Baseline Study of the Maharo Resilience 
Food Security Activity (RFSA) in Madagascar
August 2021 | Volume I 
IMPEL | Implementer-Led Evaluation & Learning Associate Award
ABOUT IMPEL
The Implementer-Led Evaluation & Learning Associate Award (IMPEL) works to improve the design and implementation of 
Bureau for Humanitarian Assistance (BHA)-funded resilience food security activities (RFSAs) through implementer-led 
evaluations and knowledge sharing. Funded by the USAID Bureau for Humanitarian Assistance (BHA), IMPEL will gather 
information and knowledge in order to measure performance of RFSAs, strengthen accountability, and improve guidance and 
policy. This information will help the food security community of practice and USAID to design projects and modify existing 
projects in ways that bolster performance, efficiency, and effectiveness. IMPEL is a seven-year activity (2019–2026) 
implemented by Save the Children (lead), TANGO International, Tulane University, Causal Design, and Innovations for Poverty 
Action.
RECOMMENDED CITATION
IMPEL. (2021). Baseline Study of the Maharo Resilience Food Security Activity (RFSA) in Madagascar (Vol. I). Washington, DC: 
The Implementer-Led Evaluation & Learning Associate Award. 
PHOTO CREDITS
Save the Children / Charlie Forgham-Bailey.
DISCLAIMER
This report is made possible by the generous support of the American people through the United States Agency for 
International Development (USAID). The contents are the responsibility of the Implementer-Led Evaluation & Learning (IMPEL) 
award and do not necessarily reflect the views of USAID or the United States Government.
CONTACT INFORMATION
IDEAL Activity
c/o Save the Children
899 North Capitol Street NE, Suite #900
Washington, DC 20002
www.fsnnetwork.org 
IMPEL@savechildren.org
PREPARED BY: 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Acknowledgments i 
ACKNOWLEDGMENTS
The Causal Design research team would like to recognize several individuals who were instrumental in 
creating and refining this document. This includes Gerard Rabesoa and Daniel Rooney with Catholic 
Relief Services, and Mohamoud Ahmed and Tom Spangler with IMPEL. Special thanks also to Mara 
Mordini, Justin Mupeyiwa, Benita O’Colmain, Arif Rashid, and Adam Trowbridge with USAID for their 
ongoing technical support. We also are grateful to the field team and their tremendous work on the 
data collection under difficult circumstances.
IMPEL | Implementer-Led Evaluation and Learning
ii Table of Contents
TABLE OF CONTENTS
Acknowledgments.................................................................................................................. i
List of Tables and Figures...................................................................................................... iii
Acronyms............................................................................................................................. vi
Executive Summary ............................................................................................................. vii
Introduction .................................................................................................................... 1
1.1 Overview of the Evaluation Research ............................................................................................ 1
Methodology and Limitations .......................................................................................... 2
2.1 Evaluation Design........................................................................................................................... 2
2.1.1 Randomization and Sampling Strategy ......................................................................... 2
2.1.2 Sampling Frame............................................................................................................. 3
2.1.3 Questionnaire Development......................................................................................... 3
2.1.4 Field Preparation........................................................................................................... 4
2.1.5 Data Collection.............................................................................................................. 4
2.2 Limitations and Challenges ............................................................................................................ 5
2.2.1 Non-response and List Issues........................................................................................ 5
2.2.2 Other Issues .................................................................................................................. 6
2.2.3 Limitations..................................................................................................................... 6
Findings........................................................................................................................... 7
3.1 Characteristics of the Study Population......................................................................................... 7
3.2 Food Security.................................................................................................................................. 9
3.3 Child Nutrition and Health ...........................................................................................................12
3.3.1 Nutrition......................................................................................................................13
3.3.2 Anthropometry ...........................................................................................................15
3.4 Women’s Health, Maternal Nutrition, and Reproductive Health................................................17
3.5 Water, Sanitation, and Hygiene (WASH) Practices......................................................................20
3.6 Agriculture....................................................................................................................................23
3.6.1 Crops...........................................................................................................................23
3.6.2 Livestock......................................................................................................................28
3.7 Poverty Measurement .................................................................................................................30
3.8 Gender Dynamics.........................................................................................................................31
3.8.1 Use of Financial Resources..........................................................................................31
3.8.2 Credit...........................................................................................................................33
3.8.3 Additional Decision-Making Areas..............................................................................34
3.9 Resilience......................................................................................................................................36
3.9.1 Ability to Recover from Shocks and Stresses Index ....................................................36
3.9.2 Social Capital Index .....................................................................................................38
3.9.3 Absorptive Capacity Index...........................................................................................38
3.9.4 Adaptive Capacity Index..............................................................................................40
3.9.5 Transformative Capacity Index ...................................................................................42
Comparison of Treatment and Control Groups ............................................................... 45
Conclusion..................................................................................................................... 46
Annex A: Balance Tables...................................................................................................... 47
Annex B: Risk Mitigation Plan.............................................................................................. 53
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
List of Tables and Figures iii
Volume II Annexes
Annex C: MAHARO RFSA Impact Evaluation Pre-Analysis Plan 
Annex D: Data Collection Tool
LIST OF TABLES AND FIGURES
Table 1. Planned versus actual survey numbers by CRS cluster and assignment......................................... 4
Table 2. Individual response rate.................................................................................................................. 4
Table 3. Percent of sampled households not located and replaced............................................................. 6
Table 4. Basic household-level statistics (weighted) .................................................................................... 7
Table 5. BL individual-level demographic information................................................................................. 7
Table 6. Summary statistics for household head.......................................................................................... 9
Table 7. Percent of households responding yes to "During the past 30 days, was there a time when you or 
others in your household..." ......................................................................................................................... 9
Table 8. Prevalence of moderate and severe food insecurity in the household based on the Food Insecurity 
Experience Scale (FIES) ...............................................................................................................................10
Table 9. Disaggregated by household type, percent of households responding yes to “During the past 30 
days, was there a time when you or others in your household…”.............................................................10
Table 10. Mean scores of households with poor, borderline, and acceptable FCS....................................12
Table 11. Percent of households with poor, borderline, and acceptable FCS............................................12
Table 12. Prevalence of children 6–23 months consuming a diet of minimum diversity (MDD-C)............13
Table 13. Percent of children ages 6–23 months receiving a minimum acceptable diet (MAD) ...............14
Table 14. Minimum meal frequency, breastfed and non-breastfed children 6-23 months.......................14
Table 15. Prevalence of exclusive breastfeeding of children under 6 months of age................................14
Table 16. Percent of children under 5 (0-59 months) who had diarrhea in the prior 2 weeks..................15
Table 17. Percent of children under 5 (0-59 months years old with diarrhea treated with oral rehydration 
therapy (ORT)..............................................................................................................................................15
Table 18. Prevalence of underweight children under 5 years old..............................................................15
Table 19. Disaggregated statistics for underweight ...................................................................................16
Table 20. Percent of women of reproductive age consuming a diet of minimum diversity ......................18
Table 21. Percent of births receiving at least four antenatal care visits during pregnancy .......................19
Table 22. Contraceptive prevalence rate (CPR)..........................................................................................19
Table 23. Percent of women in a union who have knowledge of modern family planning methods that can 
be used to delay or avoid pregnancy..........................................................................................................19
Table 24. Percent of women in a union who made decisions about modern family planning methods...20
Table 25. Percent of households using basic drinking water services........................................................20
Table 26. Water use per capita per day......................................................................................................21
Table 27. Percent of households with soap and water at a handwashing station on premises................22
Table 28. Percent of households in target areas practicing correct use of recommended household water 
treatment technologies ..............................................................................................................................22
Table 29. Percent of households in target area practicing open defecation..............................................22
IMPEL | Implementer-Led Evaluation and Learning
iv List of Tables and Figures
Table 30. Percent of households with access to basic sanitation services.................................................23
Table 31. Crops grown ................................................................................................................................23
Table 32. Percent of farmers who used financial services (savings, agricultural credit, and/or agricultural 
insurance) in the past 12 months & Percent of farmers who practiced the value chain interventions 
promoted by the activity in the past 12 months........................................................................................24
Table 33. Improved management practices/technologies for cassava ......................................................24
Table 34. Improved management practices/technologies for cowpea......................................................25
Table 35. Improved management practices/technologies for sorghum....................................................25
Table 36. Percent of producers who have applied targeted improved management practices or 
technologies................................................................................................................................................26
Table 37. Yield of targeted agricultural commodities within target areas................................................26
Table 38. Goat farming ...............................................................................................................................28
Table 39. Poultry farming............................................................................................................................29
Table 40. Fishing .........................................................................................................................................29
Table 41. Livestock ownership....................................................................................................................30
Table 42. Poverty measures........................................................................................................................30
Table 43. Disaggregated poverty measures by household type................................................................. 31
Table 44. Percent of women and men in a union who earned cash in the past 12 months......................32
Table 45. Percent of women in a union and earning cash who report participation in decision about the 
use of self-earned cash ...............................................................................................................................32
Table 46. Percent of men in union and earning cash who report spouse/partner participation in decisions 
about the use of self-earned cash ..............................................................................................................33
Table 47. Percent of women/men in a union who used credit in the previous 12 months.......................33
Table 48. Percent of women/men in a union who make decisions about credit.......................................34
Table 49. Women in a union who report medium or high amount of household input, age ....................35
Table 50. Women in a union who report medium or high amount of household input, categories.........35
Table 51. Women in a union who report medium or high amount of household input, household head's 
sex ...............................................................................................................................................................36
Table 52. Ability to recover from shocks and stresses................................................................................36
Table 53. Index of social capital at the household level .............................................................................38
Table 54. Absorptive capacity index ...........................................................................................................39
Table 55. Adaptive capacity index ..............................................................................................................40
Table 56: Transformative capacity index ....................................................................................................42
Table 57. Household roster balance table..................................................................................................45
Table 58. Food security...............................................................................................................................47
Table 59. Child nutrition and health ...........................................................................................................47
Table 60. Anthropometry............................................................................................................................48
Table 61. Women's health, maternal nutrition, and reproductive health .................................................48
Table 62. WASH...........................................................................................................................................48
Table 63. Agriculture—cassava, sorghum, and cowpea .............................................................................49
Table 64. Agriculture–yield.........................................................................................................................50
Table 65. Poverty measurements...............................................................................................................50
Table 66. Use of financial resources...........................................................................................................50
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
List of Tables and Figures v 
Table 67. Credit...........................................................................................................................................50
Table 68. Female household input..............................................................................................................50
Table 69. Resilience.....................................................................................................................................51
Figure 1. Age and gender composition (unweighted) of the sample ........................................................... 8
Figure 2. Distribution of weight-for-age z-score.........................................................................................16
Figure 3. Dietary diversity scores for women of reproductive age ............................................................18
Figure 4. Box plot of yield estimates for target crops.................................................................................27
Figure 5. Most common shocks reported by households...........................................................................37
IMPEL | Implementer-Led Evaluation and Learning
vi Acronyms
ACRONYMS
ANC Antenatal Care
BHA Bureau for Humanitarian Assistance
BL/EL Baseline/Endline
CRS Catholic Relief Services
F&M Female and Male Adults
FAO Food and Agriculture Organization 
FCS Food Consumption Score
FIES Food Insecurity Experience Scale
FNM Adult Female No Adult Male
HAZ Height-for-Age Z-Score
IE Impact Evaluation
IMPEL Implementer-Led Evaluation & Learning Associate Award
IP Implementing partner
MAD Minimum Acceptable Diet
MDD Minimum Dietary Diversity
MNF Adult Male No Adult Female
NGO Non-Governmental Organization
ORS Oral Rehydration Solution
ORT Oral Rehydration Therapy
PPP Purchasing Power Parity
RCT Randomized Controlled Trial
RFSA Resilience Food Security Activity
WASH Water, Sanitation and Hygiene
WAZ Weight-for-Age Z-Score
WHZ Weight-for-Height Z-Score
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Executive Summary vii
EXECUTIVE SUMMARY
This report captures baseline (BL) round data and observations collected from February to March 2021 
of the Maharo Resilience Food Security Activity (RFSA) that is being implemented by Catholic Relief 
Services (CRS) in southern Madagascar. This activity attempts to address and mitigate acute levels of 
food insecurity experienced by communities in this region of Madagascar. Apart from general 
demographics, study indicators include food security; child nutrition and health; women’s maternal 
nutrition and reproductive health; water, sanitation, and hygiene practice; agricultural practice and 
production; poverty measurement; gender dynamics; and resilience. A complementary endline (EL) 
survey is anticipated to be conducted from February to March of 2025. 
Baseline Study Methodology
The Impact Evaluation (IE) of the Maharo RFSA relies on a Clustered Randomized Controlled Trial (RCT) 
design to analyze differences between treatment and control groups. In the case of the Maharo RFSA IE, 
the primary difference between treatment and control is the presence of additional community tailored 
livelihood support. As a result, the analysis will focus on the additional marginal effect of these 
livelihood activities on food security and other development outcomes. Overall analysis at BL suggests 
that the IE is well placed to estimate these differences at EL, and that treatment and control groups are 
similar enough on key characteristics.
Study Limitations
Several factors posed potential challenges that the research team will adjust for or monitor throughout 
the course of the IE activity. High levels of drought condition in the southern area have contributed to 
higher-than-normal levels of outmigration. This is complicated by the challenges imposed by the 
coronavirus. The research team will work closely with implementing partners (IPs) to gauge this issue 
and take steps ahead of EL data collection to mitigate, if necessary. Finally, safety protocols that limited 
contact with beneficiaries and enforced social distancing measures ruled out the possibility of collecting 
some BL indicators, specifically height measurements for women of reproductive age and children.
Key Findings
Demographic Profiles
By the end of data collection efforts 4,595 households had been surveyed. There were little to no 
observed trends of variance between the treatment and control group on both the individual and 
household level, suggesting that the two groups are similar in terms of overall demographic 
characteristics. Demographic data from 2011 demonstrating that 24% of the population were under the 
age of 5 match the distribution found at BL (25%). Self-reported household head characteristics follow a 
similar trend as other individual indicators in showing little difference between treatment and control 
households. 
Food Security
Reported severe drought conditions in the area suggest reduced food security across the entire region 
of the study. Food security was estimated using two standard measurement approaches, the Food 
Insecurity Experience Scale (FIES) and the Food Consumption Score (FCS) index. Based on the FIES, 60% 
of the population is facing severe food insecurity, and over 97% are at least moderately food insecure. 
