Do Informed Consumers Reduce the Price and Prevalence of Counterfeit - - PowerPoint PPT Presentation

Do Informed Consumers Reduce the Price and Prevalence of Counterfeit Drugs? Evidence from the Antimalarial Market

Anne Fitzpatrick*

Department of Economics & Ford School of Public Policy fitza@umich.edu

March 22, 2015

*Funding was received from the following sources: National Science Foundation Dissertation Improvement Grant (#1260911), the Department of Afro-American and African Studies; the African Studies Center; the Rackham Graduate School International Research and Graduate Research Awards, and the Center for Public Policy in Diverse Studies Research Grant; Center for International Business Education Research Grant; Department of Economics MITRE grant; and the Center for the Education of Women Research Grant. Anne Fitzpatrick (University of Michigan) March 22, 2015 1 / 22

Private sector is primary source of treatment for malaria

Patients seek treatment and advice from “experts”: make money from recommendations

Anne Fitzpatrick (University of Michigan) March 22, 2015 2 / 22

Private sector is primary source of treatment for malaria

Patients seek treatment and advice from “experts”: make money from recommendations Asymmetric information

Particularly problematic in developing countries: unregulated markets, low human capital

⇒ High prices, low quality [lemons model- Akerlof, 1970] Meta-analysis: one-third of antimalarial drugs are low quality [Nayyar et al., 2012]

Anne Fitzpatrick (University of Michigan) March 22, 2015 2 / 22

Private sector is primary source of treatment for malaria

Patients seek treatment and advice from “experts”: make money from recommendations Asymmetric information

Particularly problematic in developing countries: unregulated markets, low human capital

⇒ High prices, low quality [lemons model- Akerlof, 1970] Meta-analysis: one-third of antimalarial drugs are low quality [Nayyar et al., 2012]

If asymmetric information causes problems in markets, will increased customer information fix the problem?

Anne Fitzpatrick (University of Michigan) March 22, 2015 2 / 22

This Paper: An Audit Study

Research Question

Do providers adjust price or quality in response to increased customer information regarding diagnosis and treatment?

Anne Fitzpatrick (University of Michigan) March 22, 2015 3 / 22

This Paper: An Audit Study

Research Question

Do providers adjust price or quality in response to increased customer information regarding diagnosis and treatment? Outline

Send pairs of mystery shoppers to purchase drugs with randomly assigned scripts

Information customers could actually present at the time of purchase

Link with surveys of vendors and real customers to examine mechanisms

Anne Fitzpatrick (University of Michigan) March 22, 2015 3 / 22

This Paper: An Audit Study

Research Question

Do providers adjust price or quality in response to increased customer information regarding diagnosis and treatment? Outline

Send pairs of mystery shoppers to purchase drugs with randomly assigned scripts

Information customers could actually present at the time of purchase

Link with surveys of vendors and real customers to examine mechanisms

Account for quality: test drugs using a handheld spectrometer

Contrast with service quality and correct dosage Theoretical framework

Anne Fitzpatrick (University of Michigan) March 22, 2015 3 / 22

Ugandan antimalarial market is ideal setting

Anne Fitzpatrick (University of Michigan) March 22, 2015 4 / 22

Ugandan antimalarial market is ideal setting

1 Customer characteristics

Low levels of customer information for a common disease Policy-relevant

Anne Fitzpatrick (University of Michigan) March 22, 2015 4 / 22

Ugandan antimalarial market is ideal setting

1 Customer characteristics

Low levels of customer information for a common disease Policy-relevant

2 Treatment recommendations are not ambiguous

If an adult has malaria, test and prescribe AL Clear interpretation of how providers respond Objective measure of quality

Anne Fitzpatrick (University of Michigan) March 22, 2015 4 / 22

Ugandan antimalarial market is ideal setting

1 Customer characteristics

Low levels of customer information for a common disease Policy-relevant

2 Treatment recommendations are not ambiguous

If an adult has malaria, test and prescribe AL Clear interpretation of how providers respond Objective measure of quality

3 Market characteristics

Little enforced institutional regulation Fee-for-service providers No prescriptions/insurance/etc.

Anne Fitzpatrick (University of Michigan) March 22, 2015 4 / 22

Shopping Protocol

Comparison to Real Customers Anne Fitzpatrick (University of Michigan) March 22, 2015 5 / 22

Drug Choice Protocol

Standardized choice behavior across & within shoppers

Avoids endogeneity that happens with real customers Recommendations affect knowledge level

1 Buy AL, taking cheapest available brand 2 Take quinine if AL not available 3 Take cheapest other option if quinine not available 4 Buy nothing if no antimalarials available

92 percent of visits resulted in an antimalarial drug purchase No difference in type of drug purchased across scripts

Anne Fitzpatrick (University of Michigan) March 22, 2015 6 / 22

Research Design: Randomly Assigned Scripts

Know AL Ask Rec TOTAL T-3 T-1 Know Malaria N = 499 0.249 0.242 T-2 Control Ask for Diagnosis N =517 0.252 0.255 TOTAL N= 510 N=506 N=1016 All Treatments together (“Any Information”) vs. Control

Comparison to Real Customers Anne Fitzpatrick (University of Michigan) March 22, 2015 7 / 22

Provider Response To Any Information

Yst = α0 + α1AnyInformationst + γv + δ′X + ǫst with Yst = price, quality, quantity of transaction t in shop s γv = village fixed effect δ′X = visit order, patient, and shopper FE ǫst = standard errors clustered at the outlet level Normalized indices to correct for multiple outcomes [Kling et al., 2007] I use the randomly assigned script in all specifications

Anne Fitzpatrick (University of Michigan) March 22, 2015 8 / 22

Counterfeit vs. Substandard

Counterfeit... but NOT substandard

17 % are counterfeit; 4 % are substandard

Anne Fitzpatrick (University of Michigan) March 22, 2015 9 / 22

Drug Prices Fall

Price

• ffered, All

Price

• ffered,

Analysis Ln Price Offer Price Index (1) (2) (3) (4) Any Information −0.128 −0.184∗∗ −0.053∗ −0.081∗∗∗ (0.095) (0.093) (0.030) (0.031) (1) (2) (3) (4) Know Only Malaria −0.166 −0.231∗∗ −0.051 −0.081∗∗ (0.116) (0.116) (0.034) (0.039) Know Only Drug −0.076 −0.129 −0.042 −0.070∗∗ (0.111) (0.107) (0.039) (0.034) KnowMalaria & Drug −0.142 −0.192∗ −0.066∗ −0.090∗∗ (0.106) (0.107) (0.035( (0.035) Pvalue Malaria= 0 0.311 0.109 0.14 0.032 Pvalue Drug= 0 0.409 0.191 0.173 0.029 R-squared 0.557 0.572 0.528 0.574 Observations 933 879 879 879 #clusters 471 459 459 459 Mean Dep Control 3.51 3.59 1.18 0.01

