Micro-loans, bednets and malaria: Evidence from a ...dse.univr.it/ssdev/documents/material2010/tarozzi_slides.pdf · Micro-loans, bednets and malaria: Evidence from a randomized controlled

Micro-loans, bednets and malaria:Evidence from a randomized controlled trial

in Orissa (India)

Alessandro Tarozzi (Duke Economics)Aprajit Mahajan (Stanford Economics)

Brian Blackburn (Stanford School of Medicine)Dan Kopf (Independent)

Lakshmi Krishnan (Yale Environmental Management)Joanne Yoong (RAND Corporation)

SUMMER WORKSHOP IN DEVELOPMENTAlba di Canazei

July 30 - August 3 2010

Tarozzi, Mahajan et al. Micro-loans, Bednets and Malaria 1 / 57

1 Background and motivation

2 Study design

3 Data

4 Results (Intent-to-Treat)Net take-upNet usageRe-treatment decisionsBiomarkers

5 Interpretation of results

6 Conclusions


Motivation: Malaria

Caused in humans by four species of Plasmodium, (P. falciparum (Pf) thedeadliest), transmitted through the bite of Anopheles mosquitoes

Huge global burden

In 2002, ≈ 1/3 of humans estimated to live in malarious areasRecent estimates of 300-660 million Pf cases annually, 80 million inIndia alone> 1 million deaths per year, mostly young children (U5)

Even when not fatal, health effects can be dire:

Debilitating fevers, anemia, respiratory problems, hypoglycemiaPregnancy complications, increased risk of low birth weightPermanent neurologic and developmental impairment for childrenLong term consequences on human capital accumulation: Barreca2010, Bleakley 2007, Cutler et al. 2010, Hong 2007a,b, Lucas 2010.

No vaccine currently. Only vector control, prophylaxis (IRS and ITNs) andtreatment


Motivation: Malaria


Huge global burden






Motivation: Malaria


Huge global burden






Motivation: Malaria


Huge global burden






Motivation: Insecticide Treated Nets (ITNs)

Multiple Randomized Evaluations have demonstrated that intensivedistribution/use bednets and especially ITNs can provide largebenefits, esp. among PW/U5 (Lengeler 2004), e.g.

Incidence, prevalence, mortality ↓Hemoglobin levels & child nutritional status ↑, premature births ↓Number and parasite loads of Anopheles ↓Community effects (with high coverage)


Public health in developing countries: Free or Cost Sharing?Free distribution of ITNs often advocated as the best public healthintervention (Sachs 2005, WHO 2007)

Demand for health-protecting technologies very steep and small co-paymentsoften wipe out demand: Cohen and Dupas 2010, Kremer and Miguel 2007,Ashraf, Berry and Shapiro 2008, Holla and Kremer 2009.

However, free provision is not always feasible and some argue not desirable

Cost-sharing to improve targeting and reduce wastage (but see Cohenand Dupas, Ashraf et al)

I Can micro-loans offer a viable “third way”?

Randomized evaluation in 141+ villages in malaria-endemic Orissa (India):

Free ITNs offered for freeMF ITNs offered at full cost on credit

C control areas

Primary Objectives:

1 Evaluate impact on ITN ownership and usage.2 Evaluate impact on health.









C control areas

Primary Objectives:










C control areas

Primary Objectives:










C control areas

Primary Objectives:










C control areas

Primary Objectives:1 Evaluate impact on ITN ownership and usage.2 Evaluate impact on health.


Unique Features of RCTwithin Public Health/Epidemiology Literature

First large-scale ITN-related RCT:

1 in India

2 low coverage & limited monitoring

3 micro-loans (“social marketing”)

4 free distribution vs. “social marketing”


Commitment devices and compliance

Several Public Health measures require compliance with appropriatefollow-up behavior (e.g. deworming drugs, nets & retreatment)

ITNs: with cost sharing, re-treatment is rare even when nets are regularlyused

I Can appropriately designed contracts help by including futurecompliance costs upfront?

Two loan contract types, endogenously chosen:

C1 Loan for ITN only. Re-treatment offered for cash laterC2 Loan for ITN + 2 re-treatments 6 and 12 months later

Re-treatment rates C2 ≈ twice as large as C1.Weak association between contract choice and hh characteristics


















Preview of findings

53% of households purchased at least one net in MF villages but totalnet ownership at follow-up is ≈ 1.5 persons/net in Free villages, ≈ 2persons/net in MF, ≈ 3 persons/net in control areas.

Self-reported ITN usage rates at follow-up

Previous night: 0.47 (Free), 0.16 (MF), 0.03 (Control).During peak mosquito seasons: 0.77 (Free), 0.36 (MF), 0.07 (Control)

No statistically significant differences in malaria prevalence orhemoglobin levels, measured with rapid diagnostic tests (RDTs):

Malaria prevalence: 0.18 (Control), 0.23 (MF), 0.22 (Free).Hemoglobin (g/dl): 11.4 (anemia 36%) in all arms.


Preview of findings







Preview of findings








2 Study design

3 Data



6 Conclusions


Study Design: LocationStudy Design: Location

Malaria: “number one public health problem” in Orissa (OHDR, 2004)

2003 Dept of Health and Family Welfare data show 417,000 cases of malaria (83%falciparum).

High % of self-reported malaria (NFHS-1999) in our study districts: 8.5% inSambalpur and Bargarh, 8.8% in Balangir, 12.3% in Keonjhar, 17.2% in Phulbani.

Tarozzi et al. (Duke, RAND and Stanford) Micro-loans, Bednets and Malaria April 2009 1 / 1

Orissa: < 5% of Indian population, but in 2007 recorded 25% of malaria cases(365K), 44% of Pf (317K), 18% deaths.


Study Design: Sampling

Random sample from ≈ 900 villages where micro-lender BISWA operates

Villages from blocks with no free distribution of nets in foreseeablefuture.District-stratified selection of 141 villages.

Up to 15 households with a BISWA member selected at baseline. Sample:10,067 individuals in 1,845 households.

25 non-intervention villages added at follow-up from same randomizationscheme + 10 new households in baseline villages (from BPL2002 lists)




















Study Design: Timeline

1 Baseline survey (May-June 2007)

Demographics, socio-economic status, debt, KAP about malaria andbednets, measures of risk aversion and time preferences, sleepingpatterns, biomarkers

2 Intervention (net sale/free delivery) (October-November 2007)

Brief Information campaign (IC) everywhere + free delivery or saleRecord take up and contract choice for sample households

3 1st retreatment with insecticide (March-April 2008)

Record re-treatment decisions and selected information on ITN usageand (self-reported) malaria status

4 2nd retreatment (September-November 2008)

5 Follow-up survey (December 2008 - April 2009)

Broadly similar to baseline, but expanded sample.














































Details of Intervention

Everywhere: short IC about bednets and malaria

Free distribution: function of demographic structure of household (4maximum)

Loans: All nets offered with interest rate 20% per year. One-year loan,flexible repayment schedule. Two contract types:

C1 ITN only. Rs 173 (single) or Rs 223 (double). Retreatment every sixmonths for cash, Rs 15 (single) or Rs 18 (double).

C2 “Commitment” product: ITN + two retreatments. Rs 203 (single)or Rs 259 (double). Retreatment at no additional cost.






























ITN Cost in PerspectiveD

ensit

y, ba

ndw

idth

=15

0

Total monthly hh. expenditure per head0 2500587173

0

.0005

.001

.0015

Official poverty line for rural Orissa(Rs 381 person/month)

Median

Price of the least expensive net(single with no re-treatment option)



2 Study design

3 Data



6 Conclusions


Selected summary statistics and randomization testsControl Free MF p-value s.dev.

