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

Service Quality Falls: Checklist Express Advised Asked Service Doubts Malaria Health Quality About Test Question Index Malaria (1) (2) (3) (4) − 0 . 068 ∗ − 0 . 047 ∗ − 0 . 128 ∗∗∗ Any Information − 0 . 044 (0 . 033) (0 . 037) (0 . 026) (0 . 028) (1) (2) (3) (4) − 0 . 070 ∗∗ Know Only Malaria − 0 . 057 − 0 . 02 − 0 . 018 (0 . 039) (0 . 047) (0 . 031) (0 . 035) − 0 . 134 ∗∗∗ Know Only Drug 0 . 009 − 0 . 022 − 0 . 026 (0 . 043) (0 . 046) (0 . 033) (0 . 035) − 0 . 082 ∗∗ − 0 . 158 ∗∗∗ − 0 . 097 ∗∗∗ − 0 . 178 ∗∗∗ Know Malaria & Drug (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 p i − e ( s i ) − c q + α q max i Π i ( p i ) p i , s i , q ∈{ G , B } (1) subject to θ s i − p i ≥ 0 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, θ i , 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 + α i Π( θ s i )) = e ′ ( s i ) (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 + α i Π( θ s i )) = e ′ ( s i ) (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 + α i Π( θ s i )) = e ′ ( s i ) (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 − c G + α i Π i ≥ A − c B (3) α i Π i ≥ c G − c B 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?” Y st = λ 0 + λ 1 AnyInformation st + γ v + ψ ′ X + µ st 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 No Difference Information Information (N=195) (N=178) (1) (2) (3) Income (USD) 152 . 03 152 . 01 − 0 . 03 − 0 . 95 ∗∗ Education 8 . 927 9 . 881 − 0 . 166 ∗∗∗ Repeat Customer 0 . 698 0 . 853 0 . 04 ∗ Bought a Full Dose 0 . 982 0 . 946 − 0 . 12 ∗∗∗ Bargained over price 0 . 610 0 . 727 Bought AL 0 . 539 0 . 670 − 0 . 13 − 0 . 23 ∗∗∗ Patient took Malaria Test 0 . 191 0 . 420 − 0 . 16 ∗ Bought Additional Product 0 . 475 0 . 632 − 0 . 65 ∗∗∗ Product Price (USD) 2 . 189 2 . 835 − 1 . 36 ∗∗∗ Total Bill (USD) 2 . 336 3 . 697 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 testing . Back to 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 Outline . Back to Anne Fitzpatrick (University of Michigan) March 22, 2015 25 / 22

Exit Interview Responses Mapped to Experiment Know Ask TOTAL Drug Rec Know Malaria 85 44 129 0.228 0.118 Ask for Diagnosis 49 195 244 0.131 0.523 TOTAL 134 239 373 Treatment . Back to Anne Fitzpatrick (University of Michigan) March 22, 2015 26 / 22

Prices Vary Bargaining . Back to 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 Bargaining . Back to 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 Bargaining . Back to Anne Fitzpatrick (University of Michigan) March 22, 2015 29 / 22

Types of Drugs Purchased All Active (1) (2) (3) (4) (5) Ingredients 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) 0 . 145 ∗∗∗ 0 . 086 ∗∗ 0 . 026 ∗∗∗ Visit Order 3 (0 . 008) (0 . 035) (0 . 044) 0 . 030 ∗∗∗ 0 . 049 Visit Order 4 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 Know Know Control Village Malaria Only Only & & AL Malaria AL shopper FE 0 . 033 ∗∗ Drug shop 0 . 529 0 . 514 0 . 500 0 . 521 0 . 585 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 0 . 015 ∗∗ Luganda language 0 . 323 0 . 294 0 . 371 0 . 289 0 . 350 Back to Balancing Anne Fitzpatrick (University of Michigan) March 22, 2015 34 / 22