Selection into prices? Prices Anne Fitzpatrick (University of Michigan) March 22, 2015 10 / 22

Observable Drug Quality Improves

Correct Dosage Diverted Drug (“stolen”) (1) (2) Any Information 0.041∗ −0.039∗∗ (0.022) (0.018) (1) (2) Know Only Malaria 0.043∗ −0.040∗∗ (0.025) (0.020) Know Only Drug 0.043 −0.035 (0.026) (0.023) KnowMalaria & Drug 0.037 −0.042∗∗ (0.025) (0.020) Pvalue Malaria= 0 0.184 0.069 Pvalue Drug= 0 0.213 0.115 Observations 879 879 R-squared 0.268 0.471 Number of clusters 459 459 Mean Dep Control 0.909 0.100

Anne Fitzpatrick (University of Michigan) March 22, 2015 11 / 22

Actual Drug Quality Falls

Counterfeit Substandard % Pills Substandard Quality Index (1) (2) (3) (4) Any Information 0.003 0.034∗∗∗ 0.025∗∗∗ −0.596∗∗∗ (0.030) (0.012) (0.008) (0.188) (1) (2) (3) (4) Know Only Malaria −0.009 0.030∗ 0.014∗ −0.352∗ (0.036) (0.016) (0.008) (0.199) Know Only Drug 0.026 0.017 0.019∗ −0.460∗∗ (0.038) (0.013) (0.010) (0.231) KnowMalaria & Drug (0.005) 0.054∗∗∗ 0.041∗∗∗ −0.959∗∗∗ (0.035) (0.019) (0.012) (0.292) Pvalue Malaria= 0 0.966 0.009 0.002 0.003 Pvalue Drug= 0 0.674 0.021 0.003 0.005 Observations 879 879 879 879 R-squared 0.217 0.208 0.248 0.239 Number of clusters 459 459 459 459 Mean Dep Control 0.174 0.013 0.000 −0.017

Know quality? Testing Anne Fitzpatrick (University of Michigan) March 22, 2015 12 / 22

Service Quality Falls: Checklist

Express Doubts About Malaria Advised Malaria Test Asked Health Question Service Quality Index (1) (2) (3) (4) Any Information −0.044 −0.068∗ −0.047∗ −0.128∗∗∗ (0.033) (0.037) (0.026) (0.028) (1) (2) (3) (4) Know Only Malaria −0.057 −0.02 −0.018 −0.070∗∗ (0.039) (0.047) (0.031) (0.035) Know Only Drug 0.009 −0.022 −0.026 −0.134∗∗∗ (0.043) (0.046) (0.033) (0.035) Know Malaria & Drug −0.082∗∗ −0.158∗∗∗ −0.097∗∗∗ −0.178∗∗∗ (0.039) (0.044) (0.033) (0.037) R-Squared 0.333 0.321 0.638 0.702 Pvalue Malaria=0 0.103 0.000 0.010 0.000 Pvalue Drug = 0 0.035 0.001 0.011 0.000 Observations 867 867 867 867 Number of clusters 459 459 459 459 Mean Dep Control 0.261 0.409 0.752 0.069

Anne Fitzpatrick (University of Michigan) March 22, 2015 13 / 22

Summary of Findings

When customers know either diagnosis or treatment...

Providers decrease prices by 5 % (noisy) Decrease substandard rate by 3.4 pp (100 %) Increase observable measures of quality (correct dosage, diverted) Substantially reduce service quality No difference by type of information More information causes a stronger response

Anne Fitzpatrick (University of Michigan) March 22, 2015 14 / 22

Summary of Findings

When customers know either diagnosis or treatment...

Providers decrease prices by 5 % (noisy) Decrease substandard rate by 3.4 pp (100 %) Increase observable measures of quality (correct dosage, diverted) Substantially reduce service quality No difference by type of information More information causes a stronger response

Mechanism

Trade-off between current and future profit losses if agency detected Price drives drive the quality results, through “penalties” Service quality is actually priced into good

Anne Fitzpatrick (University of Michigan) March 22, 2015 14 / 22

Model, Formally

max

pi,si,q∈{G,B}

pi − e(si) − cq + αq

i Πi(pi)

subject to θsi − pi ≥ 0 (1) Π: future profits from that customer θ: marginal valuation of service quality Constraint binds ⇒ service and price move together directly

Service quality is observable, in utility function Affects purchase decision

Modifications: outside options, θi, etc.

Anne Fitzpatrick (University of Michigan) March 22, 2015 15 / 22

Model Solution

First-order condition ⇒ service, price, and likelihood of returning are all positively related θ(1 + αiΠ(θsi)) = e′(si) (2)

Anne Fitzpatrick (University of Michigan) March 22, 2015 16 / 22

Model Solution

First-order condition ⇒ service, price, and likelihood of returning are all positively related θ(1 + αiΠ(θsi)) = e′(si) (2) Providers choose which drug quality maximizes profits Likelihood of returning is positively related to optimal choice of drug quality

Anne Fitzpatrick (University of Michigan) March 22, 2015 16 / 22

Model Solution

First-order condition ⇒ service, price, and likelihood of returning are all positively related θ(1 + αiΠ(θsi)) = e′(si) (2) Providers choose which drug quality maximizes profits Likelihood of returning is positively related to optimal choice of drug quality Intuition: information signals another characteristic of demand A is the revenue from the current sale A − cG + αiΠi ≥ A − cB αiΠi ≥ cG − cB (3)

Anne Fitzpatrick (University of Michigan) March 22, 2015 16 / 22

Mechanisms: Correlations of Real Customers

I compare answers between two groups of customers to establish plausible correlations

“Did you ask the vendor for a diagnosis?” “Did you ask the vendor for a specific product?”

Anne Fitzpatrick (University of Michigan) March 22, 2015 17 / 22

Mechanisms: Correlations of Real Customers

I compare answers between two groups of customers to establish plausible correlations

“Did you ask the vendor for a diagnosis?” “Did you ask the vendor for a specific product?”