SC/ST/OBC 0.945 0.933 0.912 0.421 0.256hhsize 5.466 5.593 5.309 0.138 2.221No. children U5 0.499 0.506 0.487 0.892 0.704Male household head 0.952 0.941 0.932 0.368 0.235H. Head has some schooling 0.72 0.706 0.714 0.908 0.452Head 2ndary education or above 0.084 0.075 0.114 0.123 0.287Expenditure per Head (daily) 22.291 21.156 24.238 0.085* 16.272Expenditure per Head (excl. cerem.) 19.078 18.593 20.602 0.157 12.881Nets (per capita) 0.287 0.264 0.311 0.167 0.315ITNs (per capita) 0.021 0.046 0.055 0.027** 0.157Used net last night (%) 0.131 0.116 0.162 0.195 0.304Used ITN last night (%) 0.019 0.022 0.03 0.617 0.137Use regularly nets (%) 0.564 0.512 0.572 0.304 0.458Malaria prevalence (RDT) 0.108 0.116 0.123 0.841 0.299Hemoglobin 10.962 10.743 10.983 0.132 1.821Anemia prevalence (Hb< 11, RDT) 0.527 0.569 0.504 0.121 0.457

Overall good balance, but ITN ownership/use higher in control areas.

Biomarkers, by age and gender Biomarkers, by village Blood tests


Attrition and Changes in Household Composition

Overall little attrition (4%), and similar across arms

Household attrition

Similar changes in entry/exit from household and in demographiccomposition

Entry into & exit from households

Changes in household demographic structure



2 Study design

3 Data



6 Conclusions


ITN Take-up by village, IC data, fall 2007

010

2030

0 .5 1 1.5 2 0 .5 1 1.5 2

Free MFFr

eque

ncy

Village mean of # BISWA nets received (per person)


ITN Take Up and Changes in Net Ownership

Take-up Yi,Post = βFreeFree + βMFMF + ui

Follow-up Ownership: Yi,Post = β0 + βFreeFree + βMFMF + ui

Follow-up Ownership (DD): Yi,Post − Yi,Pre = β0 + βFreeFree + βMFMF + ui

(1) (2) (3) (4) (5)ITN Take up (Fall 2007) All nets (Winter 08/09)

Intervention: Free and MF arms only Follow-upAny ITN ITNs ITNs Nets Nets owned

Dependent variable Delivered delivered delivered owned DD> 0 only (per capita) (per capita) (per capita)

Free 0.96 2.77 0.52 0.270 0.278(0.020) (0.047) (0.015) (0.027) (0.023)

MF 0.53 2.24 0.24 0.143 0.115(0.045) (0.295) (0.038) (0.029) (0.023)

Intercept (Control) 0.364 0.070(0.019) (0.015)

Difference: Free − MF 0.43 0.53 0.28 0.13 0.16p-value (H0 : MF=Free) 0.0000 0.0809 0.0000 0.0000 0.0000

Observations 1246 933 1245 1774 1762R-squared 0.81 0.66 0.55 0.12 0.10no. clusters 94 89 94 141 141









Free 0.96 2.77 0.52 0.270 0.278(0.020) (0.047) (0.015) (0.027) (0.023)

MF 0.53 2.24 0.24 0.143 0.115(0.045) (0.295) (0.038) (0.029) (0.023)












Free 0.96 2.77 0.52 0.270 0.278(0.020) (0.047) (0.015) (0.027) (0.023)

MF 0.53 2.24 0.24 0.143 0.115(0.045) (0.295) (0.038) (0.029) (0.023)












Free 0.96 2.77 0.52 0.270 0.278(0.020) (0.047) (0.015) (0.027) (0.023)

MF 0.53 2.24 0.24 0.143 0.115(0.045) (0.295) (0.038) (0.029) (0.023)






2 Study design

3 Data



6 Conclusions


Differences in Bednet Usage: Previous night

Data from pre and post-intervention surveys, baseline householdsSelf-reported

(1) (2) (3) (4) (5) (6)Previous night

Dependent variable Any Any ITN ITN Untreated Untreatednet net net net

Free 0.358 0.378 0.446 0.458 -0.085 -0.083[0.038]*** [0.036]*** [0.029]*** [0.031]*** [0.025]*** [0.026]***

MF 0.125 0.092 0.139 0.126 -0.012 -0.035[0.038]*** [0.034]*** [0.024]*** [0.026]*** [0.030] [0.026]

Intercept (Control) 0.175 0.049 0.022 0.003 0.148 0.048[0.025]*** [0.019]** [0.006]*** [0.007] [0.023]*** [0.016]***

DD no yes no yes no yes

Difference: Free − MF 0.23 0.29 0.31 0.33 -0.07 -0.05p-value (H0 : MF=Free) 0.0000 0.0000 0.0000 0.0000 0.0014 0.0995

Observations 9105 7746 9054 7686 9054 7686R-squared 0.1 0.091 0.203 0.198 0.014 0.007no. clusters 141 141 141 141 141 141


Differences in “Usual” Bednet Usage

Data from pre and post-intervention surveys, baseline householdsSelf-reported

(1) (2) (3) (4)Usual in peak season

Dependent variable Any Any ITN Untreatednet net net

Free 0.266 0.332 0.707 -0.438[0.034]*** [0.037]*** [0.031]*** [0.039]***

MF 0.171 0.178 0.293 -0.119[0.037]*** [0.035]*** [0.037]*** [0.047]**

Intercept (Control) 0.66 0.089 0.065 0.59[0.032]*** [0.022]*** [0.015]*** [0.032]***

DD no yes no no

Difference: Free − MF 0.09 0.15 0.41 -0.32p-value (H0 : MF=Free) 0.0001 0.0002 0.0000 0.0000

Observations 9525 8482 9388 9388R-squared 0.078 0.061 0.352 0.145no. clusters 141 141 141 141



2 Study design

3 Data



6 Conclusions


Re-treatment Summary

Two contracts:

C1: Net only, retreatment later for cashC2: Net + 2 retreatments

Significantly higher six-month retreatment rates with “commitment”

C1: 36%C2: 84%Free: 92%

Lower 12-month retreatment rates, even with free provision butespecially in MF communities with no “commitment”.

C1: 21%C2: 74%Free: 83%

Details



Two contracts:



C1: 36%C2: 84%Free: 92%


C1: 21%C2: 74%Free: 83%

Details



Two contracts:



C1: 36%C2: 84%Free: 92%


C1: 21%C2: 74%Free: 83%

Details



2 Study design

3 Data



6 Conclusions


Impact on malaria and anemia (Hb< 11g/dl): Panel households only

(1) (2)

(3) (4) (5) (6)

+ve Malaria

Hemoglobin Anemic (Hb< 11g/dl)

Follow-up DD

Follow-up DD Follow-up DD

Free distribution= 1 0.036 0.053

-0.036 0.217 0.01 -0.024

[0.030] [0.040]

[0.105] [0.106]** [0.022] [0.033]

Micro-loans= 1 0.044 0.061

0.023 0.045 0.005 0.035

[0.035] [0.039]

[0.094] [0.123] [0.021] [0.035]

Constant 0.184 0.065

11.434 0.278 0.384 -0.111

[0.022]*** [0.028]**

[0.064]*** [0.075]*** [0.012]*** [0.024]***

Only panel individuals No Yes

No Yes No Yes

Observations 7173 1900

7168 1873 7168 1873

No. clusters (villages) 141 141

141 141 141 141

R-squared 0.0021 0.0035

0.0002 0.0034 0.0021 0.0035

Free=MF (p-value) 0.817 0.8216

0.5852 0.1650 0.8305 0.0951

Free=MF=0 (p-value) 0.3655 0.2438

0.8610 0.1099 0.8899 0.2468

Hemoglobin distribution By age & gender



(1) (2) (3) (4)

(5) (6)

+ve Malaria Hemoglobin

Anemic (Hb< 11g/dl)

Follow-up DD Follow-up DD

Follow-up DD

Free distribution= 1 0.036 0.053 -0.036 0.217

0.01 -0.024

[0.030] [0.040] [0.105] [0.106]**

[0.022] [0.033]

Micro-loans= 1 0.044 0.061 0.023 0.045

0.005 0.035

[0.035] [0.039] [0.094] [0.123]

[0.021] [0.035]

