impact of caregiver incentives on child health: evidence from an experiment with anganwadi workers...
TRANSCRIPT
Impact of caregiver incentives on child health:Evidence from an experiment with
Anganwadi workers in India
Will MastersFriedman School of Nutrition & Department of Economics, Tufts University
Prakarsh SinghDepartment of Economics, Amherst College
POSHAN (IFPRI)10th November, 2016.
Can we incentivize salaried workers to target their services effectively?
• Performance pay is difficult to use and evaluate– Measurement of performance is costly, affected by noise, time lags and confounders
– Rewards may crowd out other motivations, and reduce effort on other tasks
– Rewards may drive selection into participation, targeting and level of effort
• Child nutrition is difficult to improve– Inputs (dietary intake and disease exposure) are usually not observed
– Outcomes (body size, disease state) are difficult to measure and compare
– Links between inputs and outcomes are unknown
• India’s ICDS program offers a large-scale opportunity to intervene– Over 1 million centers each serving over 30 preschool children, with salaried Anganwadi
worker providing mid-day meal, advice to mothers, some teaching
– Government aims to improve performance for both nutrition and education
– Objectives include reduced weight-for-age malnutrition, which is still widespread
– Low weight-for-age, defined as WAZ < -3 or -2 standard deviations below median of a healthy population, can be due to either inadequate diet or disease burden
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Can we incentivize salaried workers to target their services effectively?
Summary of results
• Trial compares a performance pay bonus (<5% of salary) to a fixed bonus of similar size and a pure control group – Population is about 4,000 children in 160 government-run ICDS day-care centers in urban slums
of Chandigarh, India
– Primary outcome is the ICDS objective of lower weight-for-age malnutrition; we also report changes in height-for-age
– Mechanism checks measure efforts of the worker and the child’s mother, with dose-response checks around thresholds
• We find that the performance bonus reduces prevalence of weight-for-age malnutrition by about 5 percentage points over 3 months– Effect is sustained with renewal of incentives, and fades when discontinued
– Mechanism is attendance and communication with mothers of at-risk children, with improved diets at home especially for children near thresholds
• Impacts imply that small bonuses can focus caregiver attention and improve targeting of efforts such as communication with mothers
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Context
Block 1 (control)
Block 2 (bonus treatments)
Block 3 (later treatments)
In urban slums of Chandigarh-- Planned city in far north India-- Capital of both Punjab and Haryana-- Income level similar to Delhi-- Population size < 2 million
Trial designed in collaboration with ICDS management-- Geographically separated blocks-- Retain 84 centers in poorer block 1 as
controls for seasonality and trends-- Split 76 centers in block 2 between
performance pay and fixed bonus-- Keep 85 centers in block 3 for later
tournament treatments (not reported here)
-- Data collected in 5 rounds at 3 monthintervals, July 2014 - July 2015, with surveys of workers, childrenand their mothers
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Treatments
• Fixed bonus is Rs. 200 per worker over three months– In block 2, workers draw randomly into performance vs. fixed bonus treatments
• Performance bonus is Rs. 200 per child for status improvements – Every worker given a goal card, with baseline weight and gains needed for each child
• Bonuses calibrated based on previous ICDS experiments– Expected gains over 3 months on the order of 2 of the 30+ children enrolled
– Expected bonus after 3 months ≈ Rs. 400, relative to salary of Rs. 4000 per month
• Treatment is designed to align with government’s ICDS objectives – Status improvements can be from severe (WAZ<-3) to moderate (WAZ<-2) or to none
– Status improvements exclude any cases of overweight relative to height (WHZ>+1)
– Bonuses are net of any declines in status into moderate or severe malnutrition
– Bonuses have lower bound of zero
• Treatment is designed for potential cost-effectiveness– Every mother given a recipe book with nutrition advice, to complement worker efforts
• Both treatments are compared to block 1, to control for common shocks
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Timeline of the experiment
Round Date Block 1 Block 2 Block 3
Baseline-I Jul-14Control*
(83)Control (76)
Control
(85)
Baseline-II Oct-14Control
(84)
Performance
Pay (38)
Fixed Bonus
(38)
Control
(85)
Endline-I Jan-15Control
(84)
Performance
Pay (38)
Control
(85)
Endline-II Apr-15Control
(84)
Endline-III Jul-15Control
(84)
Notes: * denotes that one center was not surveyed from Block 1 in Baseline-I as
it was closed. Numbers in parentheses show the number of centers in each arm.
Treatment dates shown are for start of treatment, with bonus payments made
at the end of Endline-I and Endline-II respectively.
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Average treatment effects
Short term effects (R2 to R3) over 3 mo.