Yst = λ0 + λ1AnyInformationst + γv + ψ′X + µst Any Information - Customer reported knowing either the diagnosis (malaria) or a specific product No Information - Customer reported asking for both a diagnosis and a specific product Control for adult patient, income, and education Village fixed effects

Anne Fitzpatrick (University of Michigan) March 22, 2015 17 / 22

Correlations of Prices and Outcome Variables Among Real Customers

Variable Any Information No Information Difference (N=195) (N=178) (1) (2) (3) Income (USD) 152.03 152.01 −0.03 Education 8.927 9.881 −0.95∗∗ Repeat Customer 0.698 0.853 −0.166∗∗∗ Bought a Full Dose 0.982 0.946 0.04∗ Bargained over price 0.610 0.727 −0.12∗∗∗ Bought AL 0.539 0.670 −0.13 Patient took Malaria Test 0.191 0.420 −0.23∗∗∗ Bought Additional Product 0.475 0.632 −0.16∗ Product Price (USD) 2.189 2.835 −0.65∗∗∗ Total Bill (USD) 2.336 3.697 −1.36∗∗∗

Significance calculated conditional on a village FE

Anne Fitzpatrick (University of Michigan) March 22, 2015 18 / 22

Policy Implications

Results do not imply withholding information to customers

Customers with any information receive a 5.6 percent discount

Information may not be effective at improving drug quality by itself

Providers strategically allocate low quality drugs to customers who pay less

Providers strategically use information to segment market

Uninformed are more reliant upon advice: higher valuation on service quality Good service, compensation for price May explain why customers learned the in haven’t learned information: lower benefits

Anne Fitzpatrick (University of Michigan) March 22, 2015 19 / 22

Conclusion

In this paper, I conduct an audit study to test provider response to customer information I find that presenting information to the vendor causes...

A decrease in price and profits Fewer options of additional products for symptoms Service quality decreases Drug quality to fall

Vendors segment the market to profit-maximize

Based most likely on information but also likelihood to visit again People most willing to trade off high prices for better service

Anne Fitzpatrick (University of Michigan) March 22, 2015 20 / 22

Thank you!

Send additional comments to fitza@umich.edu

Anne Fitzpatrick (University of Michigan) March 22, 2015 21 / 22

Bonus Slides

Anne Fitzpatrick (University of Michigan) March 22, 2015 22 / 22

Analogy: Burgers = meat + spices

All fast food chains basically have the same recipe, and slightly vary in spices Consider a chain (BK) with high variability so some BK burgers look like Wendy’s burgers Machine is very sensitive: Purchase from Franchise A = BK, even when it truly is: “counterfeit” Because recipes are similar, though, can see if purchase A matches any other fast food chain If not, “substandard” Advantage of testing many brands of same drug Back to

testing . Anne Fitzpatrick (University of Michigan) March 22, 2015 23 / 22

Credence goods with reputation, Schneider, 2012

2 firms (honest and dishonest), 2 periods, customer with a search cost Need to choose price and quality (unobservable)

Search costs = different costs of full information Firms use these different search costs to price discriminate

Backward induction: Period 2

Set maximum price so that customer is indifferent between accepting and seeking second firm Customers with higher costs of search pay higher prices Dishonest firm gives low quality drug to customer; Honest firm gives high quality drug

Period 1

Firms still price discriminate Both firms give high quality drugs

Back to Conceptual Anne Fitzpatrick (University of Michigan) March 22, 2015 24 / 22

Census Map

Back to

Outline . Anne Fitzpatrick (University of Michigan) March 22, 2015 25 / 22

Exit Interview Responses Mapped to Experiment

Know Drug Ask Rec TOTAL Know Malaria 85 44 129 0.228 0.118 Ask for Diagnosis 49 195 244 0.131 0.523 TOTAL 134 239 373 Back to

Treatment . Anne Fitzpatrick (University of Michigan) March 22, 2015 26 / 22

Prices Vary

Back to

Bargaining . Anne Fitzpatrick (University of Michigan) March 22, 2015 27 / 22

Shoppers Returned the Balance

In this context, returning balance may not be incentive compatible

Could buy less than a full dosage and say it’s a full dosage Could buy SP instead Beliefs about adjusting prices going forward in more expensive/cheaper areas

Concern that it doesn’t mimic real life

In this context, would be a windfall of income Like life without a budget constraint

Potential effect on prices?

Less concern of price reporting error because 2 visits to same shop If bias, against finding an effect of scripts

Back to

Bargaining . Anne Fitzpatrick (University of Michigan) March 22, 2015 28 / 22

Vendors Say Why Prices Vary

Reason stated Percent Customers are poor 0.382 Individual bargaining power 0.372 Cost price fluctuations 0.146 To remove inventory 0.080 To improve competitiveness 0.070 Regular customers 0.065 Patients are sick/altruism 0.065 Other Reasons 0.065 Back to

Bargaining . Anne Fitzpatrick (University of Michigan) March 22, 2015 29 / 22

Types of Drugs Purchased

All Active Ingredients (1) (2) (3) (4) (5) AL 806 0.86 8275 3.19 1.28 Quinine 34 0.04 6429 2.48 1.19 SP 79 0.09 2915 1.12 0.59 Other High Quality 7 0.08 12857 4.96 4.16 Other 7 0.08 4071 1.52 0.71 TOTAL 933 1.00 7757 2.99 1.25 Back to

Bargaining Anne Fitzpatrick (University of Michigan) March 22, 2015 30 / 22

Incorrect Scripts

(1) (2) (3) (4) (5) Know Only Malaria −0.002 −0.002 −0.006 −0.002 −0.005 (0.015) (0.015) (0.017) (0.017) (0.017) Know Only AL −0.013 −0.012 −0.009 −0.005 −0.006 (0.016) (0.016) (0.018) (0.017) (0.018) Know Malaria & AL (0.009) (0.008) (0.002) (0.001) (0.001) (0.017) (0.016) (0.017) (0.016) (0.017) Patient 0.004 0.003 −0.005 −0.001 −0.002 (0.012) (0.012) (0.011) (0.011) (0.011) Visit Order 2 −0.016 −0.016 −0.015 −0.015 −0.017 (0.011) (0.011) (0.012) (0.012) (0.012) Visit Order 3 0.026∗∗∗ 0.145∗∗∗ 0.086∗∗ (0.008) (0.035) (0.044) Visit Order 4 0.030∗∗∗ 0.049 0.035 (0.011) (0.048) (0.049) First 2 Visits Y Y All Visits Y Y Y Village Fixed Effects Y Y Y Interviewer Fixed Effects Y Observations 978 1016 978 1016 1016 R-squared 0.002 0.003 0.254 0.228 0.241

