Learning, Hygiene, and Traditional Medicine Daniel Bennett University of Chicago Syed Ali Asjad Naqvi Vienna University of Economics and Business Wolf-Peter Schmidt London School of Hygiene and Tropical Medicine June 25, 2014 Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 1 / 43
66
Embed
Learning, Hygiene, and Traditional Medicine Hygiene, and Traditional Medicine ... Introduction Information and health ... Introduction Unani Medicine
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Learning, Hygiene, and Traditional Medicine
Daniel BennettUniversity of Chicago
Syed Ali Asjad NaqviVienna University of Economics and Business
Wolf-Peter SchmidtLondon School of Hygiene and Tropical Medicine
June 25, 2014
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 1 / 43
Introduction
Information and health
• Many health behaviors are costly or inconvenient.
• The lack of information may constrain behavior and health.
• Information treatments have mixed effects on sexual behavior, nutrition,malaria prevention, and sanitation.(e.g. De Walque 2007, Madajewicz et al. 2007, Dupas 2011, Luo et al. 2012)
Bayesian learning
• People learn from signals that are (1) novel and (2) credible.
• Infectious disease prevention messages rely on the germ theory of disease.
• Alternative disease models that do not involve invisible pathogens.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 2 / 43
Introduction
Information and health
• Many health behaviors are costly or inconvenient.
• The lack of information may constrain behavior and health.
• Information treatments have mixed effects on sexual behavior, nutrition,malaria prevention, and sanitation.(e.g. De Walque 2007, Madajewicz et al. 2007, Dupas 2011, Luo et al. 2012)
Bayesian learning
• People learn from signals that are (1) novel and (2) credible.
• Infectious disease prevention messages rely on the germ theory of disease.
• Alternative disease models that do not involve invisible pathogens.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 2 / 43
Introduction
Unani Medicine
• Foods, objects, and actions have “hot” or “cold” properties.
• People become sick through a hot/cold imbalance.
• Treat illness by readjusting hot/cold exposure.
Diarrhea
• Diarrhea is a leading cause of infant and child mortality.
• Unani medicine: diarrhea is a hot disease. Withhold liquids and foods.
• Modern medicine: oral rehydration therapy.
Crowd-Out
• Traditional and modern medicine may be substitutes.
• Precise priors may make people unreceptive to modern medicine.
• Showing the existence of microbes may make hygiene more plausible.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 3 / 43
Introduction
Unani Medicine
• Foods, objects, and actions have “hot” or “cold” properties.
• People become sick through a hot/cold imbalance.
• Treat illness by readjusting hot/cold exposure.
Diarrhea
• Diarrhea is a leading cause of infant and child mortality.
• Unani medicine: diarrhea is a hot disease. Withhold liquids and foods.
• Modern medicine: oral rehydration therapy.
Crowd-Out
• Traditional and modern medicine may be substitutes.
• Precise priors may make people unreceptive to modern medicine.
• Showing the existence of microbes may make hygiene more plausible.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 3 / 43
Introduction
Unani Medicine
• Foods, objects, and actions have “hot” or “cold” properties.
• People become sick through a hot/cold imbalance.
• Treat illness by readjusting hot/cold exposure.
Diarrhea
• Diarrhea is a leading cause of infant and child mortality.
• Unani medicine: diarrhea is a hot disease. Withhold liquids and foods.
• Modern medicine: oral rehydration therapy.
Crowd-Out
• Traditional and modern medicine may be substitutes.
• Precise priors may make people unreceptive to modern medicine.
• Showing the existence of microbes may make hygiene more plausible.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 3 / 43
Outline
Model
• A simple Bayesian learning model.
• Traditional medical beliefs interfere with learning.
• The convenience of hygiene mediates the impact on behavior.
Experiment
• Cluster-randomized trial: Microbe Literacy and Instruction Only treatments.
• Improvements in hygiene knowledge, behavior, and health.
The Role of Traditional Medicine
• ML reduces adherence to traditional medicine.
• Only people without traditional beliefs learn from ML.
• Only non-believers with high hygiene propensities adopt better hygiene.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 4 / 43
Model
Setup
• Health is a function of hygiene effort: h(e; ω) = ω ln(e + 1).
• People have priors over ω, the effectiveness of hygiene: ω ∼ N(µ, τ)
• Hygiene education provides a signal, ω ∼ N(µ, τ), of hygiene effectiveness.