IMPEL | Implementer-Led Evaluation and Learning
viii Executive Summary
Severe food insecurity is lower in treatment areas compared to control (56% versus 63%). The FCS, 
which calculates overall consumption levels across food groups while accounting for cultural and 
regional weights for food preference and importance, finds that fewer than 20% of all households have 
an acceptable FCS and more than 45% of households are considered to have a poor FCS. Consistent with 
the FIES, control areas have slightly worse FCS outcomes (p-value = 0.05). Disaggregation by household 
adult type show some differences as well with fewer than 9% of MNF households having an acceptable 
score, compared to 19% of households with both male and female adults present.
Child Nutrition and Health
Overall quality of diet for children 6 to 23 months of age appears to be poor across the survey 
population. Only 3% of all children ages 6 to 23 months met Minimum Dietary Diversity (MDD) criteria 
and only 2% met Minimum Acceptable Diet (MAD) standards. Dietary diversity was slightly poorer in 
control areas while they were more equal in terms of children receiving a MAD. Approximately 34% of 
children under 5 years old are reported to have experienced diarrhea within the last 2 weeks. Of those 
that experienced diarrhea, only 14% reported use of oral rehydration therapy (ORT) to treat symptoms. 
There does not appear to be any observed trend across treatment status or the gender of the child. 
Given the ongoing COVID-19 pandemic, weight of children under 5 years of age was the only 
anthropometric indicator captured in the BL round of surveys. More than a third of children are 
underweight and approximately 15% are severely underweight. Observations of the distribution also 
show a high concentration of children around thresholds suggesting larger populations on the cusp of 
being underweight.
Women’s Health, Maternal Nutrition, and Reproductive Health
Observations around the health and reproductive decisions of women of reproductive age among 
surveyed households suggest poor food consumption diversity but minimally acceptable levels of access 
to health personnel during pregnancy. Altogether, less than 2% of women consumed a diet that meets 
the MDD criteria. Nearly 69% of women consumed two or fewer food groups. On average 65% of 
women received the recommended number of antenatal care (ANC) visits (at least four) during their 
most recent pregnancies. Additionally, contraceptive use is not widespread with 13% of women 
reporting using a modern method of birth control. Injectable contraception accounts for more than half 
of all contraception methods used.
Water, Sanitation, and Hygiene (WASH) Practices
Based on indicator criteria, 13% of all households have access to basic drinking water services. While 
most water sources are available year-round, fewer than a third of households have water within 30 
minutes and less than half (43%) have access to an improved source. Regarding treatment, sanitation, 
and hygiene practice only 4% of households were observed to have handwashing facilities available in 
the home and less than half (41%) report treating water with filtering or disinfecting process. Most 
households (57%) practice open defecation and only 15% have household-level improved sanitation 
facilities. Among those not practicing open defecation, 83% utilized unimproved technology (uncovered 
pit latrine), and 23% share the facility with other households.
Agriculture
More than 93% of households are engaged in farming (either crops or livestock), and over 96% of 
farmers in the Maharo region own the land they cultivate. Reported crops of focus include cassava, 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Executive Summary ix
cowpea, and sorghum. Yield estimates for farmers that were able to successfully harvest all three crops 
are in the range of what would be expected in southern Madagascar under drought conditions, 
however, because of the prolonged, severe drought, many farmers (70%) are omitted from the yield 
calculation because they had likely abandoned or stopped working their plots well before harvest. 
Regarding other support practices, use of credit and savings is not common among farmers, and crop 
insurance is almost non-existent and value chain participation is generally low—less than 1% of farmers 
report participation in value chain activities. Livestock of focus in the BL include goats and poultry, 
though the BL did also explore fishing practice among surveyed households. Of significant note is that 
42% of poultry farmers report poultry dying in the last year with most farmers (87%) reporting that they 
had not used any vaccination methods for their animals. Responses around fishing suggest that fewer
than half of the population engage in the practice, though confusion around the survey question may 
have led to underreporting.
Poverty Measurement
Based on daily per capita expenditures of less than $1.90 per day (2011 purchasing power parity (PPP)), 
the poverty rate among households surveyed is approximately 90%. The depth of poverty of the poor is 
57%, which means that the average poor person is 57% below the poverty line. In monetary terms, this 
means it would require an additional $1.08 per person per day to bring every poor person out of 
poverty.
Gender Dynamics
Gender dynamics are captured through the eight indicators in this section. Among women who are 
earning cash, a large majority (82%) reported that they participate in decisions about how to use the 
cash, whether solely or jointly with others. women and men in a union report borrowing at similar rates 
across any source (36%). However, men in a union participate in decisions about credit at a much higher 
rate than women in a union (p-value = 0.00), with 89% of men reportedly making credit decisions, which 
is consistent across age groups. Almost half of women have a medium or high input when making 
decisions in their home (50%).
Resilience
Resilience indicators were captured through several questions including indices that were constructed 
to assess overall resilience capacities. Generally, households perceived their ability to meet their current 
needs as worse than the previous year and suspect that their future ability to meet these needs will 
deteriorate. Nearly all households (98%) listed the drought as one of the shocks affecting them. Other 
common shocks listed included rising food prices and crop pests. Out of the average of 3.1 shocks 
experienced across the sample, households perceived those shocks to be severe in nature, likely 
impacting perceived ability to recover. There appears to be limited difference across treatment and 
control communities when gauging exposure to shocks and overall capacities that might mitigate their 
impact. Resilience capacity sub-scores suggest that households may have limited access to networks and 
services that could be leveraged to mitigate and adapt to further disasters.
This page is intentionally left blank. 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Introduction 1
INTRODUCTION
1.1 Overview of the Evaluation Research
Southern Madagascar is experiencing a prolonged drought and much of the population is facing a severe 
food security crisis. The Maharo RFSA aims to provide development, and nutritional support to a large 
population of households in southern Madagascar. In addition to these services, the RFSA package also 
includes a range of community-tailored livelihood support activities. The primary objective of the IE is to 
measure the impact of this livelihood support on reducing food insecurity and increasing well-being for 
households in southern Madagascar. This report summarizes the results of the BL study conducted in 
February–March 2021.
The evaluation uses a randomized controlled trial design which randomized the communities which 
would receive the additional activities. Given that general support was planned to be delivered 
unconditionally in the region of focus, while livelihood activities were designed to be more targeted and 
tailored to specific community needs, the Maharo RFSA created the potential to design an experimental 
IE to estimate the marginal effect of a set of livelihood activities on the "standard" RFSA package. The 
evaluation seeks to inform the larger knowledge base around the efficacy of the RFSA among vulnerable 
populations and how benefits to vulnerable households can be further maximized. Based on this, the IE
of the RFSA focuses on the following research question:
● What is the additional impact on food security and nutrition outcomes in communities that 
receive additional tailored livelihood activities in conjunction with development, emergency, 
and nutritional support activities?
The BL study relies on quantitative methods to measure BL indicators collected in the RFSA target area. 
The survey provides BL estimates on the status of communities and households across BHA standard 
indicators. Causal Design has worked closely with the BHA and relevant stakeholders to identify key 
learning objectives, and to ensure that the BL survey and study are able to contribute to this learning 
where possible.
IMPEL | Implementer-Led Evaluation and Learning
2 Methdology and Limitations
METHODOLOGY AND LIMITATIONS
The IE of the Maharo activity uses a clustered RCT design. Rather than a true control group with no 
intervention, however, eligible households in the control areas will receive food assistance and maternal 
support from CRS, and treatment areas will receive this assistance plus additional livelihood support 
activities, such that the estimated effect reflects the marginal impact of this additional livelihood 
support. While the project started in October 2021, a BL survey of households was conducted in 
February–March 2021 due to COVID-19. The EL survey is planned for February–March 2025. 
2.1 Evaluation Design
2.1.1 Randomization and Sampling Strategy 
Level of Randomization
After discussion with the CRS team, village or fokontany-level randomization was deemed infeasible 
because (1) a single village or fokontany is too small for practical project implementation, and (2) the 
definitions and demarcations of fokontany are highly subject to change in these areas. Furthermore, 
there are only 20 communes in the Maharo intervention areas, which eliminated this as the level of 
randomization. CRS proposed defining “clusters” as groups of adjacent fokontany that could be served 
by a CRS team. The initial list from CRS proposed around 218 clusters, each with an average of three
fokontany.
Matched Pair Randomization
We chose a matched pair randomization approach to ensure better balance across treatment and control 
cluster characteristics prior to BL data collection compared with stratified random sampling. In a matched 
pair randomization approach, units are first matched based on variables related to outcomes. In other 
words, areas that look “similar” based on available data are paired together. One unit from each pair is 
then randomly assigned to treatment and the other to control. CRS provided us with a complete list of 
clusters and potential matching criteria. Data available at the time included location, proximity to a 
market, proximity to a health clinic, proximity to the coast, road and river access, and population. 
Preliminary matches were made and confirmed with CRS staff. This resulted in 98 pairs. Unmatched 
clusters were excluded from the BL, and CRS was free to operate in these areas as they saw fit.
Sample Size and Power Calculation
● A cluster randomized design.
● Intra-class correlation of 0.10.
1
● Significance level. 
● Power level of 80%. 
● Expected reduction in poverty over the life of the project of 8 percentage points. 
 
1 The ICC estimate comes from HAZ scores in the Madagascar Enquête Anthropométrique et Développement de l'Enfant 2011 
household data. A range of ICC numbers were tested to ensure that the ultimate sample size was not sensitive to this choice. 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Methodology and Limitations 3
● Inflation Factor for number of households with children under 5 years of age to sample. Using a 
household data set from 2011, 24% of the population is under 5 in the provinces of Toliara and 
Fianarantsoa. This implies an inflation factor of 1.2. 
● Non-response factor is 5%.
To capture sufficient children under 5 years old and to account for attrition, we would need
3,724 * 1.18 * 1.05 = 4,614 households total, or 24 households per cluster.
2.1.2 Sampling Frame
CRS completed a census of households in their area of intervention in June 2020. These data were not 
available when the initial matching was done. These data provided both the population numbers for 
each fokontany and a list of all households in these areas. Based on the census, the total population was 
slightly higher in control areas (49,739 compared to 48,021).
The power calculation was based on an average of 24 households per CRS cluster, but because some of 
these clusters are large, it was not feasible to randomly sample households at that level. Sampling two 
fokontany per CRS cluster and 12 households per fokontany would be equivalent. The CRS clusters, 
however, vary widely in both size and number of fokontany (villages). The sampling strategy adopted, 
therefore, involved sampling 1–4 fokontany per cluster, based on total cluster population. Table 1 below 
summarizes the number of fokontany, and households sampled. 
Households were then randomly sampled from these fokontany based on the census list. A list of 
households in each sampled fokontany, numbered 1–100, was provided to the survey teams. Households 
1–12 were the primary households and teams were instructed to make several attempts to contact these 
households. Replacement households were then contacted in sequential order, starting with 13.
2.1.3 Questionnaire Development
The BL survey was developed using previous BL surveys used by BHA and refined in consultation with 
BHA and the IPs. 
The following survey modules were included:
● Module A: Household identification. 
● Module B: Roster. 
● Module C: Food access. 
● Module D: Child nutrition and health. 
● Module E: Women’s nutrition and health. 
● Module F: WASH. 
● Module G: Agriculture. 
● Module H: Household expenditure. 
● Module J: Gender and cash use. 
● Module K: Gender and credit. 
● Module CRS: Female household input. 
● Module R: Resilience. 
IMPEL | Implementer-Led Evaluation and Learning
4 Methdology and Limitations
In addition, we created a short, commune-level survey to capture community-level variables such as 
public service availability, the activities of development or aid projects, and the presence of local 
community groups.
2.1.4 Field Preparation
Training was conducted in the southern city of Toliara. Restrictions related to the COVID-19 pandemic 
prevented international travel. During the first week of the survey, therefore, the United States-based 
team members checked in daily via live video feed to take questions about the survey instrument and 
about the BL data collection in general. The training was conducted over a 2-week period (January 26 to 
February 10) to allow more time for refining the tablet survey and for addressing enumerator questions. 
Two doctors assisted the team for 1 day of the training and covered three topics: infant nutrition, 
maternal nutrition, and anthropometry. Two survey pilots were conducted in two separate fokontany 
accessible from Toliara. 
In terms of organization, there were 90 field staff organized into teams. Each team was composed of 
four enumerators, one supervisor, and one person charged with anthropometry and survey verification. 
Two lead supervisors oversaw the entire data collection effort and worked to resolve any problems 
(with tablets, for example).
2.1.5 Data Collection
Overview
Data collection began on February 10 and ended March 24. A total of 4,595 households were surveyed 
in the Maharo BL. Slightly more households were surveyed in the control areas (2,317) compared to 
treatment areas (2,278), but this was expected because of the higher population in control areas.
Table 1. Planned versus actual survey numbers by CRS cluster and assignment
Treatment Control
Number 
fokontany 
sampled
CRS cluster 
population 
category
Number 
of 
clusters
Number of 
fokontany 
sampled
Number of 
households to 
be surveyed
Number 
of 
clusters
Number of 
fokontany 
sampled
Number of 
households to 
be surveyed
1 <20 percentile 32 32 384 31 31 372
2 20–70 45 90 1,080 46 92 1,104
3 70–90 15 45 540 14 42 504
4 >90 6 24 288 7 28 336
Planned Total 98 191 2,292 98 193 2,316
Actual Total 98 191* 2,278 98 193 2,317
*Local authorities in one sampled treatment fokontany did not allow the team to conduct the surveys. This 
fokontany was replaced by random selection from the remaining fokontany in that cluster.
Table 2. Individual response rate
Outcome Number in roster Number surveyed Response Rate
Women of reproductive age 4,497 4,112 91.44%
Children under 5 6,145 5,985 97.40%
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Methodology and Limitations 5
Outcome Number in roster Number surveyed Response Rate
Farmers 4,748 4,478 94.31%
Women in a union 2,667 2,410 90.36%
The names, definitions, and boundaries of fokontany in this region are particularly fluid. In at least one 
case, households listed in the census were not found in the fokontany identified, and in another, the 
named fokontany was not recognized locally. In the former instance, the listed households were found 
in a different fokontany in the same cluster and were surveyed. Authorities in one sampled fokontany
refused to allow the survey team access and this fokontany was replaced.
