different strokes for different folks: evidence on …...2 key contributions that we make: test...
TRANSCRIPT
Manoj
Mohanan
(Duke /
Sanford Public
Policy )
NIPFP
Conference
“New Thinking
in Health
Policy”
Oct 26-27
2017
DIFFERENT STROKES FOR
DIFFERENT FOLKS: EXPERIMENTAL EVIDENCE ON
PERFORMANCE INCENTIVE CONTRACTS
* with
Katherine Donato (Harvard),
Grant Miller (Stanford & NBER),
Yulya Truskinovsky (Harvard SPH),
Marcos Vera Hernández (UCL & IFS)
This research was made possible by funding and support from:
3ie &DFID-India (Grant #OW2:205 co-PIs: Mohanan & Miller),
World Bank HRITF (Grant number TF099435: PI Mohanan),
Government of Karnataka.
Usual disclaimers about funders apply.
FUNDING
Performance incentives to correct principal -agent problems ( H a l l & L i e b m a n 1 9 9 8 , H o l m s t r o m & M i l g r o m 1 9 9 1 , J e n s e n & M u r ph y 1 9 9 0 , K h a l i l & L a w a r r e e 1 9 9 5 ,
L a z e a r 2 0 0 0 , P r e n d e r g a s t 2 0 0 2 , R o l a n d 2 0 0 4 , R o s e n t h a l e t a l . 2 0 0 4 ) .
Reward inputs or outputs?
Output contracts assume that agents know the production function
and can find optimal combination of inputs in their own contexts
With low skilled worker, if inputs are observable & verifiable,
rewarding inputs is first best (Khali l and Lawarree 1995; Prendergast 2002).
Do agents perform better when they are able to innovate or do they do
better when they „follow orders‟?
P4P in health globally … ( F i n a n , O l k e n , a n d P a n d e , 2 0 1 5 )
Most P4P programs in health reward measures of service delivery (Ashraf, Bandiera, and Jack 2014, Basinga et al . 2011, Celhay et al . 2015, Dupas and Miguel 2016, Gertler
and Vermeersch 2013, Mil ler and Babiarz 2014)
Evidence of P4P‟s impact on health (outcomes) is mixed (Mil ler and Babiarz 2014; Sherry, Bauhoff , and Mohanan 2017)
V low quality / provider effort in LMIC … large „Know -do‟ gaps
MOTIVATION
Wide theoretical literature on input / output contracts, but
empirical evidence is relatively scarce
2 Key Contributions that we make:
Test effectiveness of input and output incentive contracts
First to empirically compare performance of agents when contracted on
inputs or outputs (esp in health)
Study differential effectiveness of input and output contracts among
providers with varying levels of human capital
Extends literature on optimal contracts and performance incentives
2 Concerns:
Selecting low risk patients
Multitasking (Holmstrom and Milgrom 1991)
OBJECTIVES & CONTRIBUTION
Agents produce health outcomes
Agents with high and low skills; beliefs about θ
Input contract,
Principals can reward specific inputs directly regardless of agents‟
beliefs on productivity shifters θ
But –because θ are local (principals do not know), this could lead
providers in input contracts to pick inefficient levels of effort .
Output contract,
Risk: outcome is not fully under agents‟ control. Premiums need to
compensate agents for this risk
But if providers have correct beliefs about productivity shifters and
hence can choose inputs optimally.
Testable implication: performance will depend on provider
skills in output contracts; but independent under input
contracts.
BRIEF CONCEPTUAL FRAMEWORK
Field experiment: randomize three types of performance
contracts to maternal care providers:
(a) Rewards based on outputs (PPH, Sepsis, Pre-eclampsia and
Neonatal Mortality)
(b) Rewards based on adherence to inputs (WHO / G.o.I. Guidelines)
(c) Control
All providers are given identical WHO / G.o.I . Guidelines as
information material
All providers sign an agreement - A & B sign performance
contracts, C sign an agreement to participate in study on MCH
All providers receive identical participation payments (~$45
at each visit) as compensation for time to answer surveys etc.
THE EXPERIMENT
Outputs ( in PPH, Sepsis, Pre-eclampsia)
Positive payments for reductions below a pre -intervention
level of outcomes.
INR 15000 for avoiding neonatal deaths
α is set based on allocating available balance across range of
improvements in 3 outputs (Exp. 5% min incidence)
Example: pre-intervention rate of PPH = 35% (x -bar). α = INR
850 (~$17); if x i = 25% then the provider gets $170.
THE CONTRACTS
Similar for inputs, except payments for improvements in
adherence to input guidelines above a min. level of
performance on 5 domains of care:
ANC,
Childbirth,
Post Natal Maternal Counseling,
Newborn care,
Post Natal Newborn Counseling.
Providers do not know what inputs are measureable (from
validation) and what the survey questions would look like.