Bayesian learning
• People update their beliefs according to Bayes Rule.
µ′ =τ
τ + τµ +
τ
τ + τµ
τ′ = τ + τ
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 5 / 43
Model
Setup
• Health is a function of hygiene effort: h(e; ω) = ω ln(e + 1).
• People have priors over ω, the effectiveness of hygiene: ω ∼ N(µ, τ)
• Hygiene education provides a signal, ω ∼ N(µ, τ), of hygiene effectiveness.
Bayesian learning
• People update their beliefs according to Bayes Rule.
µ′ =τ
τ + τµ +
τ
τ + τµ
τ′ = τ + τ
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 5 / 43
Model
The treatment effect on ω
E [ω1i −ω0
i ] = µ′ − µ
=τ(µ− µ)
τ + τ> 0 if µ > µ
The microscope demonstration
• The microscope demonstration makes the educational signal, τ, more precise.
• A more precise signal increases learning.
∂E [ω1i −ω0
i ]
∂τ=
τ(µ− µ)
(τ + τ)2> 0 if µ > µ
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 6 / 43
Model
The treatment effect on ω
E [ω1i −ω0
i ] = µ′ − µ
=τ(µ− µ)
τ + τ> 0 if µ > µ
The microscope demonstration
• The microscope demonstration makes the educational signal, τ, more precise.
• A more precise signal increases learning.
∂E [ω1i −ω0
i ]
∂τ=
τ(µ− µ)
(τ + τ)2> 0 if µ > µ
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 6 / 43
Model
Traditional medical beliefs
• Traditional medicine may increase the precision of the prior, τ.
• A more precise prior decreases learning.
∂E [ω1i −ω0
i ]
∂τ=
τ(µ− µ)
(τ + τ)2< 0 if µ > µ
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 7 / 43
Model
Hygiene behavior
• Contingent utility u(h, c) = h+ c and budget constraint: pe + c ≤ y .
• maxe µ ln(e + 1) subject to pe + c ≤ y
e∗ =µ
p− 1
The treatment effect on hygiene
E [e1i − e0i ] =τ(µ− µ)
p(τ + τ)> 0 if µ > µ
The complementarity of τ and p
∂E [e1i − e0i ]
∂τ< 0 and
∂E [e1i − e0i ]
∂p< 0 but
∂2E [e1i − e0i ]
∂τ∂p> 0
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 8 / 43
Model
Hygiene behavior
• Contingent utility u(h, c) = h+ c and budget constraint: pe + c ≤ y .
• maxe µ ln(e + 1) subject to pe + c ≤ y
e∗ =µ
p− 1
The treatment effect on hygiene
E [e1i − e0i ] =τ(µ− µ)
p(τ + τ)> 0 if µ > µ
The complementarity of τ and p
∂E [e1i − e0i ]
∂τ< 0 and
∂E [e1i − e0i ]
∂p< 0 but
∂2E [e1i − e0i ]
∂τ∂p> 0
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 8 / 43
Model
Hygiene behavior
• Contingent utility u(h, c) = h+ c and budget constraint: pe + c ≤ y .
• maxe µ ln(e + 1) subject to pe + c ≤ y
e∗ =µ
p− 1
The treatment effect on hygiene
E [e1i − e0i ] =τ(µ− µ)
p(τ + τ)> 0 if µ > µ
The complementarity of τ and p
∂E [e1i − e0i ]
∂τ< 0 and
∂E [e1i − e0i ]
∂p< 0 but
∂2E [e1i − e0i ]
∂τ∂p> 0
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 8 / 43
Setting
Southern Punjab Province
• Four districts with heavy wheat and cotton cultivation.
• 33% of women work outside the home.
Social context
• Communities are Muslim and practice purdah.
• 70% of women “seldom or never” speak with anyone outside the village.
• 83% of women have left the village 0 or 1 times in the past week.
Adult literacy classes
• Free, government-sponsored courses cover basic literacy and numeracy.
• Classes meet six times per week for 90 minutes and last for six months.
• 89% of participants lack formal schooling.
• Classes are gender segregated.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 9 / 43
Setting
Southern Punjab Province
• Four districts with heavy wheat and cotton cultivation.
• 33% of women work outside the home.
Social context
• Communities are Muslim and practice purdah.
• 70% of women “seldom or never” speak with anyone outside the village.