Back to

Empirical Strategy Anne Fitzpatrick (University of Michigan) March 22, 2015 31 / 22

Distribution of Outlets & Village Balance

Num Visits Per Outlet N % Sample (1) (2) 1 61 0.133 2 378 0.823 3 18 0.039 4 2 0.004 Village-Level Variables Average SD F-test (1) (2) (3) HHI 0.36 0.26 0.22 Village Visit Order 8.86 11.96 0.34 Number of Outlets 8.67 9.88 0.33 Urban 0.79 0.41 0.60

Anne Fitzpatrick (University of Michigan) March 22, 2015 32 / 22

Distribution of Outlets

Back to

Empirical Strategy Anne Fitzpatrick (University of Michigan) March 22, 2015 33 / 22

Degree of Imbalance

Know Malaria & AL Know Only Malaria Know Only AL Control Village & shopper FE Drug shop 0.529 0.514 0.500 0.521 0.585 0.033∗∗ Clinic 0.391 0.379 0.411 0.405 0.369 0.371 Pharmacy 0.076 0.099 0.085 0.066 0.046 0.366 Runyankole language 0.520 0.523 0.502 0.539 0.518 0.129 English language 0.520 0.157 0.119 0.151 0.124 0.422 Luganda language 0.323 0.294 0.371 0.289 0.350 0.015∗∗

Back to Balancing Anne Fitzpatrick (University of Michigan) March 22, 2015 34 / 22

Full Balancing Table, Analysis Sample

Know Malaria & AL Know Only Malaria Know Only AL Control Cross Section Village & shopper FE Drug Shop 0.535 0.536 0.477 0.541 0.583 0.088∗ 0.043∗∗∗ Clinic 0.392 0.370 0.431 0.405 0.365 0.335 0.568 Pharmacy 0.073 0.095 0.092 0.055 0.052 0.081∗ 0.042∗ Runyankole 0.514 0.5120.505 0.518 0.522 0.978 0.235 Luganda 0.328 0.294 0.358 0.3140.343 0.405 0.058 Patient = Uncle 0.498 0.526 0.450 0.500 0.517 0.475 0.351 Weekend Visit 0.422 0.460 0.394 0.409 0.426 0.455 0.326 Afternoon Visit 0.661 0.654 0.670 0.659 0.661 0.986 0.825 Had baby in shop 0.084 0.090 0.078 0.083 0.084 0.965 0.908 Female Vendor 0.794 0.820 0.780 0.759 0.817 0.252 0.442 Shop Had No Name 0.402 0.422 0.362 0.400 0.422 0.444 0.809 Female Interviewer 0.556 0.573 0.541 0.582 0.530 0.673 −− Bargaining 0.593 0.626 0.583 0.591 0.574 0.678 0.694 Visit Order 1.557 1.545 1.500 1.568 1.613 0.466 0.430

Back to Balancing Anne Fitzpatrick (University of Michigan) March 22, 2015 35 / 22

Selection into Purchase?

Denied Sale Purchase Bought AL Bought SP Purchase Index Any Information 0.016

• 0.015
• 0.024

0.017

• 0.041

(0.014) (0.018) (0.027) (0.020) (0.036) Know Malaria Only

• 0.005

0.014

• 0.017

0.033 0.016 (0.014) (0.021) (0.033) (0.025) (0.041) Know AL Only 0.014

• 0.019
• 0.022

0.015

• 0.058

(0.016) (0.022) (0.033) (0.025) (0.043) Know Malaria & AL 0.036**

• 0.039
• 0.034

0.004

• 0.078*

(0.018) (0.024) (0.031) (0.022) (0.047) Constant 0.051** 0.933*** 0.808*** 0.004 0.023 (0.026) (0.046) (0.076) (0.042) (0.083) Pvalue Malaria= 0 0.0424 0.081 0.557 0.339 0.0984 Pvalue AL= 0 0.108 0.264 0.553 0.815 0.221 Observations 1016 1016 1016 1016 1016 R-squared 0.236 0.289 0.36 0.312 0.28 Number of clusters 495 495 495 495 495

Anne Fitzpatrick (University of Michigan) March 22, 2015 36 / 22

Some Evidence of Selection

Establishment type is related to purchase likelihood

No sale at a pharmacy was ever denied Clinics and drug shops have approximately the same purchase rates (80-85 %) Establishments charging consultation fees more likely to deny sales Purchases more likely in urban areas Language, visit order, and patient are unrelated to likelihood of purchase

Lee Bounds when using restricted sample [Lee,2009] Back to

prices Anne Fitzpatrick (University of Michigan) March 22, 2015 37 / 22

Index components

Purchase Index

• stockout, -denied, purchase, bought AL, bought quinine, bought SP

Price Index

price offer, price paid, highest price, lowest price, price variation, average price , - bargained

Menu Index

malaria test, express doubts about malaria, made a recommendation, number drugs offered, additional products

Profit Index

• ffer profitmargin, bought profitmargin, offer markup, bought markup,
• unitcost ordering , -unitcost

Service Quality

ask health questions,gave time, explain all options, very friendly, very unfriendly

Drug Quality

• gov drug, correct dosage, - counterfeit, substandard, fraction

substandard, fraction counterfeit

Anne Fitzpatrick (University of Michigan) March 22, 2015 38 / 22

Drug with Public Sector Markings

Back to

Market . Anne Fitzpatrick (University of Michigan) March 22, 2015 39 / 22

Real Customers Report Where They Typically Shop

“Where else do you typically shop?”

Visit Public Visit Private Nowhere else Buy Both Public & Private (1) (2) (3) (4) Real Customer with Information −0.104 −0.074 0.054 −0.121∗ (0.063) (0.045) (0.045) (0.064) Bought Adult 0.083 −0.005 0.016 0.085 (0.119) (0.058) (0.058) (0.120) Ln(Income) −0.068∗ −0.012 0.041∗ −0.04 (0.040) (0.037) (0.024) (0.044) Years of Education 0.009 0.001 0.004 (0.009) (0.007) (0.005) (0.010) Observations 321 322 321 321 R-squared 0.295 0.328 0.353 0.305

Back . Anne Fitzpatrick (University of Michigan) March 22, 2015 40 / 22

Do Vendors Know Quality?