Constant 0.184 0.065 11.434 0.278

0.384 -0.111

[0.022]*** [0.028]** [0.064]*** [0.075]***

[0.012]*** [0.024]***

Only panel individuals No Yes No Yes

No Yes

Observations 7173 1900 7168 1873

7168 1873

No. clusters (villages) 141 141 141 141

141 141

R-squared 0.0021 0.0035 0.0002 0.0034

0.0021 0.0035

Free=MF (p-value) 0.817 0.8216 0.5852 0.1650

0.8305 0.0951

Free=MF=0 (p-value) 0.3655 0.2438 0.8610 0.1099

0.8899 0.2468




(1) (2) (3) (4) (5) (6)+ve Malaria Hemoglobin Anemic (Hb< 11g/dl)

Follow-up DD Follow-up DD Follow-up DD

Free distribution= 1 0.036 0.053 -0.036 0.217 0.01 -0.024[0.030] [0.040] [0.105] [0.106]** [0.022] [0.033]

Micro-loans= 1 0.044 0.061 0.023 0.045 0.005 0.035[0.035] [0.039] [0.094] [0.123] [0.021] [0.035]

Constant 0.184 0.065 11.434 0.278 0.384 -0.111[0.022]*** [0.028]** [0.064]*** [0.075]*** [0.012]*** [0.024]***

Only panel individuals No Yes No Yes No YesObservations 7173 1900 7168 1873 7168 1873No. clusters (villages) 141 141 141 141 141 141R-squared 0.0021 0.0035 0.0002 0.0034 0.0021 0.0035

Free=MF (p-value) 0.817 0.8216 0.5852 0.1650 0.8305 0.0951Free=MF=0 (p-value) 0.3655 0.2438 0.8610 0.1099 0.8899 0.2468




2 Study design

3 Data



6 Conclusions


Why the lack of health benefits?

1 Changes in risk-avoiding behavior;

2 Biases in the self-reports on net usage;

3 Reliability of the tests;

4 Systematic differences in testing refusals across arms;

5 Insufficient village-wide coverage of the interventions.






























1. Changes in other precautionary behavior?

(1) (2) (3) (4)Control Free MF Test of equality

(A) Causes (p-values)Drinking contaminated water 0.105 > 0.059 0.073 0.055*Mosquito bites 0.845 < 0.892 0.854 0.058*Contaminated environment 0.116 < 0.131 0.148 0.447Don’t know 0.037 0.025 0.051 0.065*(B) Malaria-avoiding behaviorNets 0.819 < 0.866 0.830 0.139ITNs 0.023 0.023 0.017 0.718Proper clothing (long sleeves etc) 0.004 < 0.008 0.010 0.268Avoid drinking contaminated water 0.076 > 0.054 0.058 0.471Insecticides 0.009 0.008 0.017 0.352Repellents/mosquito coils 0.030 > 0.020 0.020 0.554Smoke 0.016 < 0.023 0.022 0.622Clearing stagnant water 0.028 > 0.021 0.022 0.702Cleaning drainage system/sewage 0.054 < 0.075 0.087 0.093*Avoiding contaminated environments 0.158 < 0.170 0.211 0.151Proper diet 0.051 0.039 0.037 0.618Medicine 0.042 0.033 0.066 0.058*Other ways 0.035 0.021 0.027 0.469Don’t know 0.035 0.030 0.024 0.608(C) Residual spraying of wallsInner walls sprayed in 2008-09 0.403 0.368 0.296 0.242Outer walls sprayed in 2008-09 0.531 0.481 0.442 0.580


2. Are nets really kept/used?Use information from actual observation of nets at follow-up

(1) (2) (3) (4) (5) (6)Slept under Surveyor was Slept under Slept under Slept under Slept under

a net allowed to see a net seen by a net in good a BISWA net a net seenthe net surveyor conditions, seen seen by hanging properly

by surveyor surveyor by surveyor

Free 0.375 0.075 0.360 0.293 0.472 0.037[0.039]*** [0.043]* [0.038]*** [0.024]*** [0.030]*** [0.011]***

MF 0.135 0.037 0.127 0.093 0.133 0.009[0.037]*** [0.046] [0.036]*** [0.017]*** [0.022]*** [0.008]

Intercept 0.170 0.851 0.144 0.044 0.002 0.018[0.025]*** [0.041]*** [0.024]*** [0.010]*** [0.002] [0.005]***

Observations 8018 2780 8018 8018 8018 8018Clusters 141 128 141 141 141 141R-squared 0.1077 0.0089 0.1049 0.1040 0.2406 0.0078

Free=MF 0.0000 0.1189 0.0000 0.0000 0.0000 0.0161Free=MF=0 0.0000 0.0911 0.0000 0.0000 0.0000 0.0044


3. Are the malaria RDTs reliable?

August 2009, in collaboration with S K Sharma at the Malaria ResearchCentre Field Station, Rourkela.

Compare 3 RDT readings (2 of the testers and the most senior surveymonitor) with results from microscopy. 205 tests from 3 villages.

Correlations among 3 readings and microscopy:

RDT(1) RDT(2) RDT(3)RDT(2) 0.7873*RDT(3) 0.7844* 0.8760*Microscopy 0.5274* 0.6131* 0.5968*

* Significant at 1% level.

Errors of inclusion and exclusion:

Microscopy Microscopy Microscopy-ve +ve -ve +ve -ve +ve

R1 -ve 129 1 R2 -ve 148 3 R3 -ve 146 3+ve 45 30 +ve 26 28 +ve 28 28
































Likely overestimates of current prevalence (as expected)

Possible reader-specific fixed effects

Differences somewhat decrease when we include tester-FE.

(1) (2) (3) (4)Follow-up only DD Follow-up only Malaria, DD

Free=1 0.038 0.056 0.021 0.038[0.030] [0.040] [0.026] [0.036]

MF=1 0.046 0.063 0.023 0.046[0.036] [0.039] [0.029] [0.036]

Intercept 0.184 0.064 0.379 0.227[0.022]*** [0.028]** [0.043]*** [0.047]***

Tester FE No No Yes YesObservations 7105 1886 7105 1886R-squared 0.0023 0.0039 0.0467 0.0415Clusters 141 141 141 141Free=MF 0.8140 0.8581 0.9502 0.8200Free=MF=0 0.3308 0.2179 0.6479 0.3971






Free=1 0.038 0.056 0.021 0.038[0.030] [0.040] [0.026] [0.036]

MF=1 0.046 0.063 0.023 0.046[0.036] [0.039] [0.029] [0.036]

Intercept 0.184 0.064 0.379 0.227[0.022]*** [0.028]** [0.043]*** [0.047]***







Free=1 0.038 0.056 0.021 0.038[0.030] [0.040] [0.026] [0.036]

MF=1 0.046 0.063 0.023 0.046[0.036] [0.039] [0.029] [0.036]

Intercept 0.184 0.064 0.379 0.227[0.022]*** [0.028]** [0.043]*** [0.047]***



5. Was the scale of the program insufficient to have an impact?Existing large-scale clustered RCTs achieved significantly larger coverage

35

Mala

ria p

reva

lenc

e (F

ree)

n=25

76

(A) Winter 08/09, baseline households only(# program ITNs)/(Village pop.)

Village-level prevalence Fitted values Fitted values, no outliers

0 .1 .2 .3 .4 .50

.1

.2

.3

.4

.5

.6

Mala

ria p

reva

lenc

e (F

ree)

n=65

4

(B) Change, Spring 07 to Winter 08/09(# program ITNs)/(Village pop.)