Full Balancing Table, Analysis Sample Know Know Know Control Cross Village Malaria Only Only Section & & AL Malaria AL 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 offer 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 Market . Back to 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 Nowhere Buy Both Private else 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 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 Gave Knowingly Sell Fake NDA antibiotics Drugs official when not needed List 1 List 2 (1) (2) (3) (4) 0 . 479 ∗∗ Treatment 0 . 135 0 . 153 − 0 . 059 (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) 2 . 859 ∗∗∗ 2 . 343 ∗∗∗ 2 . 515 ∗∗∗ 1 . 610 ∗∗∗ Constant (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 1) Treatment or certain death 2) Expensive for recipients Moral duty Moral duty 3) Cheap for donors to rescue to respond 4) Limited risk-taking 5) Democratic moral views  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) 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Botswana Kenya Lesotho Malawi Nigeria South Africa Uganda Zimbabwe External debt stock HIV treatment fiscal liability (% GDP)

Cost over time (Malawi) 4.0% 3.5% 3.0% 2.5% % GDP 2.0% 1.5% 1.0% 0.5% 0.0% 2015 2020 2025 2030 2035 2040 2045 2050 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)

Cost and discounting (Malawi) 800% Aggregated cost (% 2015 GDP) 700% 600% 500% 400% 300% 222% 200% 80% 100% 00% 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12 Discount rate

Prevention as an Investment Prevention = investment to reduce the fiscal liability Simple two periods SI model: • 𝑇 0 and 𝑇 1 : susceptible in period 0 and 1 • 𝐽 0 and 𝐽 1 : infected in period 0 and 1 • 𝑂 0 : incidence • 𝑞 0 : investment in prevention • 𝑐 ℎ > 𝑐 𝑚 : transmission rate of HIV without/with ART • 𝑏 0 and 𝑏 1 : share of infected who don’t need ART • 𝑢 0 and 𝑢 1 : share of infected who don’t need ART • 𝑒 : death rate without ART • 𝜌 𝑢 and 𝑑 𝑞 0 : cost of ART and cost of prevention • 𝑠 : discount rate (marginal interest rate)

Prevention as an Investment In period 1, we have: 𝑇 1 = 𝑇 0 − 𝑂 0 𝐽 1 = 𝐽 0 + 𝑂 0 − 𝐽 0 (1 − 𝑏 0 )(1 − 𝑢 0 ) 𝑒 where incidence 𝑐 𝑚 1 − 𝑏 0 𝑢 0 𝐽 0 + 𝑐 ℎ ( 𝑏 0 𝐽 0 + (1 − 𝑏 0 )(1 − 𝑢 0 ) 𝐽 0 ) 𝑇 0 . 𝑂 0 = 1 − 𝑞 0 The moral duty to rescue implies that: 𝑢 0 = 𝑢 1 = 1  Objective to minimize the total cost of ART + prevention 1 − 𝑏 0 𝐽 0 + 1 − 𝑏 1 𝑏 0 𝐽 0 + 𝑂 0 𝜌 𝑢 / 1 + 𝑠 min 𝑑 𝑞 0 + 1 − 𝑏 0 𝐽 0 + 𝑞 0

Prevention as an Investment Solution of the maximization program: 1 − 𝑏 1 𝑏 0 ( 𝑐 ℎ − 𝑐 𝑚 ) + 𝑐 𝑚 𝐽 0 (1 − 𝐽 0 ) 𝜌 𝑢 = 𝑑 𝑞 0 ′ 1 + 𝑠 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 one dollar .

Prevention as an Investment Comparative statics ∗ ∗ δp 0 δc′ > 0, δp 0 1. δπ t > 0: The optimal level of prevention is decreasing in its marginal cost and is increasing in the cost of treatment. ∗ δp 0 2. δr < 0: Prevention is decreasing in the marginal interest rate. ∗ ∗ δp 0 δb l > 0, δp 0 3. δb h > 0: The optimal level of prevention is increasing in the transmission rate of HIV, both with and without ART. ∗ δp 0 4. δa 0 > 0: The optimal prevention level is decreasing in the proportion of PLHIV who need treatment in period 0. ∗ δp 0 5. δa 1 < 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 Total cost (ART + circumcision) Large initial investment 350 CD4/mm3 Circumcision In million for long-term gains: coverage USD % GDP 2015 Baseline (22%) 4,019 80.7 • Initial investment: $267 m 30% 3,966 79.6 40% 3,900 78.3 • Reduction in NPV of ART 50% 3,836 77.0 cost: $734 m 60% 3,774 75.7 • Net pay-off: $468 m 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 600 16,000 500 14,000 400 12,000 10,000 Million US$ 300 Million US$ 8,000 200 6,000 100 4,000 0 0 200 400 600 2,000 -100 0 -200 01% 02% 03% 04% 05% 06% 07% 08% 09% 10% 11% 12% CD4 count threshold Marginal interest rate