Yes, at least some vendors know they are selling low quality drugs Supported by where they dispensed the drug from

Customers with any information are 3.9 percentage points more likely to have their drugs picked from the back of the outlet/out of sight Positive, but insignificant relationship between picking from the back and government drug/counterfeit/substandard

Most vendors know that government drugs have specific markings

Asked in pilot; little variation Some drug packages have the markings rubbed off

List randomization [Droitcour et al. 1991]

Randomly divide respondents into 2 groups, Treatment and Control Control: presented a list of non-sensitive activities Treatment: same list + 1 sensitive activity

Anne Fitzpatrick (University of Michigan) March 22, 2015 41 / 22

Do Vendors Know Quality?

Variable Bribed an NDA

• fficial

Gave antibiotics when not needed Knowingly Sell Fake Drugs List 1 List 2 (1) (2) (3) (4) Treatment 0.135 0.153 −0.059 0.479∗∗ (0.102) (0.100) (0.102) (0.238) p = 0.187 p = 0.126 p = 0.563 p = 0.047 List Randomization Version 0.055 −0.059 (0.129) (0.121) Constant 2.859∗∗∗ 2.343∗∗∗ 2.515∗∗∗ 1.610∗∗∗ (0.072) (0.072) (0.072) (0.159) Observations 448 448 362 86 R-squared 0.004 0.006 0.001 0.045

Back . Anne Fitzpatrick (University of Michigan) March 22, 2015 42 / 22

The Moral and Fiscal Implications of ART for HIV in Africa

Paul Collier, Olivier Sterck, Richard Manning

Moral duty to rescue: conditions

Moral duty to rescue 1) Treatment or certain death 2) Expensive for recipients 3) Cheap for donors 4) Limited risk-taking 5) Democratic moral views Moral duty to respond

 Less demanding than Utilitarian Universalism

Long-run implications

• Lifetime treatment
• If ART provided, moral obligation to continue
• Long-term fiscal liability
• Analogous to long term obligations for debt

service

Financial burden (350 cells/mm3)

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% Botswana Kenya Lesotho Malawi Nigeria South Africa Uganda Zimbabwe

External debt stock HIV treatment fiscal liability (% GDP)

Cost over time (Malawi)

0.0% 0.5% 1.0% 1.5% 2.0% 2.5% 3.0% 3.5% 4.0%

2015 2020 2025 2030 2035 2040 2045 2050

% GDP

Figure 1: cost of ART over time in per cent of GDP (GDP growth rate is assumed to be 4.3 per cent, which is the average growth rate of GDP in Malawi between 1960 and 2013)

222% 80%

00% 100% 200% 300% 400% 500% 600% 700% 800% 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12

Aggregated cost (% 2015 GDP) Discount rate

Cost and discounting (Malawi)

Prevention as an Investment

Prevention = investment to reduce the fiscal liability Simple two periods SI model:

• 𝑇0 and 𝑇1: susceptible in period 0 and 1
• 𝐽0 and 𝐽1: infected in period 0 and 1
• 𝑂0 : incidence
• 𝑞0: investment in prevention
• 𝑐ℎ > 𝑐𝑚: transmission rate of HIV without/with ART
• 𝑏0 and 𝑏1: share of infected who don’t need ART
• 𝑢0 and 𝑢1: share of infected who don’t need ART
• 𝑒: death rate without ART
• 𝜌𝑢 and 𝑑 𝑞0 : cost of ART and cost of prevention
• 𝑠: discount rate (marginal interest rate)

Prevention as an Investment

In period 1, we have: 𝑇1 = 𝑇0 − 𝑂0 𝐽1 = 𝐽0 + 𝑂0 − 𝐽0 (1 − 𝑏0)(1 − 𝑢0)𝑒 where incidence 𝑂0 = 1 − 𝑞0 𝑐𝑚 1 − 𝑏0 𝑢0𝐽0 + 𝑐ℎ(𝑏0 𝐽0 + (1 − 𝑏0)(1 − 𝑢0)𝐽0) 𝑇0. The moral duty to rescue implies that: 𝑢0 = 𝑢1 = 1  Objective to minimize the total cost of ART + prevention

min

𝑞0

𝑑 𝑞0 + 1 − 𝑏0 𝐽0 + 1 − 𝑏0 𝐽0 + 1 − 𝑏1 𝑏0𝐽0 + 𝑂0 𝜌𝑢/ 1 + 𝑠

Prevention as an Investment

Solution of the maximization program: Proposition 1 - Money should be devoted to prevention up to the point at which the marginal dollar spent on prevention reduces the cost of the moral duty to rescue, as measured by the discounted cost of treatment, by

• ne dollar.

1 − 𝑏1 𝑏0 (𝑐ℎ − 𝑐𝑚) + 𝑐𝑚 𝐽0(1 − 𝐽0)𝜌𝑢 1 + 𝑠 = 𝑑 𝑞0 ′

Prevention as an Investment

Comparative statics

1.

δp0

∗

δc′ > 0, δp0

∗

δπt > 0: The optimal level of prevention is decreasing in its marginal

cost and is increasing in the cost of treatment. 2.

δp0

∗

δr < 0: Prevention is decreasing in the marginal interest rate.

3.

δp0

∗

δbl > 0, δp0

∗

δbh > 0: The optimal level of prevention is increasing in the

transmission rate of HIV, both with and without ART. 4.

δp0

∗

δa0 > 0: The optimal prevention level is decreasing in the proportion of

PLHIV who need treatment in period 0. 5.

δp0

∗

δa1 < 0: Prevention is increasing in the proportion of people newly infected

who need ART.

Taken together, Propositions 4 and 5 imply that the relationship between optimal prevention and the CD4 count threshold determining eligibility is ambiguous and so can only be determined empirically.

Example: circumcision in Malawi

Large initial investment for long-term gains:

• Initial investment: $267 m
• Reduction in NPV of ART

cost: $734 m

• Net pay-off: $468 m

Total cost (ART + circumcision) 350 CD4/mm3 Circumcision coverage In million USD % GDP 2015 Baseline (22%) 4,019 80.7 30% 3,966 79.6 40% 3,900 78.3 50% 3,836 77.0 60% 3,774 75.7 70% 3,714 74.5 80% 3,657 73.4 90% 3,603 72.3 100% 3,552 71.3

Benefit of prevention

as a function of parameters

2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 01% 02% 03% 04% 05% 06% 07% 08% 09% 10% 11% 12%

Million US$

Marginal interest rate

• 200
• 100

100 200 300 400 500 600 200 400 600

Million US$

CD4 count threshold

Example: ART in Malawi

0% 20% 40% 60% 80% 100% 120%

100 200 300 400 500 600

% GDP in 2015 CD4 count threshold

Apportioning the Duty to Rescue

How to share the burden if high-prevalence countries can not afford it alone?