(mean) d_m Fitted values Fitted values, no outliers

0 .1 .2 .3 .4 .5-.4

-.2

0

.2

.4

.6

.8

Mala

ria p

reva

lenc

e (F

ree)

n=22

61

(C) Winter 08/09, baseline households only(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .50

.1

.2

.3

.4

.5

.6

Mala

ria p

reva

lenc

e (M

F)n=

626

(D) Change, Spring 07 to Winter 08/09(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .5-.4

-.2

0

.2

.4

.6

.8

Figure 4: Malaria Prevalence vs. Intensity of ITNs DistributionNote: Data from winter 2007 and winter 2008/09. All graphs make use only of biomarkers collected from individuals inbaseline households. Each circle represents a village. Each graph also shows fitted values of two village-level OLS regressionsof prevalence (or its change) on ITN coverage. The dotted lines are fitted values when we exclude villages with coverage largerthan 0.35 (corresponding to the vertical line). The point estimates and standard errors (sd, in parenthesis) of the regressionsare as follows: (A) 0.56 (0.13) and 0.12 (0.19) excluding outliers; (B) 0.39 (0.18) and 0.18 (0.56) excluding outliers; (C) -0.03(0.46) and 0.40 (0.70) excluding outliers; (D) -0.45 (0.41) and -0.52 (0.76) excluding outliers.


5. Was the scale of the program insufficient to have an impact?Existing large-scale clustered RCTs achieved significantly larger coverage

35

Mala

ria p

reva

lenc

e (F

ree)

n=25

76

(A) Winter 08/09, baseline households only(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .50

.1

.2

.3

.4

.5

.6

Mala

ria p

reva

lenc

e (F

ree)

n=65

4

(B) Change, Spring 07 to Winter 08/09(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .5-.4

-.2

0

.2

.4

.6

.8

Mala

ria p

reva

lenc

e (F

ree)

n=22

61

(C) Winter 08/09, baseline households only(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .50

.1

.2

.3

.4

.5

.6

Mala

ria p

reva

lenc

e (M

F)n=

626

(D) Change, Spring 07 to Winter 08/09(# program ITNs)/(Village pop.)


0 .1 .2 .3 .4 .5-.4

-.2

0

.2

.4

.6

.8

Figure 4: Malaria Prevalence vs. Intensity of ITNs DistributionNote: Data from winter 2007 and winter 2008/09. All graphs make use only of biomarkers collected from individuals inbaseline households. Each circle represents a village. Each graph also shows fitted values of two village-level OLS regressionsof prevalence (or its change) on ITN coverage. The dotted lines are fitted values when we exclude villages with coverage largerthan 0.35 (corresponding to the vertical line). The point estimates and standard errors (sd, in parenthesis) of the regressionsare as follows: (A) 0.56 (0.13) and 0.12 (0.19) excluding outliers; (B) 0.39 (0.18) and 0.18 (0.56) excluding outliers; (C) -0.03(0.46) and 0.40 (0.70) excluding outliers; (D) -0.45 (0.41) and -0.52 (0.76) excluding outliers.


Conclusions

Micro-loans lead to a large and significant increase in ITN ownershipand (self-reported) usage, but

did not replicate the close-to-full coverage achieved with freedistribution.also (in progress) preliminary evidence suggests that the baseline surveyled to significant “priming”

Increased ITN ownership and usage not reflected in any significantimprovement in health outcomes

Potentially important implications for actual ITN distributionprograms unable to achieve full coverage.


Conclusions






Conclusions






Biomarkers

Malaria: Binax Now fingerprick RDT

Requires < 0.5ml blood, results can be read in 15min.Well-validated, can distinguish Pf from other infections.Current or very recent infection (2-4 weeks at most).

Baseline: Test attempted with all PW, all U5 and their mothers andone randomly selected adultFollow-up: all members in baseline household, and U10 and 18-40 innew households.

Hemoglobin (Hb): HemoCue. < 0.5ml blood, 15min.

Same sample as above.

Back to timeline Back to randomization


Biomarkers








Biomarkers








The Malaria RDT

ies, to create a test strip. This test strip is mounted in a book-shaped, hinged test device, along with wash and absorbent pads, intend-ed to aid in the clearing of the membrane when the device is closed.

To perform the test, whole blood is applied to the sample pad. Malarial antigen present in the sample reacts to bind the anti-malariaconjugated antibody. Reagent A is added to the bottom of the test strip and allows the antigen-conjugate complexes to migrate alongthe test strip, where they are captured by the immobilized antibodies, forming the Test Line(s). Immobilized control antibody capturescontrol conjugate, forming the Control Line. Once the blood sample has migrated the length of the test strip, the device is closed, allow-ing Reagent A that has been added to the wash pad to clear the test strip of excess blood.

Test results are interpreted by the presence or absence of visually detectable pink-to-purple colored lines. A positive test result, read in15 minutes, will include the detection of both a Test Line (or Test Lines) and a Control Line. A negative test result, read in 15 min-utes, will produce only a Control Line, indicating that malarial antigens were not detected in the sample. Failure of the Control Line toappear, whether the Test Line(s) is present or not, indicates an invalid result.

REAGENTS AND MATERIALS

Materials Provided

BinaxNOW® Malaria Test Kit:

Test Devices: A cardboard, book-shaped, hinged test device containing the test strip

Reagent A: Tris buffer containing detergent and sodium azide

Capillary tubes: EDTA capillary tubes used to transfer whole blood samples obtained via fingerstickto the test devices

MATERIALS REQUIRED BUT NOT PROVIDEDNegative Quality Control (pool of 3 – 5 EDTA whole blood samples)Positive Quality Control (EDTA whole blood sample containing P. falciparum)Lancets, sterile wipes or pads, clock, timer or stopwatch

Note: When pipetting sample, use a calibrated pipette capable of delivering a 15 µl volume.

PRECAUTIONS

1. For in vitro diagnostic use.2. Leave test device sealed in its foil pouch until just before use.3. Do not use kit past its expiration date. 4. Do not mix components from different kit lots.5. Samples and Reagent A must be added as described in the test procedure to obtain optimal sample flow and test performance.

The following precautions should be taken when adding Reagent A to the test device.a. To ensure delivery of the appropriate volume of Reagent A to both pads on the test device, hold the vial vertically,

1/2 - 1 inch above the pads and slowly add free falling drops.

b. When adding Reagent A to the white pad directly below the purple sample pad, allow the first drop to absorb complete-ly into the pad before adding the second drop. A third drop of Reagent A may be added to this pad if necessary – seeTest Procedure, Step 3.

6. If using venous blood, mix sample by tapping the tube or vial gently, and before sampling, prime the pipette tip by drawing thesample into the tip and expelling it a couple of times.

7. If using blood obtained via fingerstick, use the capillary tubes supplied in the test kit to deliver the blood to the test device andfill the entire volume of the tube.

8. Patient samples and test devices should be handled as though they are capable of transmitting disease. Observe establishedprecautions against bloodborne pathogens. Do not reopen or reuse test cards.

9. Excessive air circulation (i.e. air conditioners, fans, etc.) can slow the flow of the sample. During testing, protecting the devicesfrom excessive air flow is recommended.

10. When interpreting test results, use a bright, unfiltered light. 11. All capillary tubes and pipette tips are single use items – do not use with multiple specimens. Contamination of dispensing equip-

ment, containers or reagents can lead to inaccurate results.12. Reagent A contains sodium azide as a preservative. Sodium azide is toxic and should be handled carefully, avoiding ingestion

or skin contact. It may react with lead or copper plumbing to form explosive metal azides. Flush with a liberal volume of waterwhen disposing of unwanted reagent.

STORAGE AND STABILITY

Store kit at 2-37°C (36-98.6°F). The BinaxNOW® Malaria Test Kit and reagents are stable until the expiration dates marked on theirouter packaging and containers when stored as specified.

QUALITY CONTROL

Warning:This test should only be used by laboratories that have or can acquire blood samples containing Plasmodiumfalciparum for use as a positive control. It is recommended that the level of the positive control used chal-lenge the assay cutoff.

Daily Quality Control:The BinaxNOW® Malaria Test has built-in procedural controls. For daily quality control, the manufacturer recommends that you recordthese controls for each test run.

Procedural Controls:A. The pink-to-purple line at the “C” (Control) position in a tested device can be considered an internal positive procedural control. If

the sample flows and the reagents work, this line will always appear. B. The clearing of background color from the result window is a negative background control. The background color in the window

should be light pink to white at 15 minutes. Background color should not hinder reading of the test.