CONTRACTS… CONTD.
135 providers
Eligibility: Pvt. rural practice in areas not served by large
public facilities.
120 from govt. survey data, 15 additional found during our
field visits
56% female, 59% have advanced OBGYN training, 47 yrs, 20
yrs experience, 17 years clinic. (Table 2; for balance see
Appendix Table A1)
PROVIDERS AND RANDOMIZATION
Timeline:
Analysis sample: 25 mothers who delivered at study facility
Additional community sample
DATA COLLECTION
Household surveys
Interview Mothers / attending family member within 2-3 weeks after
each delivery.
Not a cross section, to avoid recall problems (Das et al 2012)
Questions on health history, symptoms of outputs, and recall of
inputs provided (survey included validated and non-validated
questions).
Provider surveys
Expectations, demographics, capacity – before contract
Strategies – 2 months after contract
Follow up surveys – after contract including qualitative.
DATA - 2
Pre-analysis plan on AEA registry
https://www.socialscienceregistry.org/trials/179/history/728
OLS to estimate effect of treatment, controlling for household
and provider characteristics, district and enumerator FE
Clustered at level of provider
Multiple hypothesis testing:
4 outputs, 5 input indices, 2 treatment arms
Familywise error rate (Westfall & Young 1993)
ANALYSIS
(Table 3)
Both groups reduce PPH by about 21% ( rel to 36.5% in C)
PPH most amenable to improvement?
RESULTS - 1
Inputs (Table 4)
No significant improvements in any of the indices – especially
those activities pertaining to PPH
Improvement in postnatal maternal counseling (6) has
unadjusted p = 0.033, but after multiple outcomes correction
it is not significant.
PNCC (10) is important – we will revisit in a few slides.
RESULTS - 2
Potentially dif ficult to see improvements in inputs due to aggregation of many items into index ( Anderson 2008 )
E.g. Active Management of Third Stage of Labor (AMTSL)
Early Cord Clamping, Controlled Cord Traction, Abdominal Massage note
2 specific actions that are most closely related to PPH:
Parenteral Oxytocic Drugs and
Manual Removal of Placenta (potentially reflects complications)
Table 5:
Providers in both arms ~ 7pp more likely to stock drugs (and use them)
Manual removal of placenta is conducted less often (7/27 = 26%)
RESULTS - 3
Ex-post, we see that average payments to outputs was much higher than to input contracts ($1033 v/s $252)
Potentially reflects the risk premium for output contracts
In our setting, input contract was more efficient
Unable to make generalizable inference about efficiency.
COST OF CONTRACTS
Actual Payments to Input Actual Payments to Output
Counterfactual Payments
Our conceptual framework suggests that provider skills would
determine effectiveness of output v/s input contracts
Innovate to meet target in former, v/s follow order in the latter
We look at those with advanced OBGYN training (MBBS+) v/s others
RESULTS ON TYPE OF AGENTS
Do MBBS+ providers Innovate more?
RESULTS ON TYPE OF AGENTS
Table 6:
OUTPUT contract: MBBS+ providers reduced PPH 9pp more than other
providers
INPUT contract: No better or worse
Exploring whether the MBBS+ folks innovated more (Table 7):
OUTPUT contract: increased Pr(new strategy) for MBBS+ providers
Lincom coeff = 0.36 (se = 0.14)
INPUT contract: No increase; lincom coeff = 0.14 (se =0.17)
RESULTS ON TYPE OF AGENTS
Also relates to reduction in PNCC (col 10 in T4) in output arm
75% providers thought PPH was most important to improve
among their patients
On 9% thought NM
Context of NM in
OBGYN care in
India
Col 10 – refers to
counseling about
postnatal care
No change in input
“followed orders”
v/s
Reduction in output
WHY PPH? MULTITASKING?
Output and Input contracts can achieve comparable gains –
and also reduce PPH significantly (major health issue)
Heterogeneity based on skills:
With high skilled workers, output contracts might induce better
performance
In contrast, output contracts with low skilled workers might not be as
effective
Current focus of ongoing incentive programs globally to
reward inputs might in fact be appropriate despite lack of
previous empirical evidence on the rationale for this choice.
CONCLUSIONS & POLICY IMPLICATIONS
Acknowledgements
T h is r esea r c h wa s m a d e p o ss ib le by f u n d in g a n d su p p o r t f ro m 3 ie a n d
DF I D - I n d ia ( G r a n t n u m b er OW 2 : 2 0 5 c o - P I s : M o h a n a n a n d M i l l e r ) , Wo r ld
B a n k H R I T F ( G r a n t n u m b er T F 0 99435: P I M o h a n a n ) a n d G over n m en t o f
Ka r n a t a ka .