• Democratic or procedural approach (Daniels and Sabin,

2014)

• Debt relief thresholds (e.g. IMF: Debt/GDP = 40%)
• Mix of LIC criteria and HIV burden (Global Fund)

– counterpart from 5% for LIC to 60% for UMIC – Access for UMIC only if high disease burden

• Econometric analysis of donors’ behaviour

Dependent variable: Log(International spending for HIV per capita) Tight FE OLS Quantile reg. Neigh. Pair Random (1) (2) (3) (4) (5) Log(GDP per cap.)

• 0.622***
• 0.493***
• 0.552***
• 0.614***
• 0.685***

(0.153) (0.166) (0.163) (0.194) (0.163)

HIV prevalence

20.83*** 16.25*** 20.29*** 22.06*** 21.72*** (2.471) (3.358) (3.186) (4.003) (4.715)

Constant

4.229*** 3.715*** 2.446** 2.640* 2.980*** (1.098) (1.275) (1.237) (1.468) (1.019)

Observations

93 93 357 550 164

Sample

All All All All All

Revealed preference of donors

Dependent variable: Log(Domestic spending for HIV per capita) Tight FE OLS Quantile reg. Neigh. Pair Random (1) (2) (3) (4) (5) Log(GDP per cap.)

0.923*** 0.822*** 1.154*** 1.080*** 1.019*** (0.134) (0.144) (0.202) (0.223) (0.183)

Constant

• 7.736***
• 6.758***
• 11.12***
• 12.11***
• 12.68***

(1.053) (1.126) (1.841) (1.89) (1.719)

Observations

118 118 466 734 201

Sample

All All All All All

Capacity of recipients

Figure 7 - Local funding percentage based on revealed preference of the International Community

Apportionment principles

Extension of the model to 2 players: donor & recipient

• 1. A broadly similar apportionment between countries

and donors should be applied to treatment of future infection and prevention:

• Only this strategy for dividing responsibility can avoid moral

hazard

• This simple rule may be slightly affected by the fact that

discount rates are different in donors and recipient countries (which gives donors incentive to invest more in prevention)

• 2. Lower discount rate for donors  incentive for them

to contribute more early on.

Conclusion

• Moral duty to rescue widely recognized
• Less demanding than Universal Utilitarianism
• Creates a fiscal liability
• For many African countries this is macro-significant
• Prevention is an investment to reduce the total cost
• A broadly similar apportionment between countries

and donors should be applied to treatment of future infection and prevention

• More work needed to improve estimates

Thank you

Psychic vs. Economic Barriers to Vaccine Take-up

Ryoko Sato

University of Michigan

March 24, 2015

Ryoko Sato (Univ. of Michigan) March 24, 2015 1 / 30

Vaccinations save lives, but low take-up

Vaccinations avert millions of deaths every year (WHO, 2012) Tetanus:

◮ One of the major causes of neonatal mortality in developing countries

(Oruamabo, 2007)

◮ Cut umbilical cord with a non-sterile instrument (neonatal tetanus) ◮ Tetanus-toxoid vaccine: Most effective to prevent maternal and

neonatal tetanus (Khan et al., 2013)

Tetanus-toxoid vaccination rate

◮ Worldwide: 82 percent (WHO, 2011) ◮ Nigeria: 52.8 percent (DHS,2013) Ryoko Sato (Univ. of Michigan) March 24, 2015 2 / 30

If the benefits of vaccination are high, why is vaccination take-up low?

Psychic costs of vaccination

◮ Fear of side effects (Jheeta and Newell, 2008; Nichter, 1995) ◮ Fear of needles (Deacon and Abramowitz, 2006)

Polio Vaccine Boycott in Nigeria, 2003: “the polio vaccine could make women infertile or contract HIV” (Jegede 2007) Kenya Catholic Church tetanus vaccine fears, 2014 Oct: “Catholic bishops called to stop the rollout of the vaccination campaign, saying it was a plot to target women of child-bearing age.” (BBC)

Ryoko Sato (Univ. of Michigan) March 24, 2015 3 / 30

Research Questions

Do psychic costs of vaccination reduce vaccination take-up? What are other barriers?

◮ Cash incentives ◮ Disease message which emphasizes disease severity ◮ Social networks Ryoko Sato (Univ. of Michigan) March 24, 2015 4 / 30

Sample

Jada Local Government Area in Adamawa State, Nigeria

◮ Nigeria accounted for 16 % of world neonatal tetanus deaths ◮ Northern Nigeria has low vaccination rates

Sample

◮ 2,482 Women ◮ From 80 villages in Jada within Catchment area of 10 Health Clinics

Eligibility

◮ Women aged 15-35 ◮ Not received a tetanus vaccine in the past 6 months ◮ Priority: Pregnant women, Never received tetanus vaccine before Ryoko Sato (Univ. of Michigan) March 24, 2015 5 / 30

Sample

Baseline Characteristics (N=2482) Mean SD Age 25.1 6.12 Muslim 0.496 0.500 No Education 0.483 0.500 Household Earning Per Capita Per Day ($) 1.31 1.81 Pregnant 0.182 0.386 Ever Used Clinic 0.722 0.448 Distance to Clinic (km) 1.708 1.230 Transportation cost (one way, $) 0.412 0.658 Ever Received Tetanus Vaccine 0.398 0.490

Ryoko Sato (Univ. of Michigan) March 24, 2015 6 / 30

Research Design: Psychic costs of vaccination

Randomize conditions under which respondents receive cash rewards

◮ Cost under Vaccine CCT = Transportation cost + Psychic costs ◮ Cost under Clinic

CCT = Transportation cost

✷✱✹✽✷ ❲♦♠❡♥

❱❛❝❝✐♥❡ ❈❈❚

❈❧✐♥✐❝ ❱✐s✐t ✰ ❱❛❝❝✐♥❛t✐♦♥

❈❧✐♥✐❝ ❈❈❚

❈❧✐♥✐❝ ❱✐s✐t

❈♦♥❞✐t✐♦♥ ❢♦r ❈❛s❤ ■♥❝❡♥t✐✈❡s

Ryoko Sato (Univ. of Michigan) March 24, 2015 7 / 30

Empirical strategy: Psychic costs of vaccination

Yij = α + β1VaccineCCTij + X ′

ijµ + ǫij

Y : Clinic Attendance for woman i in village j VaccineCCT: Cash incentive (CCT) to go to clinic and vaccinate (Comparison: ClinicCCT: Cash incentive to go to clinic) Standard errors are clustered by village

Ryoko Sato (Univ. of Michigan) March 24, 2015 8 / 30

Psychic costs of vaccination

No evidence of large psychic costs of vaccination

• Dependent. Var.

