preliminary findings from a phe operations research activity in madagascar yung-ting kung, mph...
TRANSCRIPT
Preliminary Findings from a PHE
Operations Research Activity in
Madagascar
Yung-Ting Kung, MPH
October 5, 2007
Pho
tos:
Eck
hard
Kle
inau
Household Food Security and Livelihood Concept
Rice cultivation
Anti-erosion measures
Fruit Trees
Plant nurseries
Reforestation
Off season planting
Forage Crops
Fish culture
Beekeeping
Animal husbandry
Vaccination, Diarrheal Disease, Malaria
Nutrition Safe
motherhoodReproductive health Potabl
e water
Sanitation & Hygiene
Sustainable use of natural resources
Protected ecosystems
Market garden
Income generationMicrocreditsCivil society organizations
Pho
to: E
ckha
rd K
lein
au, 2
002
Community Centered P-H-E Interventions
Watershed
(Kleinau et al. presentation, May 2005)
Strategies for Working with Local Communities
Champion Community
Farmer-to- Farmer
Child-to- Community
Pho
to: E
ckha
rd K
lein
au, 2
001
Pho
to: E
ckha
rd K
lein
au, 2
001
Pho
to: R
ober
t Eng
elm
an, P
AI,
2003
(Kleinau et al. presentation, May 2005)
Partners in Sustainable Development
9 NGOs 160 Communities
8 Projects, 3 Donors3 Foundations
Government in 19 Communes6 Ministries P
hoto
s: 2
001
Eck
hard
Kle
inau
(Kleinau et al. presentation, May 2005)
Madagascar Environmental Health Project: Two Operations Research Questions
Pho
to: M
iche
le D
urye
a, P
AI,
2003
2. Does the sustainability and effectiveness of Integrated PHE depend on the organizational arrangement?
1. Is integrated PHE more effective and sustainable than unlinked, single-sector approaches?(synergies hypothesis)
Pho
to: R
ober
t Eng
elm
an, P
AI,
2003
Pho
to: R
ober
t Eng
elm
an, P
AI,
2003
(Kleinau et al. presentation, May 2005)
Program Evaluation Design Quasi-experimental design
Two groups: integration and non-integration Two household surveys: baseline 2001 and impact 2004
Integration sites: 3 organizational arrangements (4 NGOs in 3 regions) (56 communities) Type 1: Integration within 1 NGO (multidisciplinary teams) Type 2: Integration within 1 NGO (separate H & E teams) Type 3a: Integration between several NGOs Type 3b: Integration between several NGOs
Control sites: (29 communities) Single sector health or environment activities or neither
(Kleinau et al. presentation, May 2005)
Sampling All communities self-selected to participate
(Kleinau et al, 2005)
Questionnaires Five different questionnaires
Village Household Head of household Women 15-49 Children under 5
Questionnaires developed in Malagasy and pre-tested
Pho
to: c
khar
d K
lein
au, 2
001
Pho
to: R
ober
t Eng
elm
an, P
AI,
2003
Earlier Univariable Analysis
(Kleinau et al. presentation)
Dependent Variables Six population, health, environment indicators
Modern contraceptive prevalence rate Household head’s perception of food security for the
entire year Childhood (under 5 years) moderate and severe stunting Household head knowledge of soil degradation as an
effect of slash-and-burn agriculture Whether the household head has planted eucalyptus
trees Women’s community participation in village associations
All indicators are dichotomous
Difference-in-Differences (DiD) Methodology DiD estimator is defined as the difference in the mean
outcome of the treatment group after v. before treatment and the mean outcome of the control group after v. before the treatment period (Ravallion, 2005)
Both group specific and time-specific effects are controlled
for (Wooldridge, 2002)
Therefore unobservable differences between the two groups which are time invariant are controlled for
For unbiased DiD estimator, treatment cannot be related to other variables which influence the outcome (Wooldridge 2002)