(7) (8) (9)
Weight Wfa z Wfa mal
Performance 0.219*** 0.101*** -0.0561**
Pay (0.0772) (0.0370) (0.0269)
Fixed 0.123 0.0557 -0.0333
Bonus (0.0933) (0.0442) (0.0278)
N 3528 3522 3524
Medium-term effects (R3-R4) over 3 mo.
(7) (8) (9)
Weight Wfa z Wfa mal
Performance 0.231*** 0.0976*** -0.0522**
Pay (0.0687) (0.0327) (0.0219)
Fixed 0.196** 0.0878** -0.0341
Bonus (0.0776) (0.0380) (0.0241)
N 2303 2301 2302
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. Weight is in kilograms. Wfa z is the weight-for-age z score given the child's sex and age, and Wfa mal is an indicator for malnutrition status. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Note: Results are robust to checks using Lee (2009) treatment effect bounds, or Moulton standard errors for sample size
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Pre-trends and fade-out
Pre-trends (R1 to R2) over 3 mo.
(7) (8) (9)
Weight Wfa z Wfa mal
Performance -0.0991 -0.00620 -0.0305
Pay (0.119) (0.0411) (0.0223)
Fixed 0.0971 0.0694 -0.0305
Bonus (0.0884) (0.0423) (0.0285)
N 3744 3730 3739
Fade-out after treatments (R4 to R5) over 3 mo.
(7) (8) (9)
Weight Wfa z Wfa mal
Performance 0.0898 0.0355 -0.0338
Pay (0.0904) (0.0408) (0.0235)
Fixed 0.00967 0.00266 0.00262
Bonus (0.0752) (0.0357) (0.0267)
N 2230 2223 2224
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. Weight is in kilograms. Wfa z is the weight-for-age z score given the child's sex and age, and Wfa mal is an indicator for malnutrition status. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Threshold effects
“Near” and “Far” are defined around the median distance to each threshold.
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. Weight is in kilograms. Wfa z is the weight-for-age z score given the child's sex and age, and Wfa mal is an indicator for malnutrition status. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Incentive effect All gain
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Worker efforts
Short term effects (R2 to R3) over 3 mo.
Home
visits by
worker
Center
visits by
mother
Frequency of
worker talking
about the child
Performance -1.256 -1.141 4.410***
Pay (0.915) (1.438) (0.970)
Fixed -2.019* -1.223 5.012***
Bonus (1.092) (0.855) (1.029)
N 3275 2831 3062
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Type of mother-worker interactions in the past month (as reported by mother)
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Short term effects (R2 to R3) over 3 mo.
Dietary
Intake Hygiene
Growth
Chart
Harmful
Effects
Perf. 0.226*** 0.0949 0.0712 -0.0206
Pay (0.0767) (0.0832) (0.0780) (0.0866)
Fixed 0.245*** 0.0757* 0.0138 -0.0922
Bonus (0.0633) (0.0907) (0.0792) (0.0725)
N 3223 3223 3223 3223
Worker efforts
Topic of mother-worker interactions in the past month (as reported by mother)
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Short term effects (R2 to R3) over 3 mo.
Milk
Green
veg. Dessert Porridge
Perf. 0.0616*** -0.130*** 0.228*** 0.105*
Pay (0.0182) (0.0341) (0.0608) (0.0617)
Fixed 0.0666*** -0.148*** 0.213*** 0.293***
Bonus (0.0228) (0.0312) (0.0582) (0.0573)
N 3223 3223 3223 3223
Mothers’ response
Child’s diet at home: items consumed at least twice in past week (as reported by mother)
All results control for observables on children, mothers and workers, with heteroscedasticity-consistent standard errors clustered on centers. *Significant at 10%, **Significant at 5%, ***Significant at 1%.
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Other outcomes: Child height
Change in height (cm) over 3 mo.
R1 to R2 R2 to R3 R3 to R4 R4 to R5
Performance 0.381 1.077** -0.263 -0.0946
Pay (0.480) (0.502) (0.375) (0.382)
Fixed 0.571 0.988* -0.206 -0.546
Bonus (0.494) (0.511) (0.332) (0.353)
N 3721 3497 2286 2220
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms
Conclusions
• Small bonuses to staff did improve outcomes of children in their care– Total gains and threshold effects were larger when bonuses were tied to outcomes
– Some improvement even with fixed bonuses
– Complements include goal cards to guide efforts, recipe books to help mothers respond
• Magnitude of improvement was significant– Reduced weight-for-age malnutrition prevalence by about 5 pct. points over 3 months
– Average speed of additional weight gain was about 70 grams per month
– Cost-benefit ratios are roughly similar to iron, deworming, etc.
– Weight gain and also promoted linear growth
– Implications for scale-up
• Mechanisms provide insight into agents’ knowledge of relative effectiveness
– Caregivers altered frequency, content of communication with mothers
– Mothers altered composition of children’s diets
Caregiver incentives and child healthmotivation | trial design | outcomes | mechanisms