Contents: Appendix 3: Parallel Trends. Appendix

Transcription

1 Mohanan M, Babiarz KS, Goldhaber-Fiebert JD, Miller G, Vera-Hernandez M. Effect of a large-scale social franchising and telemedicine program on childhood diarrhea and pneumonia outcomes in India. Health Aff (Millwood). 2016;35(10). Appendix Contents: Appendix Exhibit 1: Study Cluster Locations in Bihar, India Exhibit Table A.1: Summary Statistics from Baseline and Follow Up Appendix 1: Robustness Checks Appendix 2: Sensitivity to non-normality in asymptotic distribution of difference-indifferences estimator Appendix 3: Parallel Trends

2 Appendix Exhibit 1: Study Cluster Loca:ons in Bihar, India Notes: The Bihar Evalua2on of Social Franchising and Telemedicine (BEST) study was conducted in 11 districts (Begusarai, Gopalganj, Khagaria, Muzaffarpur, Nalanda, Paschim Champaran, Patna, Purba Champaran, Saharsa, Samas2pur, Vaishali).

3 Characteristics of the household Appendix Exhibit A1.1: Summary Statistics Baseline Overall Control Endline Implement ation Overall Control Implement ation Proportion Categorized as Scheduled Cast, Scheduled Tribe, Other Backward Class 85.9% 86.3% 85.3% 88.9% 89.0% 88.9% Number of Children Under Proportion Hindu 84.6% 84.8% 84.4% 83.5% 82.5% 84.3% Proportion BPL 65.8% 66.4% 65.0% 68.0% 67.9% 68.2% Proportion Falling into Wealth Quantile % 18.6% 21.2% 20.2% 21.6% 19.2% Proportion Falling into Wealth Quantile % 20.3% 19.2% 20.2% 20.1% 20.2% Proportion Falling into Wealth Quantile % 19.8% 20.1% 20.1% 20.5% 19.9% Proportion Falling into Wealth Quantile % 20.9% 19.0% 20.1% 19.3% 20.6% Proportion Falling into Wealth Quantile % 20.3% 20.5% 19.4% 18.5% 20.0% Characteristics of the mother Mother's Age Proportion of Mothers Literate 29.2% 29.7% 28.6% 49.6% 48.3% 50.7% Charactersitics of the child Proportion Male 53.0% 53.0% 53.0% 52.1% 51.5% 52.6% Proportion Age % 14.5% 14.9% 17.7% 17.8% 17.6% Proportion Age % 17.3% 17.0% 18.5% 18.7% 18.3% Proportion Age % 17.7% 17.7% 21.3% 21.4% 21.3% Proportion Age % 21.5% 21.4% 21.0% 20.9% 21.0% Proportion Age % 22.6% 22.1% 20.2% 19.9% 20.5% Diarrheal Illness: Prevalence, Treatment and Expenditure Diarrhea Prevalence 21.4% 20.8% 22.2% 10.9% 9.6% 12.6% Probability of Seeking Care 69.4% 70.0% 68.6% 67.2% 66.5% 68.0% Probability of Zinc Treatment (Any Source) Illness 34.1% 33.8% 34.4% 44.1% 43.7% 44.5% Probability of Zinc Treatment (Self Treatment) Illness 10.0% 9.2% 11.1% 16.1% 14.9% 17.2% Probability of Zinc Treatment (Provider) Illness 25.2% 25.6% 24.6% 30.2% 30.6% 29.8% Probability of Zinc Treatment (Provider) Illness & Seeking Care 36.3% 36.6% 35.8% 44.9% 46.0% 43.8% Probability of ORS Treatment (Any Source) Illness 1.5% 1.6% 1.4% 11.0% 11.5% 10.5% Probability of ORS Treatment (Self Treatment) Illness 0.1% 0.1% 0.1% 1.3% 1.4% 1.3% Probability of ORS Treatment (Provider) Illness 1.4% 1.5% 1.3% 9.5% 10.0% 9.0% Probability of ORS Treatment (Provider) Illness & Seeking Care 2.0% 2.1% 2.0% 14.1% 15.0% 13.2% Probability of Zinc + ORS Treatment (Any Source) Illness 1.4% 1.5% 1.3% 7.0% 7.2% 6.7% Probability of Zinc + ORS Treatment (Self Treatment) Illness 0.0% 0.0% 0.0% 0.7% 0.7% 0.7% Probabiity of Zinc + ORS Treatment (Provider) Illness 1.4% 1.5% 1.2% 5.6% 5.9% 5.2% Probabiity of Zinc + ORS Treatment (Provider) Illness & Seeking Care 2.0% 2.1% 1.8% 8.3% 8.9% 7.7% Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care Pneumonia Illness: Prevalence, Treatment and Expenditure Pneumonia Prevalence 6.3% 6.0% 6.6% 3.4% 3.1% 3.7% Probability of Seeking Care Illness 84.5% 85.7% 83.2% 88.0% 89.6% 86.3% Probability of Antibiotic Treatment (Any Source) Illness 34.6% 35.4% 33.7% 69.4% 73.7% 64.4% Probability of Antibiotic Treatment (Self Treatment) Illness 10.1% 10.1% 10.1% 24.7% 25.7% 23.5% Probability of Antibiotic Treatment (Provider) Illness 3.4% 4.1% 2.6% 6.9% 7.3% 6.5% Probability of Antibiotic Treatment (Provider) Illness & Seeking Care 0.4% 0.4% 0.3% 1.1% 0.9% 1.3% Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness 31.6% 31.7% 31.4% 61.9% 66.1% 57.1% Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness 9.7% 9.7% 9.8% 23.6% 24.8% 22.3% Probability of Full 5 Day Antibiotic Treatment (Provider) Illness 37.3% 37.0% 37.8% 70.3% 73.7% 66.2% Probability of Full 5 Day Antibiotic Treatment (Provider) Illness & Seeking Care 11.5% 11.3% 11.8% 26.9% 27.7% 25.8% Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care