Data Quality Control
Each enumerator was introduced to a household by a supervisor or controller. The supervisor assisted at 
the beginning of each interview to ensure that the interviews began well, and that the interviewer 
recorded key household data (including number of members of the households, number of children, 
number of women of childbearing age, number of cash-earners in the household, etc.) in the interviewer’s
notebook in addition to recording these data in a tablet. The notes served to confirm that data was 
properly entered and that the correct numbers of people were included in the different modules.
Two lead supervisors were tasked with permanently monitoring the field teams until the end of the 
surveys to verify the data and to promptly solve problems. In the cases of households with unusual 
responses (a case of 0 expenses, for example), the lead supervisors confirmed responses in-person. Each 
team was visited by the lead supervisors at least twice while in the field. This was necessary not only to 
ensure the survey was conducted properly but also to solve equipment issues, which was a problem all 
teams experienced.
2.2 Limitations and Challenges
2.2.1 Non-response and List Issues
Only nine households refused to participate in the survey outright. There were some discrepancies 
between the information available in the June 2020 census and the situation on the ground at the time 
of the survey that have prevented us from being able to calculate the non-response rate. Many sampled 
households were not found in the fokontany in which they were listed on the CRS census and were 
replaced. Because the CRS census, which was finalized in June 2020, was the most current list and the 
only list with household names that focused on the areas of intervention, we did not anticipate this 
issue. This means we are unable to distinguish between households which were not residents and those 
which were residents but were not present or who otherwise were not available at the time of the 
survey. Approximately 78% of households that responded were primary households, i.e., one of the 12 
initially chosen to be surveyed in the fokontany. The reasons given by the field staff for the 
discrepancies between the CRS census and what our team found in the field include: (1) recent out￾migration as a result of the ongoing drought and poor economic conditions in the region, (2) households 
that may reside in rural areas during the agricultural season and elsewhere during the rest of the year, 
and (3) local leaders claiming additional households that may not belong to that fokontany with the goal 
of attracting additional aid. We found at least two cases of households which were listed under two 
IMPEL | Implementer-Led Evaluation and Learning
6 Methdology and Limitations
different fokontany. Table 3 displays the percent of primary households that the field team was unable 
to locate but were replaced with other randomly selected households from the same community. 
Table 3. Percent of sampled households not located and replaced
Control Treatment Total
Percent SD Percent SD Percent SD 
Percent of sampled households 
not located and replaced
22.3% 17.46 20.7% 15.82 21.5% 16.68
CRS signaled in early meetings that this local administrative unit, “fokontany,” which is roughly 
equivalent to a village, is not a fixed entity in the region. This was one reason for choosing not to 
randomize at the fokontany level. Local leaders create, divide, and rename fokontany at will for various 
political or even personal reasons. Nevertheless, because the unit of randomization (the CRS-defined 
cluster) was usually too geographically dispersed to use for the household sample, we did need to 
sample fokontany within clusters.
The movement of households and the fluid definition of fokontany in the region will likely present 
challenges at EL. We can expect a relatively high attrition rate and the need to devote extra resources to 
finding households.
2.2.2 Other Issues
Because of early problems with some of the tablets, there were several dozen duplicate surveys in the 
raw data. Some enumerators were initially unable to upload their surveys and transcribed them onto 
other tablets. It appears that most of the original surveys eventually did upload. The duplicates have 
been identified and removed.
2.2.3 Limitations
● There are several other large organizations providing food assistance and other development 
aid in the region. This will present challenges in identifying effects of the Maharo activity. CRS is 
working with these organizations to better track where organizations are working, and we are 
collecting information at the community level at both BL and EL. This information will help us to 
control for other interventions in our analysis.
● Because of issues found in the field with the census list used, we do not know the true non￾response rate. If, in fact, the local population numbers in the CRS census are higher than the 
current population, this would have implications for our BL estimates. We do not expect, 
however, that this will meaningfully affect our findings, as even adjusted population weights will 
necessarily be rough estimates.
● The survey was conducted during the COVID-19 pandemic. In order to maintain distance 
between enumerators and respondents and to minimize contact, only one anthropometric 
measure was collected at BL. After discussions with BHA and the IP, it was determined that child 
weight could be safely measured if the caretaker were asked to weigh the child. Thus, we can 
calculate the weight-for-age z-score (WAZ) for children under 5 years old. Height for adults and 
children was not collected. Because of the randomized design, however, collecting these 
measures at EL will still be informative.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 7
FINDINGS
This section provides summary statistics for the Maharo BL survey. Unless otherwise indicated, all mean 
and standard deviations are calculated using sampling weights. The number of observations reported 
are the actual (unweighted) respondent numbers. Results are not reported for cells with less than 30 
observations and are denoted with an N/A for not available.
3.1 Characteristics of the Study Population
This section provides the basic demographic information for the BL sample. As mentioned previously,
4,595 households were surveyed. Table 4 shows that most households have both a male and a female 
adult present. For the whole sample, 27% of households have only an adult female present, and only 
3.6% have only an adult male present. The average household has 5.5 people, including 1.4 children 
under 5 years old. 
Table 4. Basic household-level statistics (weighted)
Control Treatment All
Number surveyed 2,317 2,278 4,595
Percent households with adult male and female 70.1% 68.7% 69.4%
Percent households with adult female only 25.8% 28.2% 27.0%
Percent households with adult male only 4.1% 3.1% 3.6%
Household size 5.6 5.4 5.5
Household with children under 5 1.4 1.4 1.4
Households with children 5–14 1.9 1.9 1.9
Table 5 provides basic, individual demographic information on the sample. The household sample 
includes nearly 25,000 individuals. The average age is 17 and approximately 52% of those in the sample 
are female. Figure 1 shows the population by age and gender. Among adults, 46% engage in some type 
of farming or livestock activity and 42% worked for cash in the previous year.
Table 5. BL individual-level demographic information
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N 
Age 17.46 (0.20) 12,664 17.52 (0.19) 12,248 17.49 (0.14) 24,912
Percent of sample who 
are female
52.4% (0.55) 12,665 52.5% (0.55) 12,248 52.5% (0.39) 24,913
Percent of adults who are 
farmers
45% (0.86) 5,134 46.2% (0.86) 4,992 45.6% (0.61) 10,126
Percent of adults who 
worked for cash
41.3% (0.77) 5,134 41.6% (0.76) 4,992 41.5% (0.54) 10,126
Percent of school age 
household members with 
at least some schooling
52.1% (0.64) 9,550 52.3% (0.64) 9,207 52.2% (0.45) 18,757
IMPEL | Implementer-Led Evaluation and Learning
8 Findings
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N 
Percent of sample women 
of reproductive age
34.3% (0.53) 12,665 34.0% (0.52) 12,248 34.2% (0.37) 24,913
Percent of sample under 5 24.6% (0.48) 12,665 24.9% (0.48) 12,248 24.7% (0.34) 24,913
Percent of sample 5–14 34.7% (0.53) 12,665 34% (0.52) 12,248 34.3% (0.37) 24,913
Figure 1. Age and gender composition (unweighted) of the sample
The characteristics of the head of household are presented in Table 6.2 Female-headed households 
make up 38% of the sample. Roughly 68% of household heads have no formal schooling, and only 16% 
completed primary school.
 
2 Head of household is self-reported by the household.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 9
Table 6. Summary statistics for household head
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Age of the head of households 42.42 (0.44) 2,317 42.34 (0.45) 2,274 42.38 (0.32) 4,591
Percent of female head of 
households
36.8% (1.26) 2,319 38.7% (1.25) 2,275 37.8% (0.89) 4,594
Percent of head of households 
with no schooling
69% (1.21) 2,319 67.9% (1.18) 2,275 68.4% (0.84) 4,594
Percent of head of households 
with some schooling, less than 
primary
15% (0.91) 2,319 16.9% (0.94) 2,275 16% (0.66) 4,594
Percent of heads of households 
that completed primary or 
more
15.9% (0.97) 2,315 15.1% (0.90) 2,272 15.5% (0.66) 4,587
3.2 Food Security
This section presents findings on household food security. The first indicator in this section is the 
prevalence of food insecurity, which is measured using the FIES developed by the Food and Agriculture 
Organization of the United Nations (FAO).
The responses to these questions are analyzed by estimating a Rasch model using tools developed by 
the FAO in R software. Only households that responded to all questions are included in the analysis. 
Furthermore, extreme responses—those that responded either yes or no to all questions—are excluded 
from the estimation but included in the final calculation of the prevalence rates.
Overall, a majority (70%) of households responded yes to all eight questions, as illustrated in Table 6. 
This is especially notable as the recall period was changed to “During the past 30 days” from “During the 
past 12 months” to reflect the serious food insecurity situation. Only 11 households answered “no” to all 
questions. Consistent with the assumption that each question captures a progressively more severe 
experience, the proportion of “yes” answers generally declined with each question. The exception is 
that more people answered “yes” to question 4 than 5.3 4
Table 7. Percent of households responding yes to "During the past 30 days, was there a time when 
you or others in your household..."
Control Treatment All
Outcome Percent N Percent N Percent N 
Were worried you would not have enough food to 
eat because of a lack of money or other resources?
99.6% 2,241 99.1% 2,223 99.4% 4,464
Were unable to eat healthy and nutritious food 
because of a lack of money or other resources?
99.1% 2,241 99.1% 2,223 99.1% 4,464
 
3 The model produced a reliability score of 0.71, which suggests a good model fit. 
4 Cafiero et al., Methods for Estimating Comparable Prevalence Rates of Food Insecurity Experienced by Adults throughout the 
World, (Rome, Italy.: FAO, 2016)
IMPEL | Implementer-Led Evaluation and Learning
10 Findings
Control Treatment All
Outcome Percent N Percent N Percent N 
Ate only a few kinds of foods because of a lack of 
money or other resources?
99.3% 2,241 99.1% 2,223 99.2% 4,464
Had to skip a meal because there was not enough 
money or other resources to get food?
94.0% 2,241 92.7% 2,223 93.4% 4,464
Ate less than you thought you should because of a 
lack of money or other resources?
98.8% 2,241 97.1% 2,223 98.0% 4,464
Did not have food because of a lack of money or 
other resources?
82.6% 2,241 80.2% 2,223 81.4% 4,464
Were hungry but did not eat because there was not 
enough money or other resources?
83.5% 2,241 80.4% 2,223 82.0% 4,464
Went without eating for a whole day because of a 
lack of money or other resources?
79.2% 2,241 75.1% 2,223 77.2% 4,464
Table 8 summarizes the prevalence of moderate and severe food insecurity in the household based on 
the Food Insecurity Experience Scale (FIES). Approximately 60% of individuals are facing severe food 
insecurity and over 97% are considered to be at least moderately food insecure. This is unsurprising 
given the severe drought conditions. Severe food insecurity is lower in treatment areas compared to 
control (56% versus 63%).5 The answers to the eight questions are disaggregated by household type in 
Table 9. The answers to the first few questions are consistent across groups.
Table 8. Prevalence of moderate and severe food insecurity in the household based on the Food 
Insecurity Experience Scale (FIES)
Control Treatment All
Prevalence of severe food insecurity 
in households 
62.9% 56.3% 60.4%
Prevalence of moderate and severe 
food insecurity in households 
97.1% 97.7% 97.4%
Table 9. Disaggregated by household type, percent of households responding yes to “During the past 
30 days, was there a time when you or others in your household…” 
Control Treatment All
Outcome Percent N Percent N Percent N 
Were worried you would not have enough food to eat because of a lack of money or other resources?
F&M 99.7% 1,530 98.9% 1,500 99.3% 3,030
FNM 99.4% 612 99.6% 649 99.5% 1,261
MNF 100.0% 99 99.3% 74 99.7% 173
Were unable to eat healthy and nutritious food because of a lack of money or other resources?
F&M 99.0% 1,530 99.0% 1,500 99.0% 3,030
FNM 99.5% 612 99.5% 649 99.5% 1,261
 
5 The FIES model does not generate individual prevalence rates, which makes testing statistical difference challenging. 
However, the difference in the raw score (the number of "yes" responses) is statistically significant with a p-value of <0.01.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 11
Control Treatment All
Outcome Percent N Percent N Percent N 
MNF 100.0% 99 100.0% 74 100.0% 173
Ate only a few kinds of foods because of a lack of money or other resources?
F&M 99.2% 1,530 98.9% 1,500 99.1% 3,030
FNM 99.6% 612 99.7% 649 99.7% 1,261
MNF 100.0% 99 100.0% 74 100.0% 173
Had to skip a meal because there was not enough money or other resources to get food?
F&M 93.5% 1,530 92.0% 1,500 92.8% 3,030
FNM 95.3% 612 95.0% 649 95.2% 1,261
MNF 99.5% 99 95.3% 74 97.8% 173
Ate less than you thought you should because of a lack of money or other resources?
F&M 98.9% 1,530 96.8% 1,500 97.9% 3,030
FNM 98.3% 612 98.1% 649 98.2% 1,261
MNF 99.5% 99 98.6% 74 99.2% 173
Did not have food because of a lack of money or other resources?
F&M 81.7% 1,530 79.4% 1,500 80.6% 3,030
FNM 84.5% 612 82.4% 649 83.4% 1,261
MNF 95.3% 99 85.4% 74 91.3% 173
Were hungry but did not eat because there was not enough money or other resources?
F&M 82.8% 1,530 79.7% 1,500 81.2% 3,030
FNM 85.4% 612 82.5% 649 83.9% 1,261
MNF 92.6% 99 86.8% 74 90.2% 173
Went without eating for a whole day because of a lack of money or other resources?
F&M 78.3% 1,530 74.2% 1,500 76.2% 3,030
FNM 81.8% 612 77.8% 649 79.7% 1,261
MNF 87.0% 99 83.7% 74 85.7% 173
*Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
The second indicator is the percent of households with poor, borderline, and acceptable food 
consumption score (FCS). This is a weighted sum of eight food groups consumed by the household in the 
previous 7 days. The weights are based on the food group’s importance in the diet. For example, meat 
and dairy have a weight of 4, staples have a weight of 2, and sugars have a weight of 0.5. The FCS ranges 
from 0 to 112. Scores below 22 are considered to be a poor consumption score, scores 22–35 are 
considered borderline, and acceptable scores are above 35. As shown in Tables 10 and 11 the mean FCS 
is 25 and more than 45% of individuals are considered to have a poor FCS. Fewer than 20% of 
households have an acceptable FCS. As with the high FIES scores, the prevalence of households with 
IMPEL | Implementer-Led Evaluation and Learning
12 Findings
poor FCS is likely driven by the extreme drought conditions. Consistent with the FIES, control areas have 
slightly worse outcomes (p-value = 0.05).