• 83% of women have left the village 0 or 1 times in the past week.
Adult literacy classes
• Free, government-sponsored courses cover basic literacy and numeracy.
• Classes meet six times per week for 90 minutes and last for six months.
• 89% of participants lack formal schooling.
• Classes are gender segregated.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 9 / 43
Setting
Southern Punjab Province
• Four districts with heavy wheat and cotton cultivation.
• 33% of women work outside the home.
Social context
• Communities are Muslim and practice purdah.
• 70% of women “seldom or never” speak with anyone outside the village.
• 83% of women have left the village 0 or 1 times in the past week.
Adult literacy classes
• Free, government-sponsored courses cover basic literacy and numeracy.
• Classes meet six times per week for 90 minutes and last for six months.
• 89% of participants lack formal schooling.
• Classes are gender segregated.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 9 / 43
Intervention
Microbe Literacy (ML)
• Participants view bacteria from the environment under a microscope.
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressionscontrol for strata dummies. Q1: “I can tell that my hands are clean just by looking at them”; Q2: “Untreatedwater is safe to drink”; Q3: “It is safe to eat food that has been touched by flies”; Q4: “The worst thing diarrheacan do is make a child uncomfortable.” * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 21 / 43
Table 3: The Impact of Hygiene Education on Hygiene Knowledge
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressionscontrol for strata dummies. Q1: “I can tell that my hands are clean just by looking at them”; Q2: “Untreatedwater is safe to drink”; Q3: “It is safe to eat food that has been touched by flies”; Q4: “The worst thing diarrheacan do is make a child uncomfortable.” * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 22 / 43
Table 4: The Impact on Hygiene
Depdendent variable: Hand Washing AppearanceDefecation Cooking R C
Note: standard errors, which appear in parentheses, are clustered by randomization group. Allregressions control for strata dummies. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 23 / 43
Table 4: The Impact on Hygiene
Depdendent variable: Hand Washing AppearanceDefecation Cooking R C
Note: standard errors, which appear in parentheses, are clustered by randomization group. Allregressions control for strata dummies. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 24 / 43
Table 5: The Impact of Hygiene Education on Health
Dependent variable: Diarrhea Cough FeverSample: R C R C R C
Note: standard errors, which appear in parentheses, are clustered by randomization group. Odd columns show results forthe respondent and even columns show results for her children under age 5. All regressions control for strata dummies.* p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 25 / 43
Table 5: The Impact of Hygiene Education on Health
Dependent variable: Diarrhea Cough FeverSample: R C R C R C
Note: standard errors, which appear in parentheses, are clustered by randomization group. Odd columns show results forthe respondent and even columns show results for her children under age 5. All regressions control for strata dummies.* p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 26 / 43
Traditional Medical Beliefs
Unani Medicine
• Imbalance between “hot” and “cold” elements causes disease.
• Treat diarrhea by withholding certain foods and liquids.
• Hakims provide traditional medical advice and treatment.
Traditional Belief Index (TBI)
• Main: beliefs in hot/cold, withholding liquids and foods, home remedies.
• Narrow: only beliefs in hot/cold.
• Broad: add traditional medical utilization.
Key Items
• Does eating hot or cold foods causes diarrhea?
• Is withholding liquids or breast milk an effective treatment for diarrhea?
• Have you consulted a hakim in the past three months?
• Would you see a hakim if your child was having seizures or fainting?
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 27 / 43
Traditional Medical Beliefs
Unani Medicine
• Imbalance between “hot” and “cold” elements causes disease.
• Treat diarrhea by withholding certain foods and liquids.
• Hakims provide traditional medical advice and treatment.
Traditional Belief Index (TBI)
• Main: beliefs in hot/cold, withholding liquids and foods, home remedies.
• Narrow: only beliefs in hot/cold.
• Broad: add traditional medical utilization.
Key Items
• Does eating hot or cold foods causes diarrhea?
• Is withholding liquids or breast milk an effective treatment for diarrhea?
• Have you consulted a hakim in the past three months?
• Would you see a hakim if your child was having seizures or fainting?
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 27 / 43
Traditional Medical Beliefs
Unani Medicine
• Imbalance between “hot” and “cold” elements causes disease.
• Treat diarrhea by withholding certain foods and liquids.
• Hakims provide traditional medical advice and treatment.
Traditional Belief Index (TBI)
• Main: beliefs in hot/cold, withholding liquids and foods, home remedies.