4 Appendix 1: Sensitivity to Alternate Model Specifications Appendix Exhibits A1.2-A1.4 report the robustness of our findings to a range of regression specifications. We explore a variety of outcomes in addition to primary outcomes reported in the main text. These include disease prevalence, the probability of seeking care, the probability of being treated with zinc, oral rehydration solution (ORS), or both zinc and ORS from any source, through selftreatment, from a provider or from a provider (conditional on seeking care from a provider). Because pneumonia identification is complex, we use two methods for identifying suspected pneumonia. The first is symptom based, with suspected pneumonia defined as the presence of fever, cough, and difficulty breathing in the preceding 15 days. The second employs a video diagnosis system developed by the Child Health Epidemiology Reference Group (CHERG) (results shown in bottom panels of tables). If children experienced either cough or fever within the preceding 15 days, parents were shown a video depicting a child with confirmed pneumonia. Parents were asked whether their child s illness was similar to the child in the video and if so, the child was coded as a suspected case of pneumonia. We also explore robustness to the inclusion of a range of control variables including district-year fixed effects only, district-year fixed effects and child characteristics (sex and single year of age indicators), districtyear fixed effects, child characteristics, and maternal characteristics (literacy and age), district-year fixed effects, child characteristics, maternal characteristics, and household characteristics (number of children under 5, religion, caste category, BPL status and wealth quintile). All combinations of these specifications are also estimated using three approaches to measuring program implementation. Because some providers moved practice locations or did not operate Sky program activity after initial recruitment, our preferred metric for determining implementation status and intensity for each study cluster is to rely on the validated enumerated census of active providers. These results are shown in Appendix Exhibit A1.2 (with treatment defined as presence of at least one Sky provider in study cluster) and Appendix Exhibit A1.3 (with treatment defined as the count of Sky providers identified in the study cluster). However, we replicate all analyses using the locations of recruited Sky providers as recorded in WHP franchise rosters, with treatment defined as the recruitment of at least one Sky provider in the study cluster (Appendix Exhibit A1.4). Finally, because our primary analysis uses linear probability models for discrete outcomes (allowing for consistent fixed effects estimation while avoiding concerns about incidental parameters) we also assess sensitivity to functional form assumptions, Appendix Exhibit A1.5 reports results obtained using non-linear probit models fit using maximum likelihood estimation (MLE), yielding comparable findings.

5 Appendix Exhibit A1.2: BEST Primary Outcomes Results: Treatment = Any SkyHealth Provider Listed in Cluster Fixed Effects Only Fixed Effects + Fixed Effects + Fixed Effects + Sample Size Child Characteristics Child Characteristics + Child Characteristics + Mother Characteristics Mother Characteristics + Household Characteristics Diarrhea Outcomes Diarrhea Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Pneumonia Outcomes: Conventional Symptom-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness ** ** ** ** 3,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness ** ** ** ** 3,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care ** ** ** ** 2,733 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care * 2,733 [ ] [ ] [ ] [ ] Pneumonia Outcomes: CHERG-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness * 3,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care * * * ** 2,614 [ ] [ ] [ ] [ ] Note: Each cell shows the estimated effect of World Health Partners' Sky program on primary and secondary outcomes. Program effects are estimated OLS linear regression of each outcome on indicators for program implementation, survey wave, and the interaction of survey wave and program implementaiton, along with varying groupss of control variables (as noted in column headers). All specifications include and district-year fixed effects. Huber-White robust standard errors are clustered at the study cluster level. * p<0.10, ** p<0.05, *** p<0.001.