The FCS disaggregated by household type and by overall score category is illustrated in Tables 10 and 11
below. Fewer than 9% of MNF households have an acceptable score, compared to 19% of households 
with both male and female adults present.
Table 10. Mean scores of households with poor, borderline, and acceptable FCS
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N 
Household FCS (0–112) 24.87 (0.34) 2,241 25.84 (0.33) 2,223 25.36 (0.24) 4,464
F&M 25.35 (0.34) 1,530 26.01 (0.33) 1,500 25.68 (0.24) 3,030
FNM 23.67 (0.41) 612 25.46 (0.39) 649 24.6 (0.28) 1,261
MNF 19.69 (0.61) 99 23.28 (0.60) 74 21.16 (0.43) 173
* Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Table 11. Percent of households with poor, borderline, and acceptable FCS 
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households with poor consumption 
score (<22)
47.2% 2,241 43.1% 2,223 45.1% 4,464
F&M 44.7% 1,530 41.5% 1,500 43.1% 3,030
FNM 53.7% 612 47.8% 649 50.7% 1,261
MNF 69.6% 99 50.8% 74 61.9% 173
Percent of households with borderline 
consumption score (22–35)
35.8% 2,241 36.6% 2,223 36.2% 4,464
F&M 37.5% 1,530 38.1% 1,500 37.8% 3,030
FNM 30.4% 612 31.6% 649 31.0% 1,261
MNF 25.0% 99 35.7% 74 29.4% 173
Percent of households with acceptable 
consumption score (>35)
17.0% 2,241 20.3% 2,223 18.7% 4,464
F&M 17.7% 1,530 20.4% 1,500 19% 3,030
FNM 15.9% 612 20.6% 649 18.3% 1,261
MNF 5.4% 99 13.5% 74 8.8% 173
* Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
3.3 Child Nutrition and Health
This section presents findings on child nutrition and health covering aspects around quality of diet, 
breastfeeding practice, as well as incidence of diarrhea. This section also includes anthropometric data, 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 13
however, due to safety concerns brought about by COVID-19, the collection of the full range of 
anthropometric measurements was not possible at BL.6
3.3.1 Nutrition
Prevalence of children 6-23 months consuming a diet of minimum dietary diversity (MDD) and percent 
of children 6-23 months receiving a minimum acceptable diet (MAD) are the indicators of nutrition
reported for children under 2 years old. The MDD-Children (MDD-C) uses the following eight food 
groups: (1) breastmilk, (2) grains, roots, and tubers, (3) legumes and nuts, (4) dairy products (milk, 
yogurt, and cheese), (5) flesh foods (meat, fish, poultry, and liver/organ meats), (6) eggs, (7) vitamin A￾rich fruits and vegetables, and (8) other fruits and vegetables. The criterion for achieving an MDD-C is 
consuming at least five of the eight food groups. Table 12 shows the percentage of children meeting 
these criteria. Only 3% of children consumed a MDD-C and the percentage was lower for girls, 
particularly in the control areas. Looking at the consumption of the individual food groups. Grains are 
the most consumed (71%), followed by vitamin A-rich fruits and vegetables (54%), other fruits and 
vegetables (21%), and legumes and nuts (11%). Dairy and flesh foods are consumed by 6% and 7% of 
children, respectively, and eggs are consumed by less than 1%.
Table 12. Prevalence of children 6–23 months consuming a diet of minimum diversity (MDD-C)
Control Treatment All
Outcome Percent N Percent N Percent N
Children (ages 6–23 months) 2.4% 814 3.8% 823 3.1% 1,637
Male children (ages 6–23 months) 3.3% 396 3.6% 415 3.4% 811
Female children (ages 6–23 months) 1.5% 418 4.1% 408 2.8% 826
A MAD is defined as eating a certain number of times per day (minimum meal frequency) in addition to 
having a MDD. MDD for the MAD is defined using six or seven food groups depending on whether the 
child is breastfed or not.7 As shown in Table 13, only 2.3% of children are receiving a MAD. Female and 
male children are similarly likely to receive a MAD (p-value = 0.15).
Minimum meal frequency (MMF) is used to calculate the MAD. The MMF is calculated for both 
breastfed and non-breastfed children. The criteria to meet this for breastfed children is to have three or 
more feedings of solid, semi-solid, or soft foods and be between the ages of 9 and 23 months. In total 
25.3% of breastfed children met these criteria, which is illustrated in Table 14. The criterion for non￾breastfed children is to have four or more feedings of solid, semi-solid, or soft foods in addition to two 
or more milk feedings and be between the ages of 6 and 23 months. Table 14 shows that 3.8% of non￾breastfed children surveyed met this standard. The low rates of children eating a MAD could be driven 
by the extreme drought conditions.
 
6 This includes Prevalence of wasted (WHZ < -2) children under 5 years old (0–59 months), Prevalence of stunted 
children (HAZ <-2) under 5 years old (0–59 months) (BL 4), and Prevalence of underweight (BMI <18.5) women of 
reproductive age (BL 7)
7 The MAD does not include breastmilk as a food group. It includes dairy products as a food group for breastfed children and 
excludes it for non-breastfed children.
IMPEL | Implementer-Led Evaluation and Learning
14 Findings
Table 13. Percent of children ages 6–23 months receiving a minimum acceptable diet (MAD)
Control Treatment All
Outcome Percent N Percent N Percent N
Children (ages 6–23 months) 1.8% 814 2.8% 823 2.3% 1,637
Male children (ages 6–23 months) 2.7% 396 3.1% 415 2.9% 811
Female children (ages 6–23 months) 0.8% 418 2.5% 408 1.7% 826
Table 14. Minimum meal frequency, breastfed and non-breastfed children 6-23 months
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N 
Minimum meal frequency for 
breastfed children ages 6–23 
months
22.5% (0.027) 597 27.8% (0.021) 617 25.3% (0.016) 1,214
Minimum meal frequency for 
non-breastfed children ages
6–23 months
4.6% (0.017) 219 2.9% (0.013) 206 3.8% (0.011) 425
The prevalence of exclusive breastfeeding children under 6 months is illustrated in Table 15. This is 
defined as the children under the age of 6 months who were exclusively breastfed during the day 
preceding the survey, excluding any oral rehydration solution (ORS). Survey responses at BL indicate that 
exclusive breastfeeding practice is fairly even across treatment and control areas between 37–40% 
across all children under 6 months. Rates are similar across male and female children (p-value = 0.50).
Table 15. Prevalence of exclusive breastfeeding of children under 6 months of age
Control Treatment All
Outcome Percent N Percent N Percent N 
Children (under 6 months of age) 38.9% 222 38.4% 200 38.7% 422
Male children (under 6 
months of age)
40.9% 117 39.5% 94 40.3% 211
Female children (under 6 
months of age)
36.2% 105 37.5% 106 36.9% 211
The following two indicators focus on the percent of children under 5 (0-59 months) who had diarrhea in 
the prior two weeks and percent of children under five (0-59 months) with diarrhea treated with Oral 
Rehydration Therapy (ORT). A positive incidence is defined as a child experiencing an episode of 
diarrhea any time in the 2 weeks that preceded the survey while ORT is defined as receiving an ORS, 
recommended home fluids, or increased fluids.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 15
Table 16. Percent of children under 5 (0–59 months) who had diarrhea in the prior 2 weeks
Control Treatment All
Outcome Percent N Percent N Percent N 
Children (ages 0–59 months) 34.7% 2,682 34.1% 2,468 34.4% 5,150
Male children (ages 0–59 
months)
35.1% 1,333 35.8% 1,209 35.5% 2,542
Female children (ages 0–59 
months)
34.2% 1,349 32.5% 1,259 33.4% 2,608
Table 17. Percent of children under 5 (0–59 months years old with diarrhea treated with oral 
rehydration therapy (ORT)
Control Treatment All
Outcome Percent N Percent N Percent N 
Children (0–59 months) 14.5% 875 13.3% 817 13.9% 1,692
Male children (0–59 months) 15% 448 12.2% 422 13.6% 870
Female children (0–59 
months)
14% 427 14.5% 395 14.2% 822
3.3.2 Anthropometry 
Anthropometric indicators traditionally include measures of prevalence rates of wasting (weight-for￾height z-score (WHZ)), stunting (height-for-age z-score (HAZ)), and being underweight (weight-for-age z￾score (WAZ)). Given limitations to data collection due to the ongoing COVID-19 pandemic, only the 
weight of children under 5 years of age was captured in the BL round of surveys. Children without a 
known birth month and year were excluded. Children with WAZ scores of less than -2 standard 
deviations are considered underweight and those with scores of less than -3 standard deviations are 
considered severely underweight. 
More than a third of children are underweight and approximately 15% are severely underweight. Figure 
2 is a histogram showing the distribution of WAZ scores. The dark blue lines mark the cutoff points for 
underweight and severely underweight. The figure shows that there are many children close to being 
underweight with scores just above -2.
Table 18. Prevalence of underweight children under 5 years old
Control Treatment All
Outcome Mean N Mean N Mean N 
Weight-for-age z-score -1.64 2,900 -1.59 2,848 -1.62 5,748
Percent of children under 5 years old 
underweight
37.7% 2,900 35% 2,848 36.3% 5,748
Percent of children under 5 years old
severely underweight
15.6% 2,900 14.2% 2,848 14.9% 5,748
Percent of children under 5 years old 
with a normal weight for age
31.0% 2,900 30.8% 2,848 30.9% 5,748
IMPEL | Implementer-Led Evaluation and Learning
16 Findings
Figure 2. Distribution of weight-for-age z-score 
Table 19 shows the underweight statistics disaggregated by gender and by age group. Boys have slightly 
lower (worse) WAZ on average than girls and the older group (2–5 years) has lower z-scores than infants 
under 2 years (p-value = 0.00). This pattern is also reflected in the percent underweight and severely 
underweight (p-value = 0.00).
Table 19. Disaggregated statistics for underweight
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N 
Weight-for-age z-score
Female (0–59 
months)
-1.61 (0.04) 1,477 -1.49 (0.04) 1,452 -1.55 (0.03) 2,929
Male (0–59 months) -1.68 (0.05) 1,423 -1.69 (0.05) 1,396 -1.68 (0.03) 2,819
0–23 months -1.3 (0.06) 1,032 -1.34 (0.06) 1,027 -1.32 (0.04) 2,059
24–59 months -1.83 (0.04) 1,868 -1.73 (0.04) 1,821 -1.78 (0.03) 3,689
Percent children underweight
Female (0–59 
months)
37.46 (1.61) 1,477 31.35 (1.42) 1,452 34.37 (1.08) 2,929
Male (0–59 months) 37.92 (1.62) 1,423 38.69 (1.59) 1,396 38.31 (1.13) 2,819
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 17
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N 
0–23 months 31.78 (1.82) 1,032 31.42 (1.71) 1,027 31.6 (1.25) 2,059
24–59 months 40.88 (1.45) 1,868 36.9 (1.36) 1,821 38.89 (1.00) 3,689
Percent children severely underweight
Female (0–59 
months)
14.75 (1.18) 1,477 12.53 (1.02) 1,452 13.63 (0.78) 2,929
Male (0–59 months) 16.38 (1.27) 1,423 15.85 (1.13) 1,396 16.12 (0.85) 2,819
0–23 months 12.71 (1.29) 1,032 12.93 (1.18) 1,027 12.83 (0.87) 2,059
24–59 months 17.09 (1.14) 1,868 14.83 (0.98) 1,821 15.96 (0.75) 3,689
Percent children normal weight
Female (0–59 
months)
32.77 (1.52) 1,477 31.95 (1.50) 1,452 32.36 (1.07) 2,929
Male (0–59 months) 29.19 (1.50) 1,423 29.64 (1.53) 1,396 29.42 (1.07) 2,819
0–23 months 41.4 (1.91) 1,032 38.4 (1.89) 1,027 39.87 (1.34) 2,059
24–59 months 25.39 (1.25) 1,868 26.59 (1.27) 1,821 25.99 (0.89) 3,689
3.4 Women’s Health, Maternal Nutrition, and Reproductive 
Health
This section focuses on the health and reproductive decisions of women of child-bearing age. The first 
indicator measures the percent of women of reproductive age (WRA) consuming a diet of minimum
diversity (MDD-W). WRA includes all women in the household 15–49 years old. MDD is measured by 
counting the number of food groups a woman consumed during the previous day and night. The food 
groups are grains, white roots and tubers, and plantains, pulses (beans, peas, and lentils), nuts and 
seeds, dairy, meat, poultry and fish, eggs, dark green leafy vegetables, other vitamin A-rich fruits and 
vegetables, other vegetables, and other fruits. The criteria for MDD are met when a woman eats at least 
five of the 10 food groups specified. 
As shown in Table 20, fewer than 2% of women consumed a diet that meets the MDD criteria. Figure 3 
shows the distribution of the MDD score. Nearly 69% of women consumed two or fewer food groups. 
The most common food groups consumed were grains (72%) and dark green vegetables (72%). The next 
most common food group, other fruits, is only consumed by 37% of women. Other vegetables, eggs, and 
dairy were consumed by less than 2% of women. A small number of women (78) reported consuming 
none of the groups. This is plausible given the responses to the questions in the FIES discussed 
previously.