Clinic attendance Vaccine CCT 0.002 [0.016] Constant 0.454** [0.144] Observations 2482 R-squared 0.021 Mean of dependent variable 0.737 Covariates X Fixed effect by village (80 villages) X

Notes: Control group is the group of women under Clinic CCT. Ryoko Sato (Univ. of Michigan) March 24, 2015 9 / 30

Psychic costs of vaccination

95.7 percent of respondents who visited a clinic under Clinic CCT received a vaccine, even though that was not required

Ryoko Sato (Univ. of Michigan) March 24, 2015 10 / 30

Psychic costs of vaccination

Why have they never received a vaccine previously? (asked before intervention)

Reasons why respondents Reasons why respondents’ children have not received have not received any vaccines (N=195) any vaccines (N=233) (1) (2) Lack of information 0.369 0.120 Psychic costs of vaccination 0.174 0.180 Clinic too far 0.169 0.150 Supply-side problem 0.046 0.180 Not enough money 0.031 0.077 Misconception of vaccination 0.021 0.120 No particular reason 0.169 0.133 Other 0.021 0.030 Notes: Psychic costs: fear of injection, side effect, do not like vaccine, tradition does not

• allow. Supply-side problem: lack of vaccine stock, health workers do not visit villages.

Misconception: healthy people do not have to take vaccines, infants should not receive vaccines in first 40 days. Descriptive Psychic Costs Ryoko Sato (Univ. of Michigan) March 24, 2015 11 / 30

Psychic costs of vaccination

Results: No evidence of psychic costs of vaccination as the major barriers to receiving a vaccine

◮ Why? ⋆ cannot differentiate psychic costs of vaccination and psychic costs of

clinic attendance

Ryoko Sato (Univ. of Michigan) March 24, 2015 12 / 30

Psychic costs of vaccination and of clinic visit

Under Clinic CCT Not attend clinic = 0 Attend clinic but not vaccinate = Bh + τ Attend clinic and vaccinate = Bh + τ+ Bv Under Vaccine CCT Not attend clinic = 0 Attend clinic but not vaccinate = Bh Attend clinic and vaccinate = Bh + τ+ Bv

Ryoko Sato (Univ. of Michigan) March 24, 2015 13 / 30

Psychic Costs

𝐶ℎ 𝐶𝑤 −𝜐 −𝜐 1 2 0=Do Not Attend Clinic 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine Under Clinic CCT

Ryoko Sato (Univ. of Michigan) March 24, 2015 14 / 30

Psychic Costs

𝐶ℎ 𝐶𝑤 −𝜐 −𝜐 1 2 0=Do Not Attend Clinic 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine Under Vaccine CCT

Ryoko Sato (Univ. of Michigan) March 24, 2015 15 / 30

Psychic Costs

𝐶ℎ 𝐶𝑤 −𝜐 −𝜐 1 2 0=Do Not Attend Clinic 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine Clinic CCT Vaccine CCT

1 under Clinic CCT, 0 under Vaccine CCT 2 under Vaccine CCT, 1 under Clinic CCT Ryoko Sato (Univ. of Michigan) March 24, 2015 16 / 30

Other Barriers to Vaccination

Cash incentives

◮ 3 different size of cash incentives (CCT)

Disease message (priming about disease severity)

◮ Salient picture of tetanus patients vs. no picture

Social networks

◮ 3 units of social networks ⋆ Village ⋆ Neighborhood (GPS) ⋆ Friends Ryoko Sato (Univ. of Michigan) March 24, 2015 17 / 30

Overall Research Design

✷✱✹✽✷ ❲♦♠❡♥

❈❂✵✳✼✸✼ ❱❂✵✳✼✷✻

❈❧✐♥✐❝ ❱✐s✐t ✰ ❱❛❝❝✐♥❛t✐♦♥

✭❱❛❝❝✐♥❡ ❈❈❚✮ ❱❂✵✳✼✸✹

❙❝❛r❡❞✲str❛✐❣❤t

❱❂✵✳✼✶✼

✽✵✵

♥❛✐r❛ ❱❂✵✳✽✹✽

✸✵✵

♥❛✐r❛ ❱❂✵✳✼✺✺

✺

♥❛✐r❛ ❱❂✵✳✺✸✺

❈♦♥tr♦❧

❱❂✵✳✼✹✽

✽✵✵

♥❛✐r❛ ❱❂✵✳✽✻✸

✸✵✵

♥❛✐r❛ ❱❂✵✳✼✾✵

✺

♥❛✐r❛ ❱❂✵✳✺✼✺

❈❧✐♥✐❝ ❱✐s✐t

✭❈❧✐♥✐❝ ❈❈❚✮ ❈❂✵✳✼✹✸ ❱❂✵✳✼✶✷

❈♦♥tr♦❧

❈❂✵✳✼✹✸ ❱❂✵✳✼✶✷

✽✵✵

♥❛✐r❛ ❈❂✵✳✽✽✼ ❱❂✵✳✽✺✷

✸✵✵

♥❛✐r❛ ❈❂✵✳✼✺✷ ❱❂✵✳✼✷✹

✺

♥❛✐r❛ ❈❂✵✳✺✻✸ ❱❂✵✳✺✸✶ ❆♠♦✉♥t ♦❢ ❈❛s❤ ■♥❝❡♥t✐✈❡s ▼❡ss❛❣❡ ❈♦♥❞✐t✐♦♥ ❢♦r ❈❛s❤ ■♥❝❡♥t✐✈❡s

❈❂❈❧✐♥✐❝ ❆tt❡♥❞❛♥❝❡✱ ❱❂❱❛❝❝✐♥❛t✐♦♥ ❚❛❦❡✲✉♣

Ryoko Sato (Univ. of Michigan) March 24, 2015 18 / 30

Effect of CCT

Strong positive effect of CCT

• Dependent. Var.