2

External Positive and Negative Controls:Good laboratory practice recommends that positive and negative controls be run with each new shipment or lot to ensure that:• test reagents are working, and • the test is being correctly performed

For training purposes, it is recommended that all first time users of the test perform external control testing prior to running patientsamples.

For a negative control, a pool of 3 - 5 EDTA whole blood samples from presumed malaria negative individuals can be used. For a pos-itive control, an EDTA whole blood sample containing P. falciparum can be used.

Other controls must be tested in order to conform with:• local, state and/or federal regulations,• accrediting groups, and/or,• your laboratory’s standard Quality Control procedures

Refer to CLSI EP12-A and 42 CFR 493.1256 for guidance on proper QC practices (U.S. customers only).

If the correct control results are not obtained, do not report patient results. Contact Binax Technical Service during normal businesshours (EST).

SPECIMEN COLLECTION AND HANDLING

Collect venous blood, by the standard venipuncture procedure, into an EDTA tube. Test whole blood samples as soon as possible aftercollection. If the test cannot be performed immediately, the blood may be stored for up to three days at 2° to 30°C (36-86°F). Ifblood is refrigerated, allow it to come to room temperature (15-30°C) prior to testing. Mix gently before testing. If microscopy con-firmation of a BinaxNOW® negative test result is necessary on a venous blood sample that has been stored, appropriate criteria for thehandling of samples used for microscopy should be followed. In some cases, it may be necessary to obtain a fresh sample from thepatient.

To obtain capillary blood via puncture of a finger, cleanse the area with a sterile wipe or pad and dry. Use a lancet to puncture the skinand collect the blood directly into the EDTA capillary tube provided in the test kit. Fill the entire capillary tube with blood and use imme-diately.

TEST PROCEDURE

See the Specimen Collection and Handling section for information regarding sample collection. Ensure that all blood samples arewarmed to room temperature prior to use.

Remove test device from pouch just prior to use. Open the device and lay it flat on the work surface.

1. If using a capillary blood sample, slowly apply blood from the capillary tubeto cover the entire PURPLE sample pad on the right side of the device. Thisis done by holding the capillary tube vertically and gently pressing the endagainst the purple pad in several places. Once the pad is saturated, properlydiscard the capillary tube. The test may not require all of the blood that hasbeen collected into the capillary tube. Go to Step 2.

If using a venous blood sample, prime the pipette tip by drawing up sample and expelling it a couple of times. Then slowlyadd 15 µl of blood to the bottom half of the PURPLE sample pad. Go to Step 2.

IMPORTANT: Incorrect addition of sample may lead to an invalid or uninterpretable test.

2. There is a white pad immediately below the purple sample pad. Hold theReagent A bottle vertically and add two (2) free-falling drops ofReagent A to this white pad. Allow the first drop to absorb into thepad before adding the second drop. Do not add Reagent A directlyto the purple pad.

3. Allow the blood sample to run up the full length of the test strip. Do notallow the blood to run into or under the absorbent pad at the top of the strip,as doing so will hinder optimal washing (clearance) of the test strip.

NOTE: If blood flow up the test strip appears to stall or is less than halfway up thestrip after one (1) minute, add one (1) additional drop of Reagent A to the whitepad at the bottom of the test strip (below the sample pad where the blood wasadded).

4. Just before the blood sample reaches the base of the white absorbent padlocated at the top of the test strip, SLOWLY add four (4) free-fallingdrops of Reagent A to the wash pad on the top left-hand side of the testdevice, allowing each drop to absorb into the pad before adding the next.Note that the third and fourth drops may not completely absorb into thepad.

5. When the sample just reaches the base of the white absorbent pad at the top of the test strip,remove the adhesive liner from the right edge of the device, and close the device. This allows theReagent A to wash (clear) the blood sample off the test strip. To ensure good device closure and testflow, press very firmly along the entire edge to the right of the result window.

3

6. Read the test result through the viewing window 15 minutes after closing the test device. Results read before or after 15 min-utes may be inaccurate.

Note: When reading test results, tilt the device to reduce glare on the result window, if necessary.

RESULT INTERPRETATION

Valid Test ResultsThe Control Line (C) will appear on all valid tests and, when it is present, test results are interpreted as follows. Note that the appear-ance of any Test Line, even when very faint, indicates a positive result.

TEST RESULTS DESCRIPTION / INTERPRETATION

T1 Positive Positive result for P. falciparum (P.f.)

T2 Positive Positive result for P. vivax (P.v.) or P. malariae (P.m.) or P. ovale (P.o.) In somecases the appearance of only the T2 Line may indicate a mixed infection withtwo or more of P.v., P.m., and P.o.

T1 + T2 Positive Positive result for P. falciparum (P.f.) In some cases the appearance of both theT1 and T2 Lines may indicate a mixed infection of P.f. with another species.

No T1 or T2 Lines Negative result (no malaria antigens were detected)

Invalid and / or The test is invalid if the Control (C) Line does not appear, whether a Test Uninterpretable Line(s) is present or not.Test Results

The test is uninterpretable if the background color hinders reading of the testresult at 15 minutes. Invalid or uninterpretable tests can occur due to impropersample or Reagent A addition. Consult the Test Procedure section andPrecaution # 5 before repeating testing with a new device. Call TechnicalService if the problem persists.

REPORTING OF RESULTS

Result Suggested Report

T1 Positive Positive for P. falciparum protein antigen only

T2 Positive Positive for malaria protein antigen, representing P. vivax or P. malariae or P. ovale or a mix ofthese. Differentiation of the species is not possible.

T1 and T2 Positive Positive for P. falciparum protein antigen. In some cases this may represent a mix of P. falci-parum antigen with P. vivax, P. malariae, or P. ovale protein antigen. Differentiation between aP.f. only infection and a mixed infection containing P.f. and another malaria species is not possi-ble with this test. Microscopy must be performed to make this determination, as well as to dif-ferentiate among the non-falciparum Plasmodium species.

Negative Presumptive negative for malaria antigens. Infection due to malaria cannot be ruled out. Malariaantigen in the sample may be below the detection limit of the test. Negative results must beconfirmed by thin / thick smear microscopy.

LIMITATIONS

A negative test result does not exclude infection with malaria, particularly at low levels of parasitemia. Therefore, the results obtainedwith the BinaxNOW® Malaria Test should be used in conjunction with other laboratory and clinical findings to make an accurate diag-nosis. As is often done in serial microscopy testing, another sample can be collected and retested.3

The BinaxNOW® Malaria Test detects antigen from both viable and non-viable malaria organisms, including gametocytes4 andsequestered P. falciparum parasites5. Test performance depends on antigen load in the specimen and may not directly correlate withmicroscopy performed on the same specimen.

Performance of the BinaxNOW® Malaria Test has not been established for monitoring treatment of malaria. Residual plasmodium anti-gen may be detected for several days following elimination of the parasite by anti-malarial treatment.4

Samples with positive rheumatoid factor (Rf) titers may produce false positive results in the BinaxNOW® Malaria Test. Rheumatoid fac-tors are autoantibodies, and positive Rf titers are associated with acute autoimmune disorders, such as rheumatoid arthritis, as well aswith chronic viral infections (such as hepatitis C) and parasitic infections.6 In addition, positive Rf titers are present in 1 to 4% of thegeneral population.7 Like other rapid malaria antigen detection tests6, the BinaxNOW® test has been shown to generate false positiveresults in samples of some individuals with positive Rf titers (see Performance Characteristics section).

Analytical reactivity testing demonstrates that the pan malarial test line (T2) on the BinaxNOW® test is capable of detecting all fourmalaria species (P.f., P.v., P.o., or P.m.). However, during clinical trials, insufficient data was generated to support clinical performanceclaims for the detection of P.m. or P.o. Clinical performance claims for this test are made for P.f. and P.v. detection only.

The test is not intended for use in screening asymptomatic populations.