We a r e g r a te f u l , fo r c o m m en t s a n d su g g es t io n s , to A lessa n d r a Vo en a ,
A lessa n d ro Ta ro z z i , A m a r H a m o u d i , Du n c a n T h o m a s , E r i c a F ie ld , J e r r y L a
Fo r g ia , J i sh n u Da s , I m r a n R a su l , M er ed i t h Ro sen t h a l , M ic h a e l C a l len ,
N ava A sh r a f , N eer a j S o o d , O r ia n a B a n d ie r a , Pa u l G er t le r , Ro h in i Pa n d e ,
Ro b G a r l i c k , V i c to r ia B a r a n ov, X ia o Yu Wa n g , a n d to a u d ien c es a t
A E A / A S S A 2 017 , A S H E c o n 2 016, B a r c e lon a G S E 2 016 , B R E A D/ C E P R
2 016, Du ke , E r a sm u s , H a r va r d , iH E A C o n g r ess M i la n , T i l b u r g Un i ve r s i t y,
a n d Un i ve r s i t y o f S o u t h er n C a l i fo r n ia .
M a nveen Ko h l i p rov id ed exc e l len t p ro jec t m a n a g em en t . We a r e t h a n k f u l
to Ku l t a r S in g h , S wa p n i l S h ek h a r, a n d A n i l L o b o f ro m S a m b o d h i , a s we l l
a s t h e f i e ld tea m fo r p ro jec t im p lem en t a t io n a n d d a t a c o l l ec t ion .
We g r a te fu l l y a c k n ow ledg e t h e su p p o r t we r ec e i ved f ro m Wo r ld B a n k
( Pa o lo B e l l i , Pa t r i c k M u l len , a n d V ik r a m R a ja n ) a n d t h e G over n m en t o f
Ka r n a t a ka ( Va n d i t a S h a r m a , S e l va Ku m a r, S u r esh M o h a m m ed ,
R a g h aven d r a J a n n u , At u l T iwa r i , D r. N a g a r a j , D r. S r id h a r, D r. P r a ka sh
Ku m a r, D r. A m r u tesh wa r i , a n d seve r a l o t h e r s ) . We a r e esp ec ia l l y g r a te fu l
to t h e m a ny d o c to r s a n d c l in i c a l ex p er t s w h o p rov id ed va lu a b le g u id a n c e
a n d feed b a c k , in c lu d ing M a t t h ew s M a t h a i , D in esh A g a r wa l , Aya b a
Wo r jo la h , V in o d Pa u l , S h a r a d I yen g a r, K i r t i I yen g a r, A m a r j i t S in g h ,
S u n eet a M i t t a l , L a l i t B ave ja , a n d S u n esh Ku m a r
Growing literature on incentives, performance, and
personality traits
Focus on Conscientiousness and Neuroticism – correlated with
labor market outcomes (Borghans et a l . 2008, Heckman, St ixrud and Urzua
2006, Heckman and Rubinste in 2001)
Conscientiousness : dependability, organization skills , perseverance,
and achievement oriented thinking
Neuroticism: the converse of emotional stability – is associated with
anxiety, worry, anger, and insecurity
PERSONALITY TRAITS &
PERFORMANCE CONTRACTS
CONSCIENTIOUSNESS & NEUROTICISM
10/31/2017 AEA Conference - Chicago
Conscientious providers do relatively better absent incentives.
Beneficial effect of incentive is weaker among high Cons. providers
At 25th percentile of Conscientiousness (4.3/5), the incentive contracts reduce PPH risk by 13.3 %pts.
At 75th percentile of C. (5/5), no statistically significant effect
No evidence of association b/w neuroticism and performance in absence of incentives (could be due to selection into MD?)
Performance improvement from incentives is amplified among low Neuroticism (high emotional stabil ity):
13%point reduction at 25 th percentile (1.25/5),
No significant results at 75 th percentile.
“Choking under pressure” hypothesis, -- performance deteriorates due to over-arousal and distraction that accompany high stakes (Ariely et al. 2009,
Baumeister 1984, Yu 2015)
PERSONALITY TRAITS (AER P&P)
PERSONALITY TRAITS (AER P&P)
Output and Input contracts can achieve comparable gains – and also reduce PPH significantly (major health issue)
Heterogeneity based on skil ls:
With high skilled workers, output contracts might induce better performance
Output contracts with low skilled workers might not be as effective
Personality traits
Evidence of significant heterogeneity by personality traits – both high conscientiousness and high neuroticism providers don ‟t show improvements with incentives.
Among high conscientiousness providers, the dampened effect is suggestive of crowding out.
Among high neuroticism (the converse of emotional stability) these results are consistent with “choking”
Current focus of ongoing incentive programs globally to reward inputs might in fact be appropriate despite lack of previous empirical evidence on the rationale for this choice.
CONCLUSIONS & POLICY IMPLICATIONS