IMPEL | Implementer-Led Evaluation and Learning
18 Findings
Table 20. Percent of women of reproductive age consuming a diet of minimum diversity
Control Treatment All
Outcome Percent N Percent N Percent N 
Women 15–49 years with MDD 1.2% 2,069 1.9% 2,010 1.6% 4,079
Women (age 15–18) 0.5% 378 2.6% 342 1.5% 720
Women (age 19+) 1.4% 1,691 1.7% 1,668 1.6% 3,359
*Minimum dietary diversity (MDD)
Figure 3. Dietary diversity scores for women of reproductive age
Table 21 shows the percent of women of reproductive age with a birth in the past five years who 
received at least four antenatal care (ANC) visits from skilled health personnel during their most recent 
pregnancy. The recommendation is that pregnant women receive at least four ANC visits. On average 
53% of these women received at least four ANC visits during their most recent pregnancy. More than 
80% of these visits were with a midwife.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 19
Table 21. Percent of births receiving at least four antenatal care visits during pregnancy
Control Treatment All
Outcome Percent N Percent N Percent N 
ANC of WRA who had a live birth during 
the last 5 years
53.2% 1,292 52% 1,275 52.6% 2,567
* Women of reproductive age (WRA) ages 15-49
*Antenatal care (ANC)
Contraceptive use is not widespread. Table 22 depicts Contraceptive Prevalence Rate (CPR). 
Approximately 13% of non-pregnant women 15–49 years in a union are using birth control, and nearly 
all these women use a modern method. Injectable contraception accounts for more than half of all 
contraception methods used. 
Table 22. Contraceptive prevalence rate (CPR)
Control Treatment All
Outcome Percent N Percent N Percent N 
Non-pregnant aged women 15–49 in 
a union using birth control 11.7% 913 13.7% 871 12.7% 1,784
Non-pregnant women aged 
15–49 in a union using 
modern birth control 11.6% 919 13.7% 877 12.7% 1,784
Non-pregnant women aged 
15–49 in a union traditional 
birth control 0.1% 919 0% 877 0.0% 1,784
As Table 23 shows the percent of women in a union who have knowledge of modern family planning 
methods that can be used to delay or avoid pregnancy. The vast majority of women in a union are aware 
of modern family planning methods. Specifically, women are counted as having knowledge of these 
methods if they are aware of at least three modern family planning methods. Younger women are more 
likely to be aware of these methods than older women (p-value = 0.07).
Table 23. Percent of women in a union who have knowledge of modern family planning methods that 
can be used to delay or avoid pregnancy
Control Treatment All
Outcome Percent N Percent N Percent N 
Women ages 15–49 in a union 74.4% 1,056 76.2% 1,000 75.3% 2,056
Women in a union (ages
15–19) 58.1% 117 75.5% 117 67.4% 234
Women in a union (ages
20–29) 75.8% 476 77.8% 444 76.8% 920
Women in a union (ages
30–49) 76.8% 463 74.7% 439 75.7% 902
Table 24 presents findings on decision-making about family planning. Of those women using modern 
family planning methods, 90% report making that decision.
IMPEL | Implementer-Led Evaluation and Learning
20 Findings
Table 24. Percent of women in a union who made decisions about modern family planning methods 
Control Treatment All
Outcome Percent N Percent N Percent N 
Women ages 15-49 in a union 89.6% 124 89.5% 134 89.5% 258
Women in a union (ages
15–19)
n/a 8 n/a 7 n/a 15
Women in a union (ages
20–29)
88.9% 57 92.6% 72 91.2% 129
Women in a union (ages
30–49)
88.7% 59 86% 55 87.3% 114
3.5 Water, Sanitation, and Hygiene (WASH) Practices
The percent of household using basic drinking water services indicator is defined by three criteria: (1) 
having access to an improved water source, such as a public tap or protected well, (2) having that source 
within 30 minutes round-trip of the home, and (3) having that source available year-round.8 Only 13% of 
all households meet all three of these criteria. While most water sources are available year-round, less
than a third of households have water within 30 minutes and 43% have access to an improved source. 
Table 25. Percent of households using basic drinking water services
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households with access to basic 
drinking water services
9.6% 2,312 15.7% 2,272 12.7% 4,584
F&M 8.9% 1,579 15.0% 1,541 11.9% 3,120
FNM 11.4% 632 18.0% 656 14.8% 1,288
MNF 10.5% 101 12.8% 75 11.5% 176
Percent of households with improved water 
source
42.1% 2,313 42.9% 2,274 42.5% 4,587
F&M 42.7% 1,580 42.5% 1,542 42.6% 3,122
FNM 39.9% 632 44.0% 656 42.1% 1,288
MNF 46.5% 101 41.4% 76 44.3% 177
Percent of households with water source 
within 30 minutes
29.0% 2,315 35.9% 2,272 32.5% 4,587
F&M 28.0% 1,582 36.1% 1,541 32.0% 3,123
FNM 32.1% 632 35.7% 656 34.0% 1,288
MNF 27.8% 101 33.9% 75 30.4% 176
Percent of households with water available 
year-round
84.3% 2,316 80.4% 2,274 82.3% 4,590
F&M 84.0% 1,583 79.7% 1,542 81.9% 3,125
 
8 The final criteria having the water source available year-round includes a measure of the water volume used. 
However, that data is not available for all households. The indicator reported in Table 25 does not include that 
component.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 21
Control Treatment All
Outcome Percent N Percent N Percent N 
FNM 85.7% 632 81.8% 656 83.6% 1,288
MNF 81.6% 101 81.4% 76 81.5% 177
* Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Water use was calculated by asking respondents about the containers they use to carry water and the 
frequency of trips. On average households that treat water report using 8 liters of water per person per 
day9 and there is little difference between treatment and control households (Table 26).
Table 26 provides the full water availability indicator, which includes the additional criteria that 
household water use should be at least 20 liters per person per day. The water use questions were only 
asked of those who treated their water and therefore the full indicator cannot be calculated for all 
households. While nearly 13% had access to basic drinking water services definition without water use, 
this falls to less than 1% when the minimum water use criteria is added. Roughly 4% of households 
responding reported water use of at least 20 liters per person.
Table 26. Water use per capita per day
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Percent of household 
with access to basic 
drinking water services
including water use
0.8% (0.29) 1,086 0.78% (0.31) 1,016 0.77% (0.21) 2,102
F&M 0.8% (0.36) 731 1.00% (0.44) 680 0.91% (0.28) 1,411
FNM 0.7% (0.53) 317 0.28% (0.28) 301 0.50% (0.30) 618
MNF 0 (.) 38 0 (.) 35 0 (.) 73
Water use per capita per 
day (liters)
7.47 (0.28) 1,086 8.28 (0.27) 1,016 7.86 (0.19) 2,102
F&M 7.13 (0.49) 731 8.02 (0.48) 680 7.56 (0.35) 1,411
FNM 7.90 (0.60) 317 8.43 (0.55) 301 8.16 (0.41) 618
MNF 11.18 (1.59) 38 12.40 (1.33) 35 11.77 (1.05) 73
*Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Regarding water treatment and sanitation, only 4% of households were observed to have handwashing 
facilities available in the home as illustrated in Table 27. Only 41% of households report treating water, 
with flocculation and solar disinfection being the most common methods (Table 28).
 
9 Water use was estimated for those treating water. This was done by asking what containers were used to collect water and 
how often water was collected. The volume of the containers was verified by the enumerator by sight.
IMPEL | Implementer-Led Evaluation and Learning
22 Findings
Table 27. Percent of households with soap and water at a handwashing station on premises
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households with 
handwashing available
4.9% 2,039 3.6% 2,038 4.3% 4,077
F&M 4.9% 1,405 4.3% 1,380 4.6% 2,785
FNM 5.0% 545 2.3% 590 3.6% 1,135
MNF 4.4% 89 0.0% 68 2.5% 157
*Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Table 28. Percent of households in target areas practicing correct use of recommended household 
water treatment technologies
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households treating water 43.12% 2,316 39.9% 2,274 41.5% 4,590
Percent of households with treated 
water by adding bleach or chlorine 
before drinking
5.1% 2,316 4.7% 2,274 5.0% 4,590
Percent of households with treated 
water by flocculation before drinking
27.2% 2,316 24.6% 2,274 25.9% 4,590
Percent of households with treated 
water by filtration before drinking
3.5% 2,316 3.3% 2,274 3.4% 4,590
Percent of households with treated 
water by solar disinfection
13.5% 2,316 13.5% 2,274 13.5% 4,590
Percent of households with treated 
water by boiling before drinking
0.3% 2,316 0.4% 2,274 0.4% 4,590
Tables 29 and 30 summarize the use of sanitation facilities, specifically the percent of population in the 
target areas practicing open defecation and percent of households with access to basic sanitation 
service. The majority of households (57%) practice open defecation and only 15% have household-level 
basic sanitation facilities. Among those not practicing open defecation, 83% use a pit latrine without a 
slab, which is considered unimproved, and 23% share the facility with other households. Lack of water 
due to the drought conditions may be affecting household hygiene practices. 
Table 29. Percent of households in target area practicing open defecation
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households practicing open 
defecation
55.8% 2,316 58% 2,274 56.9% 4,590
F&M 54.3% 1,583 58.9% 1,542 56.6% 3,125
FNM 58.8% 632 55.7% 656 57.1% 1,288
MNF 64.0% 101 58.3% 76 61.5% 177
* Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 23
Table 30. Percent of households with access to basic sanitation services 
Control Treatment All
Outcome Percent N Percent N Percent N 
Percent of households with access to basic 
sanitation facilities
14.1% 2,316 16.3% 2,274 15.2% 4,590
F&M 14.0% 1,583 14.6% 1,542 14.3% 3,125
FNM 15.7% 632 20.9% 656 18.4% 1,288
MNF 6.5% 101 10.7% 76 8.3% 177
*Gendered Household Type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
3.6 Agriculture
3.6.1 Crops 
More than 93% of households are engaged in farming (either crops or livestock), and over 96% of 
farmers in the Maharo region own the land they cultivate. The targeted crops identified by the IP are 
cassava, sorghum, and cowpea. As shown in Table 31, cassava and cowpea are widely grown, cultivated 
by 74% and 76% of farmers, respectively. Sorghum is much less common, grown by only 9% of farmers. 
Other crops grown by at least 5% of the population are also included in the table.
Table 31. Crops grown
Control Treatment All
Percent of farmers growing … Percent N Percent N Percent N
Cassava 74.7% 2,069 73.1% 2,090 73.9% 4,159
Sorghum 9.3% 2,069 9.4% 2,090 9.3% 4,159
Cowpea 75.3% 2,069 77% 2,090 76.2% 4,159
Maize 70.6% 2,069 73% 2,090 71.8% 4,159
Peanut 8.2% 2,069 9.9% 2,090 9.1% 4,159
Sweet potatoes 31.8% 2,069 33% 2,090 32.4% 4,159
Melon 13.2% 2,069 12.6% 2,090 12.9% 4,159
Lablab beans 15.8% 2,069 13.9% 2,090 14.9% 4,159
Chickpeas 15.2% 2,069 16.7% 2,090 16.0% 4,159
The use of credit and savings is not common among farmers, and crop insurance is almost non-existent. 
Value chain participation is generally low. One-quarter of farmers purchase crop inputs and fewer than 
2% of farmers purchase inputs for livestock. None of the other practices listed, including the use of 
extension, contract farming, drying or processing produce, and formal marketing for crops or livestock, 
were used by more than 1% of farmers.
IMPEL | Implementer-Led Evaluation and Learning
24 Findings
Table 32. Percent of farmers who used financial services (savings, agricultural credit, and/or 
agricultural insurance) in the past 12 months & Percent of farmers who practiced the value chain 
interventions promoted by the activity in the past 12 months10
Control Treatment All
Outcome Percent N Percent N Percent N
Percent of farmers using agricultural credit 4.8% 2,167 4.6% 2,187 4.7% 4,354
Percent of farmers who saved 3.6% 2,196 2.5% 2,208 3.0% 4,404
Percent of farmers using insurance 0.1% 2,196 0.1% 2,208 0.1% 4,404
Percent of farmers reporting at least one value 
chain activity
23.3% 1,398 21.6% 1,385 22.4% 2,783
Percent of farmers who are in a value chain that 
purchase inputs for crops
21.6% 1,398 18.9% 1,385 20.2% 2,783
Percent of farmers who are in a value chain that 
purchase inputs for livestock
1.9% 1,398 1.8% 1,385 1.8% 2,783
Percent of farmers who are in a value chain that 
use of training and extension services
0.5% 1,398 0.4% 1,385 0.4% 2,783
Percent of farmers who are in a value chain that 
contract farming
0.0% 1,398 0.0% 1,385 0.0% 2,783
Percent of farmers who are in a value chain that 
drying produce
0.7% 1,398 0.8% 1,385 0.8% 2,783
Percent of farmers who are in a value chain that 
processing produce
0.0% 1,398 0.2% 1,385 0.1% 2,783
Percent of farmers who are in a value chain that 
trading, or marketing produce through agro 
dealers and/or community associations
0.3% 1,398 0.4% 1,385 0.3% 2,783
Percent of farmers who are in a value chain that 
use of formal marketing systems for livestock
0.0% 1,398 0.5% 1,385 0.3% 2,783
The agricultural module included a list of agricultural practices of interest to the IP. Tables 33–35
summarize the most common practices used by farmers for each crop.11 The most improved 
management practices/technologies common practice across all three crops was waiting for sufficient 
rain to plant. Intercropping, the use of wind breaks, and crop rotation were also common. However, 
41%, 36%, and 22% of cassava, cowpea, and sorghum farmers, respectively, used none of the listed 
practices.