EXPECTED VALUES

Malaria is a serious parasitic disease and is a major health problem in much of the tropics and subtropics. The rate of positive resultsfound in malaria testing is dependent on many factors including the method of specimen collection, the test method used, geograph-ic location, and the disease prevalence in specific localities. P. falciparum infection is considered to be the most serious and is oftenfatal, while infections with the other species such as P. vivax are typically less fatal.2

In a clinical study conducted in 2001 in areas considered to be endemic for malaria, the average prevalence of P. falciparum (as deter-mined by microscopy) in symptomatic patients was 14%, and the average prevalence of P. vivax was 29%. The prevalence of P. ovale,

4

P. malariae, and mixed infections of P.f. and P.v. was significantly less, totaling less than 2% in the population tested. When only thepan malarial (T2) line appears in the result window of the BinaxNOW® Malaria Test, it is likely that the infection is due to the pres-ence of P.v., rather than P.m. or P.o., given the relatively low incidence of these two species in most areas of the world. Areas of WestAfrica, where P.o. is common, and P.v. is rare, may be an exception to this general rule.8,9

In a multi-site study conducted in the eastern US in 2005-2006, 217 whole blood specimens, collected from adult hospitalized patientsand outpatients with fever or history of fever, were tested in the BinaxNOW® Malaria Test. Two hundred and sixteen (216 – 99.5%)of these presumed negative patients, who were living in areas with a low incidence of malaria, produced negative BinaxNOW® testresults.

PERFORMANCE CHARACTERISTICS

Clinical Sample Performance - BinaxNOW® Malaria Test Sensitivity & Specificity – Endemic Population:The performance of the BinaxNOW® test was compared to Giemsa malaria microscopy in a multi-center prospective study conductedin 2001 outside the U.S., in regions considered endemic for malaria. A total of 4,122 whole blood specimens collected from patientspresenting with malaria-like symptoms were evaluated on the BinaxNOW® test. Microscopy was considered positive only when asex-ual malaria forms were detected, since asexual forms (not gametocytes) are indicative of active infection.

Forty-four percent (1,796/4,122) of the tested population was microscopy positive for malaria, including 557 patients with P.f.,1,187 with P.v., 16 with P.m., 2 with P.o., and 34 with mixed P.f./P.v. infections. Fifty-nine percent of patients were male, 41%female, 19% pediatric (<18 years) and 81% adult (>18 years). BinaxNOW® test performance for detection of the individual malar-ia species and for mixed P.f./P.v. infections is summarized below.

No differences in BinaxNOW® Malaria Test performance were observed based on patient age or gender. BinaxNOW® test specificity forP.f. trends slightly lower (89.4%) in the 5% of patients who were on anti-malarial drug therapy, than in patients not receiving thera-py (94.4%), but does not achieve statistical significance.

BinaxNOW® Malaria test performance on samples with low hematocrit and with high hematocrit values was equivalent to its perform-ance on the overall study population.

Detection of P.f. InfectionBinaxNOW® test sensitivity and specificity for detection of P.f. vs. microscopy is presented below. Sensitivity was evaluated based onthe levels of parasitemia (parasites per µl) observed in microscopy.

BinaxNOW® Malaria Test Sensitivity and Specificity for P.f. vs. Microscopy

SENSITIVITY FOR P.f.

Parasitemia Level % Sensitivity 95%CI

> 5000 99.7% (326 / 327) 98 - 100%

1000 – 5000 99.2% (126 / 127) 96 - 100%

500 – 1000 92.6% (25 / 27) 76 - 99%

100 – 500 89.2% (33 / 37) 75 - 97%

0 – 100 53.9% (21 / 39) 37 - 70%

Overall 95.3% (531 / 557) 93 - 97%

SPECIFICITY FOR P.f.

% Specificity 95% CI

94.2% (3297 / 3500) 93-95%

Detection of P.v. InfectionBinaxNOW® test sensitivity and specificity for detection of P.v. vs. microscopy is presented below. Sensitivity was evaluated based onthe levels of parasitemia (parasites per µl) observed in microscopy. There were 68 samples generating two BinaxNOW® test lines thatwere microscopy positive for P.v. only. When these samples are included in the true positive calculation, BinaxNOW® test sensitivity foroverall detection of P.v. increases from 68.9% to 74.6% (886/1,187).

BinaxNOW® Malaria Test Sensitivity and Specificity for P.v. vs. Microscopy

SENSITIVITY FOR P.v.

Parasitemia Level % Sensitivity 95%CI

> 5000 93.5% (462 / 494) 91 - 96%

1000 – 5000 81.0% (277 / 342) 76 - 85%

500 – 1000 47.4% (37 / 78) 36 - 59%

100 – 500 23.6% (34 / 144) 17 – 31%

0 – 100 6.2% (8 / 129) 3 – 12%

Overall 68.9% (818 / 1187) 66 - 72%

SPECIFICITY FOR P.v.

% Specificity 95% CI

99.8% (2863 / 2870) 99–100%

Detection of P.m. and P.o. InfectionBinaxNOW® test sensitivity was 43.8% (7/16) for detection of P.m. and 50% (1/2) for detection of P.o. When five P.m. microscopypositive samples that generated two test lines in the BinaxNOW® test are included in the true positive calculation, BinaxNOW® testsensitivity for P.m. increases from 43.8% to 75.0% (12/16).

Detection of Mixed P.f./P.v. InfectionThirty-four samples were both P.f. and P.v. positive by microscopy, based on the detection of asexual forms of both species. TheBinaxNOW® test detected 32 of these samples by generating both test lines, for a sensitivity of 94.1% (95% CI of 81-98%).

5



0.2

.4.6

.8

U5 15 to 45 45 or olderFemale Male Female Male Female Male

Anemia (Hb<11) Malaria

Note: Gender differences significant at 5% level for adult anemia/Hb and for malaria

prevalence among 15-45yo. back


Frac

tion

Anemia (Hb<11)0 .2 .4 .6 .8 1

0

.05

.1

.15

.2

.25

.3

Frac

tion

Malaria0 .2 .4 .6 .8 1

0

.05

.1

.15

.2

.25

.3

Frac

tion

LF0 .2 .4 .6 .8 1

0

.05

.1

.15

.2

.25

.3

back


Household attrition between Spring 2007 and Winter 2008-09From baseline survey (May-June 2007): Dep. var.: Household not in follow-up

Constant 0.041 0.05 0.2 0.173[0.005]*** [0.013]*** [0.108]* [0.109]

Free -0.023 -0.022 -0.021[0.014] [0.014] [0.013]

Micro-loans -0.003 -0.001 0.004[0.015] [0.015] [0.015]

log(monthly exp./hhsize) 0.011 0.014[0.012] [0.011]

BISWA Debt/(Total yearly expenditure)< 0.05 -0.01 -0.021[0.016] [0.017]

BISWA Debt/(Total yearly expenditure)> 0.25 -0.006 -0.012[0.022] [0.022]

Baseline bednets per head -0.018 -0.035[0.023] [0.021]

% Slept under net last night -0.009 0.002[0.016] [0.017]

% Sleeps regularly under net 0.001 0.009[0.017] [0.017]

Household head’s age (log) -0.05 -0.053[0.019]*** [0.020]***

Household head had any schooling -0.024 -0.029[0.013]* [0.012]**

% malaria +ve in household -0.005[0.013]

% anemic (Hb< 11) in household 0.005[0.011]

Observations 1844 1844 1814 1645H0 : all coefficients = 0 (p-values) 0.11 0.21 0.14

Also included: head’s age and gender, household size, access to electricity. Back


38

Table 3: Changes in Household MembershipControl Free MF Total

Household and individual match 2,809 3,051 2,815 8,675% 82.6 81.4 81.8 81.9

New member at follow-up 175 212 221 608% 5.1 5.7 6.4 5.7

New member (visitor) at follow-up 99 121 86 306% 2.9 3.2 2.5 2.9

No longer a member at follow-up 319 363 318 1,000% 9.4 9.7 9.2 9.4

Total 3,402 3,747 3,440 10,589% 100 100 100 100

Notes: All figures are calculated for the 1,768 households re-contacted in the post-intervention survey. At standard significancelevels, we cannot reject the null hypothesis of independence between treatment and a categorical variable representing thedifferent membership status indicated along the rows of the table (p-value= 0.7157). The test is a Pearson chi-squared statisticrobust to clustering (Rao and Scott 1984).