• Narrow: only beliefs in hot/cold.
• Broad: add traditional medical utilization.
Key Items
• Does eating hot or cold foods causes diarrhea?
• Is withholding liquids or breast milk an effective treatment for diarrhea?
• Have you consulted a hakim in the past three months?
• Would you see a hakim if your child was having seizures or fainting?
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 27 / 43
Traditional Medical Beliefs
0
5
10
15
20
25
30
35
0 1 2 3 4 5
Perc
en
t
Traditional Belief Index
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 28 / 43
Table 6: The Impact of Hygiene Education on Traditional Medical Beliefs
Diarrhea Causes Diarrhea TreatmentsTBI Hot Foods Cold Foods No Food No Milk(1) (2) (3) (4) (5)
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions control forstrata dummies. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 29 / 43
Table 6: The Impact of Hygiene Education on Traditional Medical Beliefs
Diarrhea Causes Diarrhea TreatmentsTBI Hot Foods Cold Foods No Food No Milk(1) (2) (3) (4) (5)
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions control forstrata dummies. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 30 / 43
0
5
10
15
20
25
30
35
40
0 1 2 3 4 5
Perc
en
t
Traditional Belief Index
ML Control Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 31 / 43
Interaction with Traditional Medical Beliefs
Hypothesis
• Traditional beliefs imply precise priors, which weaken the treatment effect.
Estimation
• Interact ML and IO with the TBI or indicators for TBIH and TBIL.
• The TBI may be correlated with mediating factors: e.g. SES, cognitive ability.
• Control for the interaction of treatment with demographic, economic,hygiene, and health variables, which may be correlated with the TBI.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 31 / 43
Interaction with Traditional Medical Beliefs
Hypothesis
• Traditional beliefs imply precise priors, which weaken the treatment effect.
Estimation
• Interact ML and IO with the TBI or indicators for TBIH and TBIL.
• The TBI may be correlated with mediating factors: e.g. SES, cognitive ability.
• Control for the interaction of treatment with demographic, economic,hygiene, and health variables, which may be correlated with the TBI.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 31 / 43
Table 7: Baseline Characteristics of Respondents by TBI
Post · ML · TBI -0.17∗∗ -0.28 -0.16∗∗ -0.16∗∗ -0.15∗
(0.079) (0.17) (0.069) (0.075) (0.076)
Panel B: By High and Low TBI
Post · ML · TBIL 0.38∗∗∗ 0.31∗∗∗ 0.42∗∗∗ 0.50∗ 0.53∗∗
(0.12) (0.090) (0.12) (0.27) (0.26)
Post · ML · TBIH 0.12 0.063 0.11 0.21 0.28(0.10) (0.16) (0.10) (0.27) (0.24)
Post · Treatment:· Demo. and Economic Controls - - - Yes -· Hygiene and Health Controls - - - - Yes
Traditional Belief Index Main Narrow Broad Main MainObservations 7516 7516 7516 7516 7516
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions include stratadummies and levels and interactions for IO, Post, and TBI. Demographic and economic controls appear in Table 1. Hygieneand health controls include the baseline appearance, diarrhea, fever, and cough for the respondent and her children under age5. In Panel B, we cut the TBI at the median of 2. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 34 / 43
Table 8: The Interaction Between Microbe Literacy and Traditional Beliefs for Hygiene Knowledge
Post · ML · TBI -0.17∗∗ -0.28 -0.16∗∗ -0.16∗∗ -0.15∗
(0.079) (0.17) (0.069) (0.075) (0.076)
Panel B: By High and Low TBI
Post · ML · TBIL 0.38∗∗∗ 0.31∗∗∗ 0.42∗∗∗ 0.50∗ 0.53∗∗
(0.12) (0.090) (0.12) (0.27) (0.26)
Post · ML · TBIH 0.12 0.063 0.11 0.21 0.28(0.10) (0.16) (0.10) (0.27) (0.24)
Post · Treatment:· Demo. and Economic Controls - - - Yes -· Hygiene and Health Controls - - - - Yes
Traditional Belief Index Main Narrow Broad Main MainObservations 7516 7516 7516 7516 7516
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions include stratadummies and levels and interactions for IO, Post, and TBI. Demographic and economic controls appear in Table 1. Hygieneand health controls include the baseline appearance, diarrhea, fever, and cough for the respondent and her children under age5. In Panel B, we cut the TBI at the median of 2. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 35 / 43