Clinic attendance Vaccinated (1) (2) Vaccine CCT

• 0.011

0.021 [0.032] [0.036] CCT=300 0.168*** 0.171*** [0.039] [0.041] CCT=800 0.284*** 0.282*** [0.038] [0.039] CCT=300 * (Vaccine CCT) 0.047 0.044 [0.042] [0.046] CCT=800 * (Vaccine CCT)

• 0.000

0.001 [0.038] [0.041] Constant 0.326** 0.357** [0.145] [0.145] Observations 2482 2482 R-squared 0.113 0.110 Mean of dependent variable 0.737 0.726 Covariates X X Fixed effect by village (80 villages) X X Notes: Control group is the group of women under Clinic CCT and under CCT=5. Robust standard errors clustered by villages (80 villages) are presented. *** Significant at the 1 percent level, ** Significant at the 5 percent level, * Significant at the 10 percent level Ryoko Sato (Univ. of Michigan) March 24, 2015 19 / 30

Priming about disease severity

Scared-straight Flipchart Tetanus is very dangerous and painful (esp for babies) Typical symptoms of tetanus: (1) Severe pain (2) Muscle spasm Control Flipchart Tetanus is very dangerous and painful (esp for babies) Typical symptoms of tetanus: (1) Severe pain (2) Muscle spasm

Ryoko Sato (Univ. of Michigan) March 24, 2015 20 / 30

Effect of priming

No effect of priming on vaccination take-up

• Dependent. Var.

Vaccinated (1) Vaccine CCT& Fear

• 0.025

[0.016] Constant 0.520** [0.144] Observations 2482 R-squared 0.022 Mean of dependent variable 0.726 Covariates X Fixed effect by village (80 villages) X

Notes: Control group is the group of women under Vaccine CCT. Robust standard errors clustered by villages (80 villages) are presented. *** Significant at the 1 percent level, ** Significant at the 5 percent level, * Significant at the 10 percent level Ryoko Sato (Univ. of Michigan) March 24, 2015 21 / 30

Effect of priming

Priming increased the perceived severity of tetanus

• Dependent. Var.
• No. people die

Very worried Tetanus Protection Heart rate Perception on tetanus

• f tetanus

about tetanus is very bad very important (1) (2) (3) (4) (5) Vaccine CCT& Fear 2.529** 0.143*** 0.138*** 0.104*** 6.270*** [1.175] [0.028] [0.026] [0.026] [0.701] Constant 47.037*** 0.426* 0.018 0.419** 39.829*** [11.693] [0.235] [0.190] [0.171] [6.230] Observations 2280 2283 2283 2283 2091 R-squared 0.090 0.147 0.111 0.119 0.404 Mean of dependent variable 38.159 0.612 0.697 0.771 89.554 Covariates X X X X X Fixed effect by village (80 villages) X X X X X Notes: Control group is the group of women under Vaccine CCT. Robust standard errors clustered by villages (80 villages) are presented. *** Significant at the 1 percent level, ** Significant at the 5 percent level, * Significant at the 10 percent level. Ryoko Sato (Univ. of Michigan) March 24, 2015 22 / 30

Effect of Social networks

To measure the effect of social network on vaccination uptake: Yij = α + β1NumVaccinatedij + β2NumWomenij + X ′

ijµ + εij

First stage NumVaccinatedij = α + δ1OfferedCCT800ij + δ2NumWomenij +X ′

ijµ + ǫij

NumVaccinated: Number of respondents in a social network who received a vaccine NumWomen: Total number of respondents in a social network OfferedCCT800: Number of respondents in a social network who were offered the highest amount of CCT (CCT=800)

First stage result Ryoko Sato (Univ. of Michigan) March 24, 2015 23 / 30

Social networks (IV)

Strong positive peer effects on vaccination uptake

• Dependent. Var.

Vaccinated (1) (2) (3) Num Vaccinated in village 0.024*** [0.004] Num Vaccinated in 100 meters 0.036 *** [0.013] Num friends Vaccinated 0.172* [0.091] Constant 0.494** 0.440**

• 0.164

[0.142] [0.150] [0.149] Observations 2482 2482 2482 R-squared 0.255 0.278 0.350 Mean of dependent variable 0.726 0.726 0.726 Covariates X X X Fixed effects X X X Ryoko Sato (Univ. of Michigan) March 24, 2015 24 / 30

Conclusion

Psychic costs of vaccination: NOT large barriers to vaccination

◮ No additional incentives needed for vaccination at the clinic

Strong effect of cash incentives and peers

◮ Small cash increased vaccination by 17.1 percentage points ◮ One additional friend increased one’s vaccination by 17.2 percentage

points

Ryoko Sato (Univ. of Michigan) March 24, 2015 25 / 30

Thank you!

Ryoko Sato (Univ. of Michigan) March 24, 2015 26 / 30

Additional Slides

Ryoko Sato (Univ. of Michigan) March 24, 2015 27 / 30

Psychic costs of vaccination

(asked before intervention)

Do you agree that: (1) Vaccine gives HIV 0.184 Needles for injections are scary 0.618 Vaccine causes headache and fever 0.662 Vaccine gives me disease 0.271

Back Ryoko Sato (Univ. of Michigan) March 24, 2015 28 / 30

Comparison of My Sample and DHS Sample

Sample: My Sample DHS Sample (N=2,482) (N=23,306) Mean (1) (2) Age 25.1 26.7 Muslim 0.496 0.573 No Education 0.483 0.496 Not married 0.153 0.021 Pregnant 0.182 0.140 Have children 0.764 0.963 Ever Received Tetanus Vaccine 0.398 0.318

Back Ryoko Sato (Univ. of Michigan) March 24, 2015 29 / 30

Social networks: first stage

• Dependent. Var.

Num Vaccinated Num Vaccinated Any friends in village in 100 meters Vaccinated (1) (2) (3) Offered CCT800 in village 2.292*** [0.503] Offered CCT800 in 100 meters 0.542 *** [0.135] Friends offered CCT800 0.245*** [0.032] Constant 2.033** 2.267**

• 0.206**

[2.621] [1.398] [0.115] Observations 2482 2482 2482 R-squared 0.958 0.852 0.745 Mean of dependent variable 36.956 9.947 0.268 Mean of independent variable 16.988 4.605 0.133 Covariates X X X Fixed effect by health clinic (10 clinics) X X Fixed effect by village (80 villages) X Notes: Robust standard errors clustered by villages (80 villages) are presented. *** Significant at the 1 percent level, ** Significant at the 5 percent level, * Significant at the 10 percent level Back Ryoko Sato (Univ. of Michigan) March 24, 2015 30 / 30