Example: ART in Malawi 120% 100% 80% % GDP in 2015 60% 40% 20% 0% 100 200 300 400 500 600 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

Revealed preference of donors 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

Capacity of recipients 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

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 ✷✱✹✽✷ ❲♦♠❡♥ ❈❧✐♥✐❝ ❈❈❚ ❱❛❝❝✐♥❡ ❈❈❚ ❈♦♥❞✐t✐♦♥ ❢♦r ❈❛s❤ ■♥❝❡♥t✐✈❡s ❈❧✐♥✐❝ ❱✐s✐t ❈❧✐♥✐❝ ❱✐s✐t ✰ ❱❛❝❝✐♥❛t✐♦♥ Ryoko Sato (Univ. of Michigan) March 24, 2015 7 / 30

Empirical strategy: Psychic costs of vaccination α + β 1 VaccineCCT ij + X ′ Y ij = ij µ + ǫ ij 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 = B h + τ Attend clinic and vaccinate = B h + τ + B v Under Vaccine CCT Not attend clinic = 0 Attend clinic but not vaccinate = B h Attend clinic and vaccinate = B h + τ + B v Ryoko Sato (Univ. of Michigan) March 24, 2015 13 / 30

Psychic Costs Under Clinic CCT 0=Do Not Attend Clinic 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine 𝐶 ℎ 1 0 2 −𝜐 0 𝐶 𝑤 −𝜐 0 Ryoko Sato (Univ. of Michigan) March 24, 2015 14 / 30

Psychic Costs Under Vaccine CCT 0=Do Not Attend Clinic 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine 𝐶 ℎ 1 2 0 −𝜐 0 𝐶 𝑤 −𝜐 0 Ryoko Sato (Univ. of Michigan) March 24, 2015 15 / 30

Psychic Costs Clinic CCT 0=Do Not Attend Clinic Vaccine CCT 1= Attend Clinic but Refuse Vaccine 2=Attend Clinic and Receive Vaccine 𝐶 ℎ 2 under 1 Vaccine CCT, 1 under Clinic CCT 0 2 1 under Clinic CCT, 0 under Vaccine CCT −𝜐 0 𝐶 𝑤 −𝜐 0 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 ❈❧✐♥✐❝ ❱✐s✐t ✰ ❱❛❝❝✐♥❛t✐♦♥ ❈♦♥❞✐t✐♦♥ ✭❈❧✐♥✐❝ ❈❈❚✮ ✭❱❛❝❝✐♥❡ ❈❈❚✮ ❢♦r ❈❛s❤ ❈❂✵✳✼✹✸ ■♥❝❡♥t✐✈❡s ❱❂✵✳✼✶✷ ❱❂✵✳✼✸✹ ❈♦♥tr♦❧ ❈♦♥tr♦❧ ❙❝❛r❡❞✲str❛✐❣❤t ▼❡ss❛❣❡ ❈❂✵✳✼✹✸ ❱❂✵✳✼✶✷ ❱❂✵✳✼✹✽ ❱❂✵✳✼✶✼ ✺ ✸✵✵ ✽✵✵ ✺ ✸✵✵ ✽✵✵ ✺ ✸✵✵ ✽✵✵ ❆♠♦✉♥t ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐r❛ ♥❛✐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 Control Flipchart Tetanus is very dangerous and Tetanus is very dangerous and painful (esp for babies) painful (esp for babies) Typical symptoms of tetanus: Typical symptoms of tetanus: (1) Severe pain (1) Severe pain (2) Muscle spasm (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 of 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: α + β 1 NumVaccinated ij + β 2 NumWomen ij + X ′ Y ij = ij µ + ε ij First stage NumVaccinated ij = α + δ 1 OfferedCCT 800 ij + δ 2 NumWomen ij + X ′ ij µ + ǫ ij NumVaccinated : Number of respondents in a social network who received a vaccine NumWomen : Total number of respondents in a social network OfferedCCT 800: 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