Table 8: The Interaction Between Microbe Literacy and Traditional Beliefs for Hygiene Knowledge
Post · ML · TBI -0.17∗∗ -0.28 -0.16∗∗ -0.16∗∗ -0.15∗
(0.079) (0.17) (0.069) (0.075) (0.076)
Panel B: By High and Low TBI
Post · ML · TBIL 0.38∗∗∗ 0.31∗∗∗ 0.42∗∗∗ 0.50∗ 0.53∗∗
(0.12) (0.090) (0.12) (0.27) (0.26)
Post · ML · TBIH 0.12 0.063 0.11 0.21 0.28(0.10) (0.16) (0.10) (0.27) (0.24)
Post · Treatment:· Demo. and Economic Controls - - - Yes -· Hygiene and Health Controls - - - - Yes
Traditional Belief Index Main Narrow Broad Main MainObservations 7516 7516 7516 7516 7516
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions include stratadummies and levels and interactions for IO, Post, and TBI. Demographic and economic controls appear in Table 1. Hygieneand health controls include the baseline appearance, diarrhea, fever, and cough for the respondent and her children under age5. In Panel B, we cut the TBI at the median of 2. * p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 36 / 43
Traditional Beliefs and Learning
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
≤1 2 3 ≥4
Co
eff
icie
nt
Traditional Belief Index
Absolute Effect of ML Differential Effect of ML
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 37 / 43
The Interaction with Hygiene Propensities
Predictions
• The treatment effect on behavior and health should be the strongest forpeople with low values of both τ and p.
Estimation
• Estimate a hygiene propensity score with baseline data.
• Isolate respondents with high hygiene propensities.
• Regressors: house type, roof type, food insecurity, age, religious adherence,latrine characteristics, water source characteristics, presence of animals.
• Dummy variables for all categorical responses.
• R2 ≈ 0.10.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 38 / 43
The Interaction with Hygiene Propensities
Predictions
• The treatment effect on behavior and health should be the strongest forpeople with low values of both τ and p.
Estimation
• Estimate a hygiene propensity score with baseline data.
• Isolate respondents with high hygiene propensities.
• Regressors: house type, roof type, food insecurity, age, religious adherence,latrine characteristics, water source characteristics, presence of animals.
• Dummy variables for all categorical responses.
• R2 ≈ 0.10.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 38 / 43
Hygiene and Hygiene Propensity
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 39 / 43
Table 9: The Interaction Between Microbe Literacy and Traditional Beliefs for Hygiene and Health
Respondent ChildrenAppearance Health Appearance Health
ML · TBIL − ML · TBIH (p-value) 0.02 0.01 0.10 0.11 0.30 0.35 0.77 0.75
Sample Full HP Full HP Full HP Full HPObservations 3836 2008 3836 2008 2823 2640 5696 5278
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions include strata dummies and levels and interactionsof ML, IO, and TBI. In Panel B, we cut the TBI at the median of 2. Even columns show results for respondents with above-median hygiene propensities. *p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 40 / 43
Table 9: The Interaction Between Microbe Literacy and Traditional Beliefs for Hygiene and Health
Respondent ChildrenAppearance Health Appearance Health
ML · TBIL − ML · TBIH (p-value) 0.02 0.01 0.10 0.11 0.30 0.35 0.77 0.75
Sample Full HP Full HP Full HP Full HPObservations 3836 2008 3836 2008 2823 2640 5696 5278
Note: standard errors, which appear in parentheses, are clustered by randomization group. All regressions include strata dummies and levels and interactionsof ML, IO, and TBI. In Panel B, we cut the TBI at the median of 2. Even columns show results for respondents with above-median hygiene propensities. *p < 0.1, ** p < 0.05, *** p < 0.01.
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 41 / 43
Hygiene Behavior by TBI and Propensity Score
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
≤1 2 3 ≥4
Mic
rob
e L
itera
cy
Co
eff
icie
nt
Traditional Belief Index
High Hygiene Propensity Low Hygiene Propensity
Bennett, Naqvi, and Schmidt (UC, WU, LSHTM) Learning, Hygiene, and Traditional Medicine June 5, 2014 42 / 43
Conclusion
Impact evaluation
• The microscope demonstration improves learning, hygiene, and health.
• Stronger results for the respondent than for her children.