Since treatment not randomized, it is essential to control for initial heterogeneity between the two groups (Ravallion 2005)
Dprobit Regression Model Used to measure the DiD
estimator since all outcomes are dichotomous
Reports the marginal effect, that is change in probability of an outcome For infinitesimal change in a
continuous variable Or for discrete change in
categorical variable Standard errors adjusted for
clustering on ID
Pho
to: E
ckha
rd K
lein
au, 2
001
Baseline Characteristics: Women’s questionnaireTable 1: Women's Questionnaire relevant baseline descriptive statistics
Integration Control p-valueVariable (%/mean) (%/mean) (chi2/t-test)Dependent Variable
Modern contraceptive prevalence rate 11.67 2.39 0.000Women's community participation 29.48 30.99 0.670
Independent VariablesWomen's age 29.47 30.69 0.396Women currently married 58.81 67.52 0.022Women's education level 0.720
none 6.37 7.48primary - cannot read or write 10.20 7.81primary - can read only 59.13 68.51primary - can read and write 24.30 16.21secondary/superior
Previous live births 80.72 86.10 0.0739
Baseline Characteristics:Children’s QuestionnaireTable 2: Child's questionnaire and relevant baseline descriptive statistics
Integration Control p-valueVariable (%/mean) (%/mean) (chi2/t-test)Dependent Variable
Moderate/severe childhood stunting 50.86 44.83 0.1431
Independent VariablesChild's age 1.84 2.03 0.0746Responsible's age 30.66 32.42 0.1474Child's sex 0.6427
Girl 48.56 46.75Boy 51.44 53.25
Number of children in the household 0.50431 37.73 35.062 42.71 46.753 12.83 12.994 5.63 5.195 1.10 0.00
Responsible's education 0.0279none 27.28 37.23primary - cannot read or write 10.77 6.49primary - can read only 6.40 5.19primary - can read and write 40.84 40.26secondary/superior 14.71 10.82
Responsible received health/nutrition information 69.51 61.04 0.0203Health center close by 98.02 100 0.0520Wealth Index Terciles 0.0004
First 27.61 35.40Second 31.80 38.94Third 40.59 25.66
Baseline Characteristics:Head of Household Questionnaire
Table 3: Head of Household questionnaire and relevant baseline descriptive statistics
Integration Control p-valueVariable (%/mean) (%/mean) (chi2/t-test)Dependent Variable
Food security 15.60 15.13 0.8647Eucalyptus tree planting 58.22 40.00 0.0000Knowledge of soil degradation as effect of slash-and-burn 61.56 46.61 0.0001
Independent VariablesAge of head of household 43.49 45.71 0.0382Marital status (married=1) 75.16 77.96 0.3877Education level 0.0010
none 22.29 34.51primary - cannot read or write 5.33 6.67primary - can read only 5.82 4.31primary - can read and write 46.15 39.22secondary/superior 20.41 15.29
Sex of household head (female=1) 19.12 15.51 0.2155Principal activity of head of household 0.0000
Agriculture 81.79 83.27Animal Husbandry/Fishing 1.65 8.98Bee-Keeping 0.13 0.41Commerce 2.18 0.82Salaried 7.95 4.08Artisan 2.82 1.22Forest Exploitation 2.30 0.00arboriculture 0.43 0.00Gardening 0.13 0.00Other 1.22 0.62
Cultivable space 97.27 98.37 0.3376
Dprobit Model: Modern Contraceptive Prevalence Rate
Table 5a: dprobit model for modern methods CPR (2363 Observations)Variable dF/dx p-value
Year 0.0910 0.014Integration site 0.1209 0.000Program effect (Integration*year) -0.0482 0.229
Adjusted for sampling weights and clustering on id
Table 5b: dprobit model for modern methods CPR (1784 Observations)Variable dF/dx p-value
Year 0.0877 0.060Integration site 0.0897 0.030Program effect (Integration*year) -0.0447 0.373Education 0.0520 0.000Previous live births (yes=1) 0.1177 0.000Marital status (not in union=1) -0.0325 0.083Haute Matsiatra region* -0.0548 0.005Anosy Region* 0.0030 0.907Anosibe An'Ala Fivondronana** 0.0848 0.005Ambalvao Fivondronana** 0.0341 0.436Amboasary Fivondronana** -0.0977 0.008