Table 33. Improved management practices/technologies for cassava
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Percent of farmers 
using at least one 
practice for cassava
60.5% (1.55) 1,536 57.5% (1.54) 1,509 59.0% (1.09) 3,045
 10 The full list of value chain activities in the survey is: Purchase inputs for crops, purchase inputs for livestock, use of training 
and extension services, contract farming, drying produce, processing produce, trading or marketing produce through agro 
dealers and/or community associations, and use of formal marketing systems for livestock 11 Only those practiced by at least 5% of farmers are shown. The full list of practices included in the survey for all three crops is: 
Organic manure, compost, performing weeding, sowing after useful rain, crop association, crop rotation, use of improved 
seeds, use of climate information (rain forecast, disaster risks, etc.), wind break, and soil cover.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 25
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Percent of farmers 
interplanting for 
cassava
36.6% (1.52) 1,564 34.8% (1.50) 1,524 35.7% (1.07) 3,088
Percent of farmers 
sowing after significant 
rain for cassava
27.7% (1.46) 1,564 27.3% (1.44) 1,524 27.5% (1.02) 3,088
Percent of farmers wind 
break for cassava
16.1% (1.15) 1,564 13.4% (1.03) 1,524 14.7% (0.77) 3,088
Percent of farmers that 
did not use modern 
practices one for 
cassava
40.2% (1.54) 1,564 42.6% (1.54) 1,524 41.4% (1.09) 3,088
Table 34. Improved management practices/technologies for cowpea
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Percent of farmers using 
at least one practice for 
cowpea
64.7% (1.56) 1,525 64.4% (1.48) 1,602 64.5% (1.07) 3,127
Percent of farmers 
interplanting for cowpea
46.4% (1.60) 1,532 43.5% (1.51) 1,613 44.9% (1.10) 3,145
Sowing after significant 
rain for cowpea
34.4% (1.54) 1,532 34.2% (1.44) 1,613 34.3% (1.05) 3,145
Wind break for cowpea 16.3% (1.14) 1,532 15.7% (1.09) 1,613 16.0% (0.79) 3,145
Crop rotation for 
cowpea
7.0% (0.80) 1,532 5.4% (0.65) 1,613 6.1% (0.51) 3,145
None for cowpea 35.9% (1.56) 1,532 35.9% (1.47) 1,613 35.9% (1.07) 3,145
Table 35. Improved management practices/technologies for sorghum
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Percent using at least one 
practice for sorghum
85.3% (2.67) 192 72.3% (3.79) 198 78.7% (2.38) 390
Interplanting for sorghum 70.8% (3.53) 193 59.6% (4.05) 200 65.1% (2.73) 393
Sowing after significant rain 
for sorghum
49.8% (4.04) 193 44.1% (4.08) 200 46.9% (2.87) 393
Crop rotation for sorghum 30.2% (3.74) 193 22.0% (3.31) 200 26.0% (2.50) 393
Wind break for sorghum 23.8% (3.55) 193 20.6% (3.02) 200 22.2% (2.33) 393
None for sorghum 14.6% (2.65) 193 28.1% (3.77) 200 21.5% (2.37) 393
Table 36 presents the percent of producers who have applied targeted improved management practices 
or technologies. These practices include purchasing inputs for crops or livestock, use of training or 
extension services, contract farming, drying or processing produce or trading or marketing produce. 
Overall, the majority of farmers (78%) are not implementing any improved practices. Approximately 
one-fifth (21%) are implementing one practice. Less than 1% are implementing two or more.
IMPEL | Implementer-Led Evaluation and Learning
26 Findings
Table 36. Percent of producers who have applied targeted improved management practices or 
technologies 
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Percent of farmers implementing no 
improved practices 77.0% (0.03) 1,413 79.0% (0.03) 1,381 78.0% (0.02) 2,794
Percent of farmers implementing one 
improved practice 22.0% (0.03) 1,413 20.0% (0.03) 1,381 21.0% (0.02) 2,794
Percent of farmers implementing two or 
more improved practices 1.0% (0.00) 1,413 1.0% (0.00) 1,381 1.0% (0.00) 2,794
The yield estimates for the three crops are presented in Table 37. As a result of the prolonged, severe 
drought, most farmers who planted one of the targeted crops reported no production. These farmers 
are omitted from the yield calculation because many of these farmers may have abandoned or stopped 
working their plots well before harvest. Because the yield calculation is based on farmer recall of 
production and the farmer’s estimate of the area planted, these numbers should be taken as rough 
estimates.12
Yield estimates for all three crops are in the range of what would be expected in southern Madagascar 
under drought conditions. Figure 4 shows the distribution of yields. The estimated mean yield for 
cassava is 1,338 kg per hectare for the 867 farmers reporting any production. One study of cassava 
production in southwestern Madagascar cited historical yields in the region from various studies, most 
ranging from 3,000–6,000 kg per hectare.13 Thus, our lower estimate is reasonable given the current 
drought.
The cowpea yield was estimated to be 362 kg per hectare. A technical sheet from GIZ on cowpea 
production in the region estimated yields of 400–1,000 kg per hectare in 2011, Sorghum is not widely 
cultivated and only 65 farmers reported any production. While this small number might raise concerns 
about the yield estimate, 351 kg/ha is within the range of yields (300–1,000) cited by another GIZ 
technical document for the Androy region.
Table 37. Yield of targeted agricultural commodities within target areas
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Percent of 
producers
reporting zero 
cassava production
66.7% (3.58) 1,548 67.5% (3.58) 1,522 67.1% (2.53) 3,070
Producers of 
cassava reporting
yield in kg per ha
1,401.99 (144.37) 458 1,269.36 (131.26) 409 1,337.67 (98.61) 867
 12 The survey first asks farmers about the number of plots, the area of each plot, and how much of each plot was devoted to 
each of the three crops. Next, farmers are asked to recall total production of each crop across all plots in the previous year. 
While most farmers measure land in acres (1/100 of a hectare), many farmers measure production in volume, not weight, and 
this must be converted to kilograms.
13 Jacques Arrivets, Situation actuelle de la culture du manioc dans le Sud-Ouest malgache perspectives d’amélioration, 
(Montpellier, France.: Centre de coopération internationale en recherche, 1996)
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 27
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Percent of 
producers
reporting zero 
cowpea production
73.8% (2.99) 1,529 75.7% (2.44) 1,611 74.8% (1.93) 3,140
Producers of 
cowpea reporting
yield in kg per ha
386.71 (57.02) 304 336.36 (43.34) 320 362.12 (36.38) 624
Percent of 
producers
reporting zero 
sorghum 
production
69.% (3.84) 193 73% (4.58) 211 71.3% (3.04) 404
Producers of 
sorghum reporting 
yield in kg per ha
364.6 (83.72) 37 n/a n/a 28 350.94 (66.16) 65
*Kilogram (kg))
*Hectare (ha)
Figure 4. Box plot of yield estimates for target crops
IMPEL | Implementer-Led Evaluation and Learning
28 Findings
3.6.2 Livestock
The primary targeted livestock are goats and poultry. The summary statistics for goat farmers are in 
Table 38. In our sample, 37% of farmers raised goats. The average herd size for goat farmers was 11 
animals. Goat yield is calculated as the estimated weight of animals sold or consumed divided by the 
herd size.14 The most commonly used modern livestock practices were vaccinations, treatment for 
parasites, and castration.
Table 38. Goat farming
Control Treatment All
Outcome Mean N Mean N Mean N
Percent of all farmers raising goats 37.7% 2,235 37% 2,243 37.4% 4,478
Average herd size for goat farmers 11.41 859 10.57 798 10.99 1,657
Average number of adult male goats
per farmer
2.5 859 2.33 798 2.42 1,657
Average number of adult female goats
per farmer
4.81 859 4.29 798 4.55 1,657
Average number of young male goats
per farmer
1.75 859 1.72 798 1.73 1,657
Average number of young female 
goats per farmer
2.36 859 2.23 798 2.29 1,657
Average livestock weight in kg 208.47 859 192.1 798 200.33 1,657
Goat yield (weight of offtake in 
kg/herd size)
16.61 859 18.94 798 17.77 1,657
Percent of farmers using vaccinations 32.0% 892 26.1% 837 29.0% 1,729
Percent of farmers using anti-parasitic 
treatments
13.5% 892 8.5% 837 11.0% 1,729
Percent of farmers using castration 10.6% 892 6.5% 837 8.6% 1,729
Percent of farmers using none of the 
practices
61.3% 892 69.8% 837 65.6% 1,729
Nearly 48% of farmers raise poultry and on average farmers raised 13 birds in the last year. Only 12% of 
farmers reported any egg production from chickens in the previous week. Farmers are much more likely 
to have sold poultry in the last year (81%) than to have consumed any (21%). A significant share of 
farmers (42%) reported poultry dying in the last year. Vaccinations were the only practice used by more 
than 5% of farmers and most farmers (87%) did not use any of the listed practices. Poultry yield is 
calculated as the total weight of poultry sold and consumed divided by the number raised.15
 14 Weights are based on the "Fiche Technique Chaîne de Valeur Caprin/Ovin," developed for the regions of Anosy and Androy, 
which estimates weights of 1–2 year old goats and sheep to be 15–30 kg. Based on this, 25kg is used for adult males, 20 for 
adult females, 15 for young males and 10 for young females. 15 Based on the following, 2 kilograms is used for the weight of poultry. 
https://agritrop.cirad.fr/585447/7/Fiches%20produits_march%C3%A9s%20%20VF%2020_10_2017.pdf
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 29
Table 39. Poultry farming
Control Treatment All
Outcome Mean N Mean N Mean N
Percent of all farmers raising poultry 47.2% 2,235 48% 2,243 47.6% 4,478
Number of poultry raised by farmers in the last 
year
12.58 1,123 13.06 1,134 12.83 2,257
Percent of farmers reporting eggs in previous week 12.4% 1,122 12.5% 1,131 12.4% 2,253
Percent of farmers selling poultry in the last year 81.3% 1,123 81% 1,134 81.2% 2,257
Percent of farmers consuming own poultry in the 
last year
18.9% 1,121 23% 1,133 21% 2,254
Percent of farmers experiencing poultry dying in 
the last year
40.7% 1,122 43.9% 1,132 42.4% 2,254
Poultry yield in kg per bird 1.41 (0.17) 1,123 1.24 (0.03) 1,134
Percent of farmers vaccinating 13.3% 1,066 12.8% 1,086 13.1% 2,152
Percent of farmers not using any of the poultry 
practices
86.6% 1,066 86.6% 1,086 86.6% 2,152
Some households in the coastal areas in the region fish, and the IP is interested in developing this 
sector. In our sample, however, only 246 individuals fished.16 More than half of the communes have no 
fishermen. One-quarter of fishermen are in the commune of Itampolo. The vast majority of fishermen 
fish for both own consumption and for sale (Table 40).
Table 40. Fishing
Control Treatment All
Outcome Percent N Percent N Percent N
Percent of fishers fishing for food only 12.7% 108 4.6% 138 7.7% 246
Percent or fishers fishing for market only 4.7% 108 0.8% 138 2.3% 246
Percent of fishers fishing for both food and 
market
82.6% 108 94.6% 138 90% 246
Percent of fishers using a pirogue 43.1% 108 63.6% 136 55.6% 244
Percent of fishers using nets 39.6% 108 60.9% 136 52.6% 244
Percent of fishers using containers 30.3% 108 34.7% 136 33% 244
Table 41 presents the average number of cattle and sheep owned by households.17 On average, 
households own 0.41 cattle and 0.29 sheep.
 16 It is possible that this is because the survey was conducted during the rainy season when many fishermen self-identify as 
farmers and because the survey question was interpreted as whether one currently fishes. 17 Note that sheep and cattle are captured in the section on household productive assets and were not included in the livestock 
section so we do not have disaggregated information about male versus female sheep and cattle.
IMPEL | Implementer-Led Evaluation and Learning
30 Findings
Table 41. Livestock ownership
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Average number of cattle
per household
0.435 (0.06) 2,312 0.391 (0.05) 2,276 0.413 (0.03) 4,588
Average number of sheep
per household 
0.267 (0.04) 2,311 0.309 (0.04) 2,277 0.288 (0.02) 4,588
3.7 Poverty Measurement
This section presents three measures of poverty, all based on household expenditure. The measures are 
daily per capita expenditure, the percent living on less than $1.90 per day (2011 PPP) and the depth of 
poverty of the poor. The equivalent of $1.90 was determined to be 2,443 Malagasy Ariary.18
The poverty measures are summarized in Table 42. The mean per capita expenditure is 1,272 Ariary, or 
less than $1 per day. The poverty rate is approximately 90%. The depth of poverty of the poor is 57%, 
which means that the average poor person is 57% below the poverty line. In monetary terms, this 
means it would require an additional $1.08 per person per day to bring every poor person out of 
poverty.
Table 42. Poverty measures
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Daily per capita 
expenditure (Ariary)
per household
1,243.12 (23.77) 2,317 1,302.02 (21.83) 2,278 1,272.47 (16.15) 4,595
Daily per capita 
expenditure (2011 
Dollars) per 
household
0.97 (0.02) 2,317 1.01 (0.02) 2,278 0.99 (0.01) 4,595
Prevalence of 
Poverty: Percent of 
people living on less 
than $1.90/day
89.9% (0.80) 2,317 89.7% (0.74) 2,278 89.8% (0.54) 4,595
Depth of Poverty of 
the Poor: Mean 
percent shortfall of 
the poor relative to 
the $1.90/day 2011 
PPP poverty line
58.1% 0.00 2,007 55.2% 0.00 1,944 56.7% (0.03) 3,951
Table 43 presents measures further broken down by household type. Interestingly, expenditure is higher 
and poverty is lower among both female-headed households and male-headed households compared to 
those households with both adult males and adult females present (p-value = 0.00). There has been 
 18 The expenditure module of the survey followed standard practices for expenditure calculation. Frequent items, primarily 
foods, used 7-day recall. Less frequent purchases used 30-day or 12-month recall. A rental equivalent is used to value housing 
and durable goods. Enumerators reported a few extreme cases of people with essentially zero expenditures who lived on what 
they could collect, forage, or receive from begging.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 31
significant migration out of this region and the difference could be due to remittances. The BL survey, 
however, did not include direct questions on remittances. Remittances were listed as an option in 
response to a question about how households coped with shocks, but only 2% of households cited this 
as a strategy.
Table 43. Disaggregated poverty measures by household type
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Prevalence of 
Poverty: Percent of 
people living on 
less than 
$1.90/day
89.9% (0.80) 2,317 89.7% (0.74) 2,278 89.8% (0.54) 4,595
F&M 91.6% (0.88) 1,584 91.8% (0.83) 1,545 91.7% (0.61) 3,129
FNM 83.9% (1.96) 632 82.7% (1.69) 657 83.3% (1.29) 1,258
MNF 85.0% (3.89) 101 86.0% (3.80) 76 85.4% (2.76) 177
Daily per capita 
expenditure 
(Ariary)
1,243.12 (23.77) 2,317 1,302.02 (21.83) 2,278 1,272.47 (16.15) 4,595
F&M 1,191.42 (26.81) 1,584 1,253.66 (24.81) 1,545 1,222.27 (18.30) 3,129
FNM 1,436.01 (53.97) 632 1,464.09 (50.15) 657 1,450.53 (36.70) 1,289
MNF 1,307.26 (135.53) 101 1,427.8 (120.11) 76 1,357 (93.70) 177
Daily per capita 
expenditure 
(Dollars)
0.97 (0.02) 2,317 1.01 (0.02) 2,278 0.99 (0.01) 4,595
F&M 0.93 (0.02) 1,584 0.98 (0.02) 1,545 0.95 (0.01) 3,129
FNM 1.12 (0.04) 632 1.14 (0.04) 657 1.13 (0.03) 1,289
MNF 1.02 (0.11) 101 1.11 (0.09) 76 1.06 (0.07) 177
*Gendered household type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no Adult 
Female (MNF)
The 2011 PPP used was 700.228 Ariary/dollar. The Consumer Price Index used for 2011 was 72.11 and 
132.5 for 2020. This gives us the poverty line = 1. 9 * 700.228 * 132.5 / 72.18 = 2,243.