Back to attrition & changes in household composition


39

Table 4: Changes in Household Demographic Composition

Regression Coefficients Value atConstant Free MF Baseline

(1) (2) (3) (4)

Males, U5 0.007* −0.004 −0.005 0.044(0.0035) (0.0043) (0.0049)

Females, U5 0.008*** −0.007 −0.010** 0.042(0.0029) (0.0040) (0.0046)

Males, 5 to 15 0.003 0.001 0.004 0.096(0.0031) (0.0043) (0.0047)

Females, 5 to 15 0.000 0.006 −0.002 0.089(0.0032) (0.0050) (0.0043)

Males, 15 to 45 −0.004 0.008 −0.001 0.254(0.0048) (0.0071) (0.0065)

Females, 15 to 45 0.005 0.004 0.006 0.256(0.0042) (0.0061) (0.0058)

Males, over 45 −0.011** 0.001 0.012** 0.114(0.0044) (0.0058) (0.0058)

Females, over 45 −0.007** −0.008* −0.005 0.106(0.0031) (0.0048) (0.0053)

Cross-equation joint tests Statistic p-valueFree = 0 F(8,129)= 1.4187 0.2031MF = 0 F(8,129)= 2.0016 0.0596*Free = MF = 0 F(16,121)= 1.5921 0.0904*

Notes: All figures are calculated for the 1,768 households re-contacted in the post-intervention survey. Each row reportscoefficients of a separate OLS regression estimated at the household level, where the dependent variable is the change—between baseline and follow-up—in the fraction of the household who belongs to the specified age-gender group (so thecoefficients sum up to one in columns 1 to 3). Standard errors (in brackets) and tests are robust to intra-village correlation.Asterisks indicate significance at the 10 (*), 5 (**) and 1 percent (***) level. The joint tests are robust to the presence ofcross-equation correlation of residuals.

Back to attrition & changes in household composition


Impact on hemoglobin levels: Panel households only

(1) (2) (3)Hemoglobin

Free distribution= 1 -0.036 -0.034 0.217[0.105] [0.107] [0.106]**

Micro-loans= 1 0.023 0.034 0.045[0.094] [0.097] [0.123]

Constant 11.434 11.512 0.278[0.064]*** [0.067]*** [0.075]***

Only panel individuals No Yes YesObservations 7168 6490 1873No. clusters (villages) 141 141 141R-squared 0.0002 0.0002 0.0034

Free=MF (p-value) 0.5852 0.5315 0.1650Free=MF=0 (p-value) 0.8610 0.8198 0.1099

Back Hb densities


Bednets per head: by treatment, “sample” and “BISWA” status.Data from post-intervention survey (08-09) Back

New householdsBaseline households non-BISWA BISWA hhs.

(1) (2) (3)

(A) Control (47 villages)

I

0.364 (570)

I

0.365 (371)

I

0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)

(D) New villages (25)

I

0.296 (300)

I

0.317 (75)

A1=A2=A3 (All groups equal in Control areas), p = 0.997.

D2=D3, BISWA = non-BISWA, New villages, p = 0.597.

B1=B3, Baseline=non-baseline, Free, p = 0.033.

C1=C3, Baseline=non-baseline, MF, p = 0.037.

A1=A3, Baseline=non-baseline, Control, p = 0.536.




(1) (2) (3)


I

0.364 (570)

I

0.365 (371)

I

0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)


I

0.296 (300)

I

0.317 (75)









(1) (2) (3)

(A) Control (47 villages) I0.364 (570) I0.365 (371) I0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)


I

0.296 (300)

I

0.317 (75)









(1) (2) (3)


I

0.364 (570)

I

0.365 (371)

I

0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)

(D) New villages (25) I0.296 (300) I0.317 (75)









(1) (2) (3)


I

0.364 (570)

I

0.365 (371)

I

0.366 (95)

(B) Free (47) I0.634 (612) 0.366 (391) I0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)


I

0.296 (300)

I

0.317 (75)









(1) (2) (3)


I

0.364 (570)

I

0.365 (371)

I

0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47) I0.507 (592) 0.390 (395) I0.389 (86)


I

0.296 (300)

I

0.317 (75)









(1) (2) (3)

(A) Control (47 villages) I0.364 (570)

I

0.365 (371) I0.366 (95)

(B) Free (47)

I

0.634 (612) 0.366 (391)

I

0.542 (87)

(C) MF (47)

I

0.507 (592) 0.390 (395)

I

0.389 (86)


I

0.296 (300)

I

0.317 (75)







Re-treatment rates (Details)

Data from 1st (spring 2008) and 2nd (winter 2008) re-treatment

First re-treatment Second re-treatment(six months) (12 months)

MF, “Commitment contract” (C2) -0.08 -0.09[0.030]*** [0.054]

MF, C1 -0.56 -0.62[0.059]*** [0.059]***

Intercept 0.92 0.83[0.013]*** [0.016]***

Observations 875 875R-squared 0.346 0.293Clusters 89 89

Obvious potential endogeneity of contract choice.

96% of hhds purchased only 1 type of contract (evenly split) Back


Endogenous Contract Choice?

MAY 20094 AEA PAPERS AND PROCEEDINGS

the choice of C2 is more likely when respon-dents’ preferences appear to be consistent with “hyperbolic discounting” (David Laibson 1997). Hyperbolic discounting models have been proposed to explain the finding that, in laboratory contexts, some individuals choose a reward at date t over a larger one at date t + s, but revert to choosing the later reward if the two dates are shifted by an equal time period. This form of “preference reversal” is not con-sistent with standard expected utility models. Ashraf, Karlan, and Yin (2006) find that female clients of a Philippines bank are more likely to take up a savings commitment product when their responses to hypothetical time discount-ing questions display preference reversals. We evaluate the extent of preference reversals by asking respondents to choose between smaller but earlier (actual) rewards and larger (actual) ones to be paid three months later. The earlier date is either one or four months. In Table 2, we include as regressors both a binary variable equal to one when the earlier reward is always chosen, and the number of preference reversals expressed by the respondent. The former is a measure of impatience (equal to one for a quar-ter of the sample), while the latter is a measure of “hyperbolic discounting” (with 28 percent of the sample showing at least one preference reversal). Both predict an increase in the prob-ability of choosing C2, but the coefficients are close to zero. The only regressor that predicts

a large change in the probability of choosing the commitment product is the number of ITNs owned at baseline, whose coefficient is very large (!0.312). On the one hand this is surpris-ing, because we expected buyers who already own retreated nets to value retreatment more, but on the other hand less than 10 percent of buyers owned ITNs at the time of the base-line, and the coefficient is very imprecisely estimated.

Interestingly, while we find that preference reversals only weakly predict the choice of con-tract, they do significantly decrease the fraction of nets retreated among buyers who choose the noncommitment product. When the fraction of bednets retreated is regressed on the number of preference reversals among buyers who chose C1 using OLS, the slope is negative and signifi-cant ( p-value = 0.022). The coefficient is also large in magnitude: while 40 percent of nets are retreated among buyers of C1 who display no preference reversal, the proportion is 15 percent lower per each preference reversal in the time preference questions.3 This result is consistent with a situation where some buyers are “hyper-bolic” but “naïve” in the sense that such indi-viduals do not recognize that the high discount

3 When the “unlikely retreaters” are excluded as in col-umn 2 of Table 1, the slope is even larger in absolute value (slope = !0.19, p-value = 0.002).