6 Appendix Exhibit A1.3: BEST Primary Outcomes Results: Treatment = Number of SkyHealth Providers Listed Per ClusterProviders Listed Per Cluster Fixed Effects Only Fixed Effects + Fixed Effects + Fixed Effects + Sample Size Child Characteristics Child Characteristics + Child Characteristics + Mother Characteristics Mother Characteristics + Household Characteristics Diarrhea Outcomes Diarrhea Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Pneumonia Outcomes: Conventional Symptom-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness ** ** ** ** 3,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness * * * * 3,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care * * * * 2,733 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] Pneumonia Outcomes: CHERG-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Note: Each cell shows the estimated effect of World Health Partners' Sky program on primary and secondary outcomes. Program effects are estimated OLS linear regression of each outcome on indicators for program implementation, survey wave, and the interaction of survey wave and program implementaiton, along with varying groupss of control variables (as noted in column headers). All specifications include and district-year fixed effects. Huber-White robust standard errors are clustered at the study cluster level. * p<0.10, ** p<0.05, *** p<0.001.

7 Appendix Exhibit A1.4: BEST Primary Outcomes Results: Treatment = SkyHealth Provider Contracted to Work in Cluster per WHP Roster Records Fixed Effects Only Fixed Effects + Fixed Effects + Fixed Effects + Sample Size Child Characteristics Child Characteristics + Child Characteristics + Mother Characteristics Mother Characteristics + Household Characteristics Diarrhea Outcomes Diarrhea Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probability of ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness ,764 [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] Pneumonia Outcomes: Conventional Symptom-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness * * * ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness * ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness ,196 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care * * * * 2,733 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Probability) Illness ,196 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Probability) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] Pneumonia Outcomes: CHERG-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Antibiotic Treatment (Provider) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Provider) Illness ,014 [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Probability) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Out of Pocket Expenditure (Treatment and Medicine) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] Note: Each cell shows the estimated effect of World Health Partners' Sky program on primary and secondary outcomes. Program effects are estimated OLS linear regression of each outcome on indicators for program implementation, survey wave, and the interaction of survey wave and program implementaiton, along with varying groupss of control variables (as noted in column headers). All specifications include and district-year fixed effects. Huber-White robust standard errors are clustered at the study cluster level. * p<0.10, ** p<0.05, *** p<0.001.

8 Appendix Exhibit A1.5: BEST Primary Outcomes Results: OLS and Equivalent Probit Specifications District-Year Level Fixed Effects Treatment = Sky Provider Identified in Cluster During Provider Census Treatment = Count of Sky Providers Identified in Cluster During Provider Census Treatment = Sky Provider Contracted to Work in Cluster per WHP Roster Records Sample Size OLS Probit OLS Probit OLS Probit Diarrhea Outcomes Diarrhea Prevalence ** ,246 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Seeking Care ,764 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Zinc Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Any Source) Illness ,764 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Zinc + ORS Treatment (Self Treatment) Illness ,764 [ ] [ ] [ ] [ ] [ ] [ ] Probabiity of Zinc + ORS Treatment (Provider) Illness & Seeking Care ,391 [ ] [ ] [ ] [ ] [ ] [ ] Pneumonia Outcomes: Conventional Symptom-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Seeking Care Illness * * * ,196 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,196 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness * ,196 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Probability) Illness & Seeking Care ,733 [ ] [ ] [ ] [ ] [ ] [ ] Pneumonia Outcomes: CHERG-Based Diagnosis Pneumonia Prevalence ,246 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Seeking Care Illness ,014 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Any Source) Illness ,014 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Self Treatment) Illness ,014 [ ] [ ] [ ] [ ] [ ] [ ] Probability of Full 5 Day Antibiotic Treatment (Probability) Illness & Seeking Care ,614 [ ] [ ] [ ] [ ] [ ] [ ] Note: Each cell shows the estimated effect of World Health Partners' Sky program on primary and secondary outcomes. Program effects are estimated OLS linear regression and probit regressions of each outcome on indicators for program implementation, survey wave, and the interaction of survey wave and program implementaiton, along with controls for child, mother and household characteristics and district-year fixed effects. Huber-White robust standard errors are clustered at the study cluster level. * p<0.10, ** p<0.05, *** p<0.001.

9 Appendix 2: Sensitivity to non-normality in the asymptotic distribution of the difference-in-differences estimator. Serial correlation might lead to incorrect inference in the difference-in-differences estimator (Bertrand, Duflo et al. 2004). To check that this is not a problem in our analysis, we estimate the Sky program effects on primary outcomes under each of 1000 randomly drawn synthetic implementation-control allocation scenarios, using implementation-control proportions consistent with both our census and WHP franchise rosters. Figures 1-10 in this Appendix plot the distribution of these simulated program effects, showing distribution of possible treatment effects.