Adjusted for covariates, sampling weights, and clustering on id*Comparison to Mangoro region**Comparison to Moramanga Fivondronana
Dprobit Model: Women’s Community Participation
Table 6a: dprobit model for Women's Participation (2363 Observations)Variable dF/dx p-value
Year -0.0546 0.201Integration site -0.0152 0.671Program effect (Integration*year) 0.0930 0.057
Adjusted for sampling weights and clustering on id
Table 6b: dprobit model for Women's Participation (1784 Observations)Variable dF/dx p-value
Year -0.0366 0.503Integration site -0.0006 0.990Program effect (Integration*year) 0.0573 0.350Age 0.0065 0.000Education 0.0259 0.083Previous live births (yes=1) 0.1223 0.001Marital status (Not in union=1) -0.0107 0.712Haute Matsiatra region* 0.1140 0.001Anosy Region* 0.0688 0.110Anosibe An'Ala Fivondronana** 0.2997 0.000Ambalvao Fivondronana** 0.0515 0.311Amboasary Fivondronana** 0.0829 0.193
Adjusted for covariates, sampling weights, and clustering on id*Comparison to Mangoro region**Comparison to Moramanga Fivondronana
Dprobit Model:Moderate/Severe Childhood Stunting
Table 7a: dprobit model for moderate/severe childhood stunting (2057 Observations)Variable dF/dx p-value
Year 0.0907 0.047Integration site 0.0604 0.143Program effect (Integration*year) -0.1150 0.031
Adjusted for sampling weights and clustering on id
Table 7b: dprobit model for moderate/severe childhood stunting (1616 Observations)Variable dF/dx p-value
Year 0.0949 0.090Integration site 0.0697 0.148Program effect (Integration*year) -0.1351 0.024Child's age 0.0845 0.000Child's sex (female=1) -0.0365 0.019# children in household -0.0004 0.983Responsible's education -0.0045 0.678Responsible's age -0.0015 0.421Received info on health/nutrition (yes=1) 0.0317 0.300Health center in village -0.0163 0.730Grandmother responsible* 0.1847 0.018Aunt responsible* 0.1674 0.507Father responsible* 0.0600 0.563Other responsible* 0.3510 0.017Wealth tercile -0.0670 0.000Haute Matsiatra region** -0.0837 0.056Anosy Region** -0.2802 0.000Anosibe An'Ala Fivondronana*** -0.1460 0.009Ambalvao Fivondronana*** 0.0565 0.343Tolagnaro Fivondronana*** 0.1306 0.021
Adjusted for covariates, sampling weights, and clustering on id*Comparison to mother responsible** Comparison to Mangoro region*** Comparison to Moramanga Fivondronana
Dprobit Model: Food Security (Reported by Head of HH)
Table 8a: dprobit model for food security (2073 Observations)Variable dF/dx p-value
Year 0.1239 0.000Integration site 0.0057 0.864Program effect (Integration*year) -0.0491 0.234
Adjusted for sampling weights and clustering on id
Table 8b: dprobit model for food security (2069 Observations)Variable dF/dx p-value
Year 0.1073 0.003Integration site 0.0013 0.971Program effect (Integration*year) -0.0440 0.297Age of household head -0.0014 0.056Marital status of household head (Not in union=1) 0.0603 0.096Education of household head 0.0038 0.600Sex of household head (Female=1) -0.0443 0.232Haute Matsiatra region* 0.1853 0.000Anosy Region* 0.0199 0.521Anosibe An'Ala Fivondronana** 0.1378 0.001Ambalvao Fivondronana** -0.1071 0.004Amboasary Fivondronana** 0.0181 0.641Principal activity of household head***
animal husbandry/fishing -0.0336 0.676Bee-keeping 0.2209 0.312Commerce 0.0041 0.963Salaried -0.0358 0.448Artisan -0.1396 0.047Forest exploitation -0.0272 0.770Other 0.0639 0.496
Adjusted for covariates, sampling weights, and clustering on id*Comparison to Mangoro region**Comparison to Moramanga Fivondronana***Comparison to agriculture as principal activity
Dprobit Model: Eucalyptus Tree Planting
Table 9a: dprobit model for eucalyptus tree planting (1872 Observations)Variable dF/dx p-value