AQ 4

Table 2—Predictors of the Choice of Commitment (Contract 2)

Log (monthly expenditure per head) −0.028 (0.09)Number of household members aged < 5 years −0.032 (0.04)Years of schooling of most educated household member 0.007 (0.01)Any household member tested positive for malaria at baseline −0.003 (0.08)Time preference: number of preference reversals 0.018 (0.05)Time preference: always chooses earliest reward 0.055 (0.08)Risk aversion: chose no-risk lottery −0.038 (0.06)Subjective belief about protective power from retreatment −0.001 (0.01)Number of bednets per head owned at baseline 0.013 (0.15)Number of ITNs per head owned at baseline −0.312 (0.31)

Notes: Robust standard errors in parentheses allow for intravillage correlation. The regression is estimated using a linear probability model, and also includes household size, and age and gender of the household head (none of which is statistically significant). Sample size is 297; R2 = 0.06. The regression is estimated on the same sample used for column 1 of Table, with 23 observations lost because of missing values in some of the regressors. The dependent variable is a dummy equal to one if the buyer choose the commitment product (C2). For the definition of the other variables see text. Further details are avail-able upon request from the authors.

AQ 3

P20090005.indd 4 2/10/09 8:02:41 AM

No compelling evidence of endogenous contract choice

“Preference reversals” do not predict choice but predict lower retreatmentsrates (“hyperbolic but naive”?) Back


Follow-up distribution of Hb by treatment status

NP

kern

el-e

stim

ated

den

sity

Hemoglobin

Control Free MF

4 7 11 15 180

.05

.1

.15

.2

.25

Back


4. Missed blood tests: Linear Probability ModelsOmitted Category: Male adults (15-45y)

(1) (2) (3) (4) (5) (6)Absent Absent Absent Refusal Refusal Refusal

Free -0.001 -0.001 -0.009 -0.01[0.018] [0.018] [0.015] [0.015]

MF 0.006 0.005 0.018 0.017[0.018] [0.019] [0.016] [0.016]

Male, 0-5 -0.212 0.017[0.020]*** [0.013]

Female, 0-5 -0.205 0.045[0.023]*** [0.017]***

Male, 5-15 -0.121 0.017[0.018]*** [0.010]*

Female, 5-15 -0.136 0.008[0.019]*** [0.010]

Female, 15-45 -0.187 0.011[0.015]*** [0.006]*

Male, > 45 -0.133 0.003[0.017]*** [0.006]

Female, > 45 -0.212 0.036[0.018]*** [0.009]***

Constant 0.194 0.193 0.32 0.057 0.054 0.043[0.007]*** [0.013]*** [0.018]*** [0.006]*** [0.011]*** [0.012]***

Observations 9589 9589 9555 9589 9589 9555R-squared 0.0000 0.0001 0.0404 0.0000 0.0023 0.0052Clusters 141 141 141 141 141 141

Free=MF=0 0.9209 0.9343 0.2303 0.2355M=F,0-5 0.7449 0.1558M=F,5-15 0.4402 0.4505M=F,Over 45 0.0000 0.0010

Back


Usage and Mode of Acquisition

Self selection into purchase, quality signaling and “sunk-cost effects” should leadto higher usage among buyers, conditional on ownership.

Figure 13: Fraction of actually observed BISWA nets used the previous night

.667.554 .562 .487

.3.474 .471 .542

.625

00.2

.4.6

.8

1 2 3 4 5 6 7 8# BISWA nets observed by surveyor

Free MF

.193 .083 .02 -.138.3

-1-.5

0.5

1


Free - MF 95% C.I. band 95% C.I. band

19

BISWA nets seen by surveyor


Usage and Mode of Acquisition

Self selection into purchase, quality signaling and “sunk-cost effects” should leadto higher usage among buyers, conditional on ownership.Figure 13: Fraction of actually observed BISWA nets used the previous night

.667.554 .562 .487

.3.474 .471 .542

.625

00.2

.4.6

.8


Free MF

.193 .083 .02 -.138.3

-1-.5

0.5

1



19

BISWA nets seen by surveyor


Usage and Mode of Acquisition (Continued)Figure 10: Fraction of nets used the previous night

.471.527 .524

.388 .404

.245 .274

.167

.343.289

.221

.329.265 .273 .286

.438

.2.3

.4.5

.6

1 2 3 4 5 6 7 8# Nets (all nets)

Free MF

.128.238 .303

.059.139

-.028 -.012

-.271

-.4-.2

0.2

.4

1 2 3 4 5 6 7 8# Nets (all nets)


16

All nets


Figure 1: Distribution of reported Age at baseline

010

020

030

040

0Fr

eque

ncy

0 20 40 60 80 100Age at baseline (years)

Figure 2: Net usage at baseline, by gender

0.2

.4.6

.8

0 5 10 20 30 40 50 60 70Age at baseline (in years)

F - Regular use in peak season M - Regular use in peak seasonF - Used a net last night M - Used a net last nightF - Used an ITN last night M - Used an ITN last night

11


Figure 5: Usage of any net at follow-up, previous night

Age at baseline (in years)

Used a net last night (C) Used a net last night (Free) Used a net last night (MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

Figure 6: Usage of ITNs at follow-up, previous night


Used ITN last night (C) Used ITN last night (Free) Used ITN last night (MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

13


Figure 5: Usage of any net at follow-up, previous night


Used a net last night (C) Used a net last night (Free) Used a net last night (MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

Figure 6: Usage of ITNs at follow-up, previous night


Used ITN last night (C) Used ITN last night (Free) Used ITN last night (MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

13


Figure 7: Usage of any net at follow-up, peak season


Regular use (peak, C) Regular use (peak, Free) Regular use (peak, MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

Figure 8: Usage of ITNs at follow-up, peak season


Regular use (ITN, peak, C) Regular use (ITN, peak, Free) Regular use (ITN, peak, MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

14


Figure 7: Usage of any net at follow-up, peak season


Regular use (peak, C) Regular use (peak, Free) Regular use (peak, MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

Figure 8: Usage of ITNs at follow-up, peak season


Regular use (ITN, peak, C) Regular use (ITN, peak, Free) Regular use (ITN, peak, MF)

0 700

.1

.2

.3

.4

.5

.6

.7

.8

.9

1

14


Hb, malesAge

Control Free MF

U5 5 to 14 15 to 45 Over 45

9.5

10

10.5

11

11.5

12

12.5

13

13.5

Hb, femalesAge

Control Free MF

U5 5 to 14 15 to 45 Over 45

9.5

10

10.5

11

11.5

12

12.5

13

13.5

Malaria, malesAge

Control Free MF

U5 5 to 14 15 to 45 Over 45

.1

.15

.2

.25

.3

.35

Malaria, femalesAge

Control Free MF

U5 5 to 14 15 to 45 Over 45

.1

.15

.2

.25

.3

.35

Back


Correlates of ITN Purchase Decisions

Dependent variable: The household purchased at least one ITNMF villages only, data from fall 2007 (intervention), Obs. = 527, R2 = 0.11

ln(PCE) -0.100 (0.053)*ITNs per head -0.257 (0.206)Untreated nets per head -0.031 (0.135)Usually use net (% HH members) -0.052 (0.074)Used net last night (% HH members) 0.233 (0.089)**Used ITN last night (% HH members) -0.034 (0.269)U5 -0.010 (0.035)5 − 15 0.024 (0.023)15 − 60 -0.021 (0.016)> 60 0.029 (0.034)Someone +ve Malaria RDT 0.119 (0.062)*Any malaria episode last 6mts (self-reported) 0.160 (0.033)***Higher no. years of schooling -0.003 (0.008)Years of schooling of head 0.003 (0.006)Head is a woman 0.008 (0.079)P(malaria|untreated net)−P(malaria|no net) -0.009 (0.012)P(malaria|no net)−P(malaria|untreated net) -0.018 (0.011)E(malaria cost), working man (quartic root) -0.002 (0.022)E(malaria cost), working woman (quartic root) -0.003 (0.009)E(malaria cost), non working (quartic root) 0.013 (0.016)Intercept 1.229 (0.350)***


Documents

Micro-loans, bednets and malaria: Evidence from a ...dse.univr.it/ssdev/documents/material2010/tarozzi_slides.pdf · Micro-loans, bednets and malaria: Evidence from a randomized controlled