) (3.956)2. Variance of Ex-post Subjective Distribution 30.760

-1.161 7.858 3.008 5.979 -23.370170

(11.328) (12.508) (9.652) (16.437) (18.417)B. Endline Survey3. Mean of Ex-post Subjective Distribution 18.990

-3.676 5.084 2.200 -2.346 -6.680170

(3.934) (3.931) (3.572) (5.762) (5.339)4. Variance of Ex-post Subjective Distribution 20.570

-2.410 -10.208 -13.437 -9.081 17.757170

(8.002) (8.785) (11.239) (12.716) (13.376)NOTES. Data source: authors' survey. Table shows estimations for the effect of incentive contracts and large grants on the distribution of school principal subjective expectations over anemia reductions. Panel A shows principal subjective expectation at baseline (just after learning of their incentives and budget amounts) of how many students would be anemic at the end of the school year. Panel B shows results for a similar question at endline that asked principal expectations for the current anemia rate (before they learned the result). In each case, principals were asked the minimum, maximum and percent probability above the median. Mean and variance were derived assuming a triangular distribution. Each row shows coefficient estimates (and robust standard errors) from a seperate regression estimated using equation (1). *, **, and *** indicate significance at 10%, 5% and 1%.

can bureaucrats be paid like ceos? school managers incentives for anemia reduction in rural china...

Documents