3.8 Gender Dynamics
Gender dynamics are captured through the eight indicators in this section. Because the survey targeted 
only men and women in a union, they are the only ones included in this section. These indicators 
explore male and female financial resources, access to credit, and female decision-making within the 
households.
3.8.1 Use of Financial Resources
This section presents findings on participation in cash earning activities across men and women in a 
union as well as women’s participation in decisions around their own incomes and their husbands’ 
IMPEL | Implementer-Led Evaluation and Learning
32 Findings
incomes. As illustrated in Table 44, approximately half of men and women in a union (52%) participated 
in a cash earning activity in the past year. It is far more common, however, for men in a union (71%) to 
participate in cash earning activities than for women in a union (33%) to do so (p-value = 0.00). Fewer 
young adult women (ages 15–19) are participating in cash earning activities relative to older women 
(20–49) (p-value = 0.16). A majority of men across all age brackets are participating in cash earning 
activities.
Table 44. Percent of women and men in a union who earned cash in the past 12 months
Control Treatment All
Outcome Percent N Percent N Percent N
Cash Earners in a union 51.8% 2,607 52.2% 2,543 52% 5,150
Women in a union (ages 15–49) 33.5% 1,299 32.3% 1,268 32.9% 2,567
Women in a union (ages 15–19) 19.3% 125 26.9% 132 23.4% 257
Women in a union (ages 20–29) 30.2% 508 30.5% 482 30.4% 990
Women in a union (ages 30–49) 39.3% 483 32.7% 468 35.9% 951
Men in a union (15+) 69.8% 1,308 72% 1,275 70.9% 2,583
Men in a union (ages 15–19) 66.4% 17 49.2% 13 58.1% 30
Men in a union (ages 20–29) 76.8% 307 72.2% 339 74.2% 646
Men in a union (ages 30+) 67.7% 984 72.4% 923 69.9% 1,907
As shown in Table 45, among women who are earning cash, a large majority (82%) reported that they 
participate in decisions about how to use the cash, whether solely or jointly with others. Women
participate in decisions about the use of cash income at similar rates across ages.
Table 45. Percent of women in a union and earning cash who report participation in decision about 
the use of self-earned cash
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Cash earning 
women in a union
(ages 15–49)
81.1% (0.038) 456 82.8% (0.026) 416 82.0% (0.019) 872
Women in a 
union (ages 
15–19)
53.8% (0.086) 32 75.7% (0.091) 35 67.4% (0.069) 67
Women in a 
union (ages 
20–29)
84.2% (0.036) 171 87% (0.037) 151 85.7% (0.021) 322
Women in a 
union (ages 
30–49)
83.6% (0.039) 197 82.3% (0.029) 157 83.0% (0.024) 354
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 33
Table 46 highlights the percent of cash earning men in a union who report that their spouse participates 
in decision-making about the use of self-earned cash. Approximately 75% of men perceive their wives to 
be involved in decision-making.
19
Table 46. Percent of men in union and earning cash who report spouse/partner participation in 
decisions about the use of self-earned cash
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Cash Earning Men in 
a union (ages 15+)
73.6% (0.028) 913 75.3% (0.024) 929 74.5% (0.017) 1,842
Men in a 
union (ages 
15–19)
n/a n/a 12 n/a n/a 9 n/a n/a 21
Men in a 
union (ages 
20–29)
70.6% (0.044) 230 74.2% (0.037) 262 72.6% (0.028) 492
Men in a 
union (ages 
30+)
75.0% (0.029) 671 75.6% (0.024) 658 75.3% (0.018) 1,329
3.8.2 Credit
This section presents findings on the use of credit among men and women in a union. Table 47 shows 
that approximately one-third (36%) of men and women in a union borrowed in the previous 12 months. 
Women and Men in a union report borrowing at similar rates across any source. It is far less common for 
men and women in a union to borrow from microfinance or lending based groups (3%).
Table 47. Percent of women/men in a union who used credit in the previous 12 months
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Men and women in a 
union (lending groups)
3.0% (0.01) 1,213 3.0% (0.01) 1,208 3.0% (0.01) 2,421
Men and women in a 
union (any source)
36.8% (0.03) 2,356 35.7% (0.02) 2,328 36.2% (0.02) 4,684
Men in a union (ages 
15–49)
37.2% (0.03) 1,143 36.5% (0.03) 1,159 36.8% (0.02) 2,302
Men in a union
(ages 15–19)
n/a n/a 14 n/a n/a 13 n/a n/a 27
Men in a union
(ages 20–19)
33.6% (0.04) 276 41.7% (0.03) 315 38.2% (0.03) 591
Men in a union
(ages 30+)
38.1% (0.03) 853 33.9% (0.03) 831 36.1% (0.02) 1,684
Women in a union
(ages 15–49)
36.4% (0.02) 1,213 34.9% (0.02) 1,169 35.6% (0.02) 2,382
 19 The sample for BL 34 indicator (Percent of women in a union and earning cash who report spouse/partner 
participation in decisions about the use of self-earned cash), was not going to be reflective of all cash earning women in a union 
due to a survey coding error, the research team decided to excluded the indicator from the report.
IMPEL | Implementer-Led Evaluation and Learning
34 Findings
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Women in a 
union (ages 
15–19)
22.0% (0.05) 116 25.5% (0.04) 113 23.8% (0.03) 229
Women in a 
union (ages 
20–29)
36.2% (0.03) 474 41.2% (0.03) 447 38.7% (0.02) 921
Women in a 
union (ages 
30–49)
40.7% (0.03) 452 34.0% (0.04) 434 37.3% (0.02) 886
Of the men and women in a union who report borrowing, almost two-thirds (66%) of them participate in 
decisions about credit. As illustrated in Table 48, men in a union participate in decisions about credit at a 
much higher rate than women in a union (p-value = 0.00), with 89% of men reportedly making credit 
decisions, which is consistent across age groups.
Table 48. Percent of women/men in a union who make decisions about credit
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Men and women in a 
union 
66.2% (0.019) 887 65.4% (0.015) 821 65.8% (0.012) 1,708
Men in a union (ages 15+) 89.3% (0.023) 441 88.8% (0.018) 419 89.0% (0.012) 860
Men in a union
(ages 15–19)
n/a n/a 9 n/a n/a 5 n/a n/a 14
Men in a union
(ages 20–29)
92.5% (0.029) 102 91.2% (0.028) 128 91.7% (0.021) 230
Men in a union
(ages 30+)
88.7% (0.024) 330 87.1% (0.022) 286 88.0% (0.013) 616
Women in a union (ages 
15-49)
43.5% (0.037) 446 41.4% (0.036) 402 42.5% (0.025) 848
Women in a 
union (ages 
15–19)
30.9% (0.087) 33 34.3% (0.094) 34 32.8% (0.058) 67
Women in a 
union (ages 
20–29)
45.4% (0.043) 174 39.0% (0.044) 171 42.0% (0.030) 345
Women in a 
union (ages 
30–49)
40.8% (0.048) 178 43.8% (0.053) 154 42.2% (0.037) 332
3.8.3 Additional Decision-Making Areas
This section presents findings on the input women have in their household, disaggregated by age and 
household type. The questions were broken down into the following categories: nutrition, health, 
agriculture, livestock/fisheries, savings and lending, women’s empowerment, and WASH. Women would 
rate on a scale of 1 to 4 the level of input they had in making decisions about those different categories. 
If women averaged between a 3 and 4, they were considered to have a medium to high level of input.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 35
Table 49 shows that almost half of women have a medium or high input when making decisions in their 
home (50%). Table 50 shows which categories of decisions women have medium or high input into. The 
large majority of women (94%) have medium or high input into agency related decisions. It is far less 
common for women to have medium or high input into savings and lending decisions, where 21% of 
women have this level of input.
Table 49. Women in a union who report medium or high amount of household input, age
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Women in a union 
(ages 15+)
47.7% (0.031) 1,250 51.9% (0.022) 1,181 49.8% (0.019) 2,431
Women in a 
union (ages 
15–19)
37.7% (0.045) 115 46.8% (0.060) 113 42.5% (0.040) 228
Women in a 
union (ages 
20–29)
45.1% (0.042) 483 48.4% (0.033) 453 46.7% (0.024) 936
Women in a 
union (ages 
30–49)
50.4% (0.038) 469 56% (0.034) 441 53.2% (0.023) 910
Women in a 
union (ages 
50+) 
52.9% (0.064) 183 53.9% (0.045) 174 53.4% (0.039) 357
Table 50. Women in a union who report medium or high amount of household input, categories
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Nutrition 48.7% (0.04) 913 50.7% (0.03) 926 49.7% (0.02) 1,839
Health 86.9% (0.02) 1,176 89.7% (0.01) 1,119 88.3% (0.01) 2,295
Agriculture 41.1% (0.02) 1,083 44.4% (0.02) 1,053 42.8% (0.01) 2,136
Livestock 50.0% (0.03) 1,099 53.2% (0.02) 1,067 51.6% (0.02) 2,166
Saving and Lending 22.0% (0.04) 327 19.5% (0.03) 353 20.7% (0.02) 680
Agency 94.4% (0.01) 1,223 93.8% (0.01) 1,153 94.1% (0.01) 2,376
Water and 
Sanitation
65.5% (0.02) 1,146 66.7% (0.03) 1,094 66.1% (0.01) 2,240
Table 51 highlights women in a union’s input into household decisions, disaggregated by the gender of 
the head of household. Women in female-headed households reported having a similar level of input 
into household decisions (56%) relative to women in male-headed households (50%). However, there 
are only 71 cases in which the head of the household is a woman. These are most likely women whose 
husbands are not currently present.
IMPEL | Implementer-Led Evaluation and Learning
36 Findings
Table 51. Women in a union who report medium or high amount of household input, household 
head's sex
Control Treatment All
Outcome Percent SE N Percent SE N Percent SE N
Women in a union 47.7% (0.031) 1,250 51.9% (0.022) 1,181 49.8% (0.019) 2,431
Female Head 
of Household
68.4% (0.086) 40 39.9% (0.099) 31 56.1% (0.070) 71
Male Head of 
Household
47.0% (0.031) 1,210 52.2% (0.023) 1,150 49.6% (0.019) 2,360
3.9 Resilience
3.9.1 Ability to Recover from Shocks and Stresses Index
The ability to recover from shocks and stresses index reflects the ability to recover from negative events 
that have impacted the household. On average, households score 2.2 on this index. This index is 
composed of indices that reflect how households perceived their recovery as well as the total number 
and severity of shocks the household experienced over the past year. Table 52 below illustrates that 
there was little to no observable difference across treatment and control household responses (see 
Annex A for statistical comparisons). There is also little variation in the perceived ability to recover 
across households with both male and female adults present.
Table 52. Ability to recover from shocks and stresses
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household ability to 
recover from shocks & 
stresses index
2.178 (0.047) 2,012 2.237 (0.051) 1,989 2.208 (0.019) 4,001
F&M 2.183 (0.048) 1,376 2.232 (0.049) 1,338 2.208 (0.022) 2,714
FNM 2.181 (0.061) 548 2.229 (0.058) 583 2.207 (0.022) 1,131
MNF 2.070 (0.058) 88 2.394 (0.122) 68 2.209 (0.055) 156
Household ability to 
recover index (2–6)
2.179 (0.048) 2,013 2.242 (0.052) 1,991 2.211 (0.019) 4,004
F&M 2.184 (0.048) 1,377 2.238 (0.050) 1,339 2.211 (0.022) 2,716
FNM 2.182 (0.061) 548 2.235 (0.059) 584 2.210 (0.022) 1,132
MNF 2.082 (0.058) 88 2.392 (0.124) 68 2.214 (0.054) 156
Household total shocks 
experienced (0–15)
3.145 (0.071) 2,317 3.112 (0.067) 2,278 3.128 (0.035) 4,595
F&M 3.185 (0.072) 1,584 3.132 (0.075) 1,545 3.159 (0.037) 3,129
FNM 3.085 (0.088) 632 3.043 (0.071) 657 3.063 (0.046) 1,289
MNF 2.831 (0.154) 101 3.298 (0.191) 76 3.033 (0.122) 177
Household shock 
exposure index (0–120)
20.846 (0.493) 2,311 20.322 (0.424) 2,268 20.582 (0.231) 4,579
F&M 21.057 (0.493) 1,579 20.322 (0.476) 1,540 20.690 (0.252) 3,119
FNM 20.528 (0.616) 632 20.245 (0.530) 652 20.379 (0.325) 1,284
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 37
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
MNF 19.249 (1.088) 100 21.046 (1.160) 76 20.029 (0.746) 176
*Gendered household type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
The perceived ability to recover index reflects a household’s perspective on their ability to meet food 
needs now, relative to the previous year, as well as their ability to meet their future food needs. 
Households score an average of 2.2 (range of 2–6). In other words, households perceived their ability to 
meet their current needs as worse than the previous year and suspect that their future ability to meet 
these needs will deteriorate. Households experienced an average of 3.1 out of 15 possible shocks in the 
previous year. The most commonly reported shocks are illustrated below in Figure 5, which shows the 
shocks experienced by at least 10% of households. Nearly all households (98%) listed the drought as one 
of these shocks. Other common shocks listed included rising food prices and crop pests.
In terms of the number and intensity of shocks experienced, households score an average of 20.6 out of 
120, which accounts for 15 shocks and four different levels of severity (regarding both the impact on the 
household economic situation and impact on household consumption).20 This suggests that out of the 
average of 3.1 shocks experienced, households perceived those shocks to be severe. This perception is 
likely driven by the ongoing extreme drought which has pushed many households to the brink and likely 
explains households’ perceived low ability to recover.