10 Appendix Figure 1: Diarrhea Prevalence Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

11 Appendix Figure 2: Probability of Seeking Care for Diarrhea Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

12 Appendix Figure 3: Probability of Zinc Treatment Through Self Treatment Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

13 Appendix Figure 4: Probability of Zinc Treatment by Provider (Conditional on Seeking Care) Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

14 Appendix Figure 5: Probability of Receiving both Zinc Therapy and Oral Rehydration Solution Through Self Treatment Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

15 Appendix Figure 6: Probability of Receiving both Zinc Therapy and Oral Rehydration Solution from Provider (Conditional on Seeking Care) Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

16 Appendix Figure 7: Pneumonia Prevalence Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

17 Appendix Figure 8: Probability of Seeking Care for Pneumonia Symptoms Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

18 Appendix Figure 9: Probability of Receiving Full 5 Day Course of Antibiotics Through Self Treatment Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

19 Appendix Figure 10: Probability of Receiving Full 5 Day Course of Antibiotics From Provider (Conditional on Seeking Care) Panel A: Implementation and Control Proportion According to Provider Census Panel B: Implementation and Control Proportion According to WHP Provider Roster Records observed in the provider census. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the reported in the WHP Sky provider roster. Blue line corresponds the effect size measured using the implementation-control allocation observed and confirmed in the provider census. Red line corresponds to the effect size measured using the

20 References Bertrand, M., E. Duflo and S. Mullainathan (2004). "How Much Should We Trust Differences-In- Differences Estimates?" The Quarterly Journal of Economics 119(1):

21 Appendix 3: Checking Parallel Trends Assumptions for DD Analysis To check for preexisting trends correlated with SkyHealth program implementation, we use birth history data collected in Bihar by the District Level Household Survey, waves II and III. The data was collected in and 2008 and are representative of local populations at the district level. They have been widely used for a variety of studies of maternal and child health and are deemed to be of high quality. Of particular use to us, mothers are asked to provide details about each child she has delivered in the last 4 years, the outcome of the birth, and the age of death if the child did not survive. From these data we are able to construct a child-level cohort data set including indicators the year of a child s birth, whether or not a child survived past 28 days of age, whether a child survived past 12 months of age, and a variety of maternal characteristics likely to affect survival. Pooling cohorts across these two survey waves, the combined dataset includes information on 15,391 children born between 2000 and 2008 in our 11 study districts in Bihar. We then combine these cohort child health data with information on SkyHealth implementation collected through our field team s provider census. Because DLHS does not provide geographic identifiers below the district level, we create measures of districtlevel SkyHealth program intensity (which would not be adequate for computing the impact of the program due to the small number of districts, but it is useful here as a illustration of the parallel trend assumption). For each study district, we specifically construct two different measures of program intensity: the total number of Sky providers operating within each district, and the share of private providers recruited into the Sky program in each district. We assess the degree to which non-random WHP implementation is correlated with preexisting child mortality trends. Specifically, we estimate the relationship between district-level child mortality trends and future Sky program intensity using following basic estimating equation: y!!!" = α! + β! Intensity! Birth Year! + α! X!!!" + α! Intensity! + γ! + θ! + ε!!!" where y!!!" refer to the outcome variable of interest for child i born in district d in year t,, Intensity! is a continuous measure of future intensity of Sky program implementation in district d (as defined above), X!"# are child i and household h observable characteristics (mother s age at deliver, mother s age at marriage, mother s education, household caste, and household religion), γ! are district-level fixed effects that control for time invariant characteristics of each district, and θ! are time period fixed effects that control for unobservable shocks that affect all babies born in each year. The parameters of interest, which estimates the extent to which future Sky intensity is correlated with preexisting trends are the vector β!. After accounting for the multiple comparisons that we conduct in estimating these equations, failing to distinguish the β! estimates from zero would be consistent with the assumption of parallel trends (or WHP program implementation not being correlated with pre-existing trend differences in child health).

22 Appendix Figures A3.1 and A3.2 the vector β! estimated for each of two child mortality outcomes (probability of neonatal death and probability of infant death) and each of our two intensity measures. Overall we do not find systematic correlations between intensity of Sky program implementation and pre-existing trends in infant and neonatal mortality.

23 Appendix Exhibit A3.1: Parallel pre-existing trends in child mortality associated with the future total number of Sky providers

24 Appendix Exhibit A3.2: Parallel pre-existing trends in child mortality associated with the share of private providers recruited into the Sky program in each district