Year 0.1631 0.000Integration site 0.1805 0.000Program effect (Integration*year) -0.0432 0.416
Adjusted for sampling weights and clustering on id
Table 9b: dprobit model for eucalyptus tree planting (1865 Observations)Variable dF/dx p-value
Year 0.1164 0.040Integration site 0.0509 0.235Program effect (Integration*year) 0.0056 0.930Age of household head 0.0040 0.000Marital status of household head (Not in union=1) -0.0579 0.246Education of household head 0.0388 0.000Sex of household head (Female=1) -0.2290 0.000Cultivable land (yes=1) 0.1286 0.229Haute Matsiatra region* -0.1943 0.001Anosy Region* -0.1264 0.003Anosibe An'Ala Fivondronana** 0.0636 0.239Ikongo Fivondronana** -0.2600 0.000Amboasary Fivondronana** -0.3096 0.000Principal activity of household head***
animal husbandry/fishing -0.0957 0.395Commerce 0.0061 0.956Salaried 0.0374 0.559Artisan -0.1177 0.222Forest exploitation -0.0776 0.471Other -0.0238 0.229
Adjusted for covariates, sampling weights, and clustering on id*Comparison to Mangoro region**Comparison to Moramanga Fivondronana***Comparison to agriculture as principal activity
Dprobit Model: Knowledge of Soil Degradation from slash-and-burn
Table 10a: dprobit model for knowledge of soil degradation from slash-and-burn (2288 Observations)Variable dF/dx p-value
Year 0.1075 0.010Integration site 0.1461 0.000Program effect (Integration*year) -0.0243 0.618
Adjusted for sampling weights and clustering on id
Table 10b: dprobit model for knowledge of soil degradation from slash-and-burn (2249 Observations)Variable dF/dx p-value
Year 0.0997 0.026Integration site 0.1472 0.001Program effect (Integration*year) -0.0107 0.835Age of household head 0.0003 0.716Marital status of household head (Not in union=1) -0.0248 0.562Education of household head 0.0521 0.000Sex of household head (Female=1) -0.1300 0.006Cultivable land (yes=1) 0.1772 0.005Haute Matsiatra region* 0.0395 0.214Anosy Region* -0.1376 0.000Anosibe An'Ala Fivondronana** -0.0728 0.091Ambalvao Fivondronana** 0.1048 0.043Amboasary Fivondronana** -0.0034 0.942Principal activity of household head***
animal husbandry/fishing -0.0455 0.620Bee-Keeping 0.1639 0.533Commerce -0.0990 0.164Salaried -0.0117 0.815Artisan 0.0177 0.812Forest exploitation 0.0497 0.631Other 0.0662 0.420
Adjusted for covariates, sampling weights, and clustering on id*Comparison to Mangoro region**Comparison to Moramanga Fivondronana***Comparison to agriculture as principal activity
Limitations to the Quasi-Experiment
Not randomized Short follow-up period
(2001-2004) Wide definition of
integration and control DiD analyses assume “time
invariant selection bias” (Ravallion, 2005)
Uncontrollable external events affecting some sites but not others
Pho
to: E
ckha
rd K
lein
au, 2
001
Conclusions Results are inconclusive, especially given
some limitations in design and data 5/6 of indicators showed no program effect Integration decreased moderate/severe
childhood stunting by 13.51% More research is needed
Need to clearly define and operationalize integration and control
Missing piece: cost-effectiveness
References Kleinau, E., Randriamananjara, O., & Rosenweig, F. (2005).
Healthy People in a Healthy Environment: Impact of an Integrated Population, Health, and Environment Program in Madagascar. Washington, D.C./USAID: Environmental Health Project.
Ravallion, Martin. (2005) Evaluating Anti-Poverty Programs (06/01/2005). World Bank Policy Research Working Paper Series. Available at SSRN: http://ssrn.com/abstract=922915 (Retrieved August 23, 2007).
Wooldridge, Jeffrey. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.
Many thanks to Heather D’Agnes and Eckhard Kleinau for their guidance and help!