Figure 5. Most common shocks reported by households
 20 Note that the SEI was created out of 15 possible shocks instead of 16 given available data.
97.8
77.5
55.1
26.5
16.1 13.1 10.1
0
20
40
60
80
100
Percent of Households Experiencing Shock
IMPEL | Implementer-Led Evaluation and Learning
38 Findings
3.9.2 Social Capital Index
This index conveys the ability of households to draw on social networks to get support to reduce the 
impact of shocks and stresses on their households. It measures both the degree of bonding—social 
capital among households within their own communities—and the degree of bridging—social capital 
between households in the area and households outside their own community. On average, households 
score 40.4 out of 100,
21 indicating that their ability to draw on their networks for support is moderate 
(Table 53). There is little observable difference in this score across treatment and control households. 
Similarly, F&M, FNM and MNF households report similar scores.
Table 53. Index of social capital at the household level
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household index of social 
capital at household level 
(0–100)
40.357 (1.348) 2,317 40.507 (1.370) 2,278 40.433 (0.532) 4,595
F&M 40.965 (1.419) 1,584 40.346 (1.434) 1,545 40.655 (0.605) 3,129
FNM 39.462 (1.371) 632 40.596 (1.518) 657 40.059 (0.684) 1,289
MNF 35.641 (3.460) 101 43.293 (3.203) 76 38.952 (2.275) 177
Household bonding sub￾index (0–100)
43.708 (1.320) 2,317 43.742 (1.231) 2,278 43.725 (0.547) 4,595
F&M 44.158 (1.412) 1,584 43.522 (1.327) 1,545 43.84 (0.643) 3,129
FNM 43.125 (1.398) 632 44.004 (1.351) 657 43.588 (0.704) 1,289
MNF 39.702 (3.811) 101 46.248 (2.989) 76 42.534 (2.400) 177
Household bridging sub￾index (0–100)
37.007 (1.542) 2,317 37.273 (1.615) 2,278 37.141 (0.611) 4,595
F&M 37.771 (1.602) 1,584 37.17 (1.674) 1,545 37.47 (0.671) 3,129
FNM 35.798 (1.618) 632 37.188 (1.841) 657 36.53 (0.835) 1,289
MNF 31.581 (3.412) 101 40.339 (3.655) 76 35.37 (2.356) 177
* Gendered household type: Female and Male Adults (F&M), Adult Female no Adult Male (FNM), Adult Male no 
Adult Female (MNF)
Findings suggest that households are able to draw on households within their community slightly more 
(average score of 43.7) than they are able to draw on households outside of their community (average 
score of 37.0).
3.9.3 Absorptive Capacity Index
The absorptive capacity index reflects the ability of households to prepare for, deal with, and mitigate 
the impact of shocks and stressors on well-being outcomes through preventive measures and positive 
coping strategies. Overall, households score 33.6 out of 100 on this index, reflecting a low ability to 
absorb shocks, an ability that has likely been hampered by the ongoing drought. Below we discuss the 
results of the components of the absorptive capacity index. As discussed below, beyond humanitarian 
 21 We followed precedent with TANGO and did not exclude factors that loaded negatively on the first component in the 
construction of this index. This is because we want the indices to be comparable across baseline and EL (where at EL the factors 
that load negatively may be different). However, the two scores are comparable on average.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 39
assistance and some community organizations, there are few resources that households have access to 
that would enable them to better mitigate shocks. 
A key aspect of the absorptive capacity index captures financial resources that households have access 
to in order to absorb shocks. Overall, very few households have access to financial resources for 
absorbing shocks, lowering their ability to mitigate the impact of shock on well-being outcomes. 
Specifically, only 3% of households have cash savings, 2% have reported receiving remittances and 1% 
have access to insurance. Households on average own 5.8 different types of household and productive 
assets (out of 16), although this does not necessarily mean that households have a large asset stock.
Table 54. Absorptive capacity index
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household absorptive 
capacity index (0–100)
33.626 (0.737) 2,130 33.496 (0.711) 2,091 33.561 (0.364) 4,221
Household access to cash 
savings index (0–1)
0.034 (0.008) 2,313 0.029 (0.007) 2,277 0.031 (0.004) 4,590
Household asset 
ownership index - total 
type (0–31)
5.793 (0.108) 2,290 5.747 (0.111) 2,246 5.770 (0.063) 4,536
Household remittances 
index (0–1)
0.022 (0.004) 2,272 0.015 (0.003) 2,214 0.019 (0.003) 4,486
Household access to 
insurance index (0–1)
0.001 (0.001) 2,176 0.001 (0.001) 2,141 0.001 (0.001) 4,317
Household bonding social 
capital index (0–6)
1.751 (0.050) 2,317 1.752 (0.049) 2,278 1.752 (0.021) 4,595
Household access to 
informal safety nets index 
(0–6)
2.494 (0.168) 2,317 2.484 (0.163) 2,278 2.489 (0.042) 4,595
Household shock 
preparedness & 
responsiveness index 
(0–3)
0.792 (0.051) 2,317 0.794 (0.054) 2,278 0.793 (0.014) 4,595
Household access to 
humanitarian assistance 
index (0–1)
0.813 (0.029) 2,317 0.826 (0.025) 2,278 0.820 (0.015) 4,595
The absorptive capacity index also captures the level of social capital that households have access to in 
order to help them absorb shocks. Overall, findings suggest that most households have low social 
capital. The bonding social capital index reflects the number of types of individuals households could 
draw on the inside of their communities (out of six groups). On average, households feel able to draw on 
1.8 of these types of individuals. Moreover, households reported that they have moderate access to, 
and have been active in, community organizations that typically serve as informal safety nets. On 
average, households have access to 2.5 of six types of safety nets.
IMPEL | Implementer-Led Evaluation and Learning
40 Findings
Another element of absorptive capacity is how well a household is prepared to mitigate shocks22
through the availability of disaster preparedness groups in the community, as well as other household 
shock mitigation strategies. On average, households score a 0.79 out of 3 on this index, suggesting a 
lower ability to mitigate shocks.
The last dimension of absorptive capacity is availability of humanitarian assistance in the community. A 
large majority of households (82%) reported that they have received emergency food or cash assistance 
from the government or a non-governmental organization (NGO).
23 The presence of humanitarian 
assistance might be heightened by the ongoing drought. Nevertheless, it is one of the few absorptive 
capacity resources that households currently have access to.
3.9.4 Adaptive Capacity Index
The adaptive capacity index measures the ability of households to manage resources and make pro￾active and informed choices in order to better prepare for and adapt to future shocks. The index is 
composed of several components that reflect different resources or adaptive abilities. On average, 
households score a 39.7 out of 100 on this index,
24 which suggests that households have a limited ability 
to manage resources and adapt to future shocks. Below we discuss the results of the components of the 
adaptive capacity index. Similar to the absorptive capacity index, the low score on the adaptive capacity 
index likely reflects the ongoing humanitarian emergency situation. Households across the treatment 
and control groups perform similarly on this score.
Table 55. Adaptive capacity index
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household adaptive 
capacity index (0–100)
39.604 (0.790) 1,946 39.765 (0.738) 1,959 39.686 (0.420) 3,905
Household bridging social 
capital index (0–6)
1.488 (0.062) 2,317 1.498 (0.065) 2,278 1.493 (0.025) 4,595
Household linking social 
capital (0–4)
0.035 (0.008) 2,317 0.036 (0.007) 2,278 0.035 (0.005) 4,595
Household social 
network index (0–6)
1.758 (0.087) 2,317 1.810 (0.082) 2,278 1.784 (0.025) 4,595
Household 
education/training index 
(0–8)
0.750 (0.053) 2,308 0.726 (0.052) 2,271 0.738 (0.035) 4,579
Household asset 
ownership index - total 
type (0–31)
5.793 (0.108) 2,290 5.747 (0.111) 2,246 5.770 (0.063) 4,536
 22 This index does not include whether household reports participating in any of the following activities: soil conservation 
activities, flood diversion structures (i.e., protection of land/infrastructure from flooding), planting trees on communal land, or 
improving access to health services given available data.
23 Note that this index does not capture whether NGO/government assistance is available in their community but they have not 
received it.
24 We followed precedent with TANGO and did not exclude factors that loaded negatively on the first component in the 
construction of this index. This is because we want the indices to be comparable across BL and EL (where at EL the factors that 
load negatively may be different). However, the two scores are comparable on average.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 41
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household access to 
financial resources (0–2)
0.799 (0.069) 2,317 0.804 (0.062) 2,278 0.801 (0.019) 4,595
Household livelihood 
diversification index 
(0–20)
1.958 (0.049) 2,317 2.033 (0.057) 2,278 1.996 (0.028) 4,595
Household adoption of 
improved practices index 
(0–1)
0.110 (0.017) 2,176 0.123 (0.019) 2,141 0.117 (0.008) 4,317
Household exposure to 
information index (0–19)
2.436 (0.179) 2,317 2.485 (0.183) 2,278 2.460 (0.077) 4,595
Household 
aspirations/confidence to 
adapt index (0–16)
8.773 (0.132) 2,082 8.815 (0.113) 2,084 8.794 (0.070) 4,166
One aspect of the adaptive capacity index captures households’ social capital and networks, given that 
households that are able to leverage these networks more effectively may better prepare for and adapt 
to future shocks. Across these indicators, households score poorly, suggesting that households are not 
able to effectively leverage social capital and networks in order to adapt to shocks. In particular, the 
bridging social capital index reflects the number of types of individuals that households could draw on 
outside of their communities (out of six groups). On average, households feel able to draw on 1.5 of 
these types of individuals. The linking social capital index reflects how well-connected households are to 
government or NGO leaders and whether they can draw on them for help. Households score very low 
(0.04 out of 4) on this suggesting that the majority of households neither know leaders nor are they able 
to ask leaders for help. Finally, the social network index captures household access to and participation 
in various support groups. Households score a 1.8 out of 6 suggesting that only a minority of households 
have access to and/or participate in these groups.
Another aspect of the adaptive capacity index captures the human resources, assets and financial 
resources available to households in order to mitigate shocks. Overall, households have low levels of 
human capital and asset resources, suggesting constraints on the overall resource pool they are able to 
draw on in the face of shocks. The education/training index reflects the level of human capital in the 
household, specifically adult literacy, whether any adult has surpassed primary school, and the number 
of trainings household adults have participated in. Households score very low (0.8 out of 8) on this 
indicator, reflecting that overall household human capital is low. The asset ownership index illustrates 
the number of different types of assets a household owns (out of 31 types). On average, a household 
owns 5.8 different types of assets. This could mean that overall household asset stock is low, although 
this does not reflect the value of each asset. Finally, the access to financial resources index reflects the 
financial resources available in the village through credit and savings institutions. Households have on 
average 0.8 out of two of these institutions available to them.
A third aspect of the adaptive capacity index reflects how diversified and improved household livelihood 
activities are. In summary, household activities are not well-diversified and few households have 
adopted improved practices. The livelihood diversification index reflects the number of different 
livelihood activities households were engaged in over the past year. Overall, households were engaged 
in an average of 2.0 out of 20 activities indicating that activities are not well-diversified. The adoption of 
IMPEL | Implementer-Led Evaluation and Learning
42 Findings
improved practices index25 reflects whether households adopted improved crop or livestock practices, 
natural resource management practices, or improved storage practices. Overall, households score 0.11 
out of 1 indicating that only a minority of households have adopted improved practices. The exposure to 
information index captures the number of topics that households have received information on in the 
past year, which relates directly to a household’s ability to make informed choices in order to better 
prepare for shocks. On average, households have received information on 2.5 out of 19 available topics, 
highlighting that households have had limited exposure to information to help inform shock mitigation 
strategies.
Finally, the aspirations/confidence to adapt index reflects a household adult’s aspirations, confidence to 
adapt, and a sense of control over her life. On average, adults score 8.8 out of 16 on this index reflecting 
a moderate sense of confidence to adapt.
3.9.5 Transformative Capacity Index
The transformative capacity index26 captures system-level resources, governance, and institutions that 
comprise the enabling environment which promotes or limits a household’s capacity to respond to 
shocks and stressors. On average, households score 70.3 out of 100 on this index27 indicating that there 
are moderately strong institutions available to enhance household capacity to respond to shocks. 
Households across the treatment and control groups perform similarly on this score. Below we discuss 
the results of the components of the transformative capacity index.
Table 56: Transformative capacity index
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household 
transformative capacity 
index (0–112)
70.101 (3.431) 2,153 70.459 (3.600) 2,154 70.283 (0.711) 4,307
Household access to 
infrastructure index (0–3)
0.958 (0.060) 2,317 0.953 (0.056) 2,278 0.955 (0.011) 4,595
Household basic services 
index (0–4)
2.542 (0.081) 2,317 2.581 (0.081) 2,278 2.561 (0.021) 4,595
Household access to 
markets index (0–1)
0.91 (0.026) 2,317 0.899 (0.030) 2,278 0.904 (0.008) 4,595
Household access to 
communal natural 
resources index (0–4)
0.486 (0.077) 2,317 0.480 (0.074) 2,278 0.483 (0.023) 4,595
 25 This index was created differently given available data. We create a binary variable if respondents report three or more 
improved crop or livestock practices in total versus a binary if respondents report three or more crop practices or three or more 
livestock practices.
26 Note that this index does not include the following sub-indices given data availability: access to livestock services and 
collective action.
27 We followed precedent with TANGO and did not exclude factors that loaded negatively on the first component in the 
construction of this index. This is because we want the indices to be comparable across BL and EL (where at EL the factors that 
load negatively may be different). For the case of the transformative index, the index that adjusts for negative factors is 35.11 
on average across the sample. The main driver between these scores is that the infrastructure factor gets more weight in the 
score we show here.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Findings 43
Control Treatment All
Outcome Mean SE N Mean SE N Mean SE N
Household access to 
agricultural services index
(0–1)
0.257 (0.055) 2,317 0.287 (0.056) 2,278 0.272 (0.018) 4,595
Household bridging social 
capital index (0–6)
1.488 (0.062) 2,317 1.498 (0.065) 2,278 1.493 (0.025) 4,595
Household linking social 
capital index (0–4)
0.035 (0.008) 2,317 0.036 (0.007) 2,278 0.035 (0.005) 4,595
Household social 
cohesion index (0–3)
0.109 (0.037) 2,317 0.120 (0.037) 2,278 0.114 (0.014) 4,595
Household local decision￾making index (0–1)
0.228 (0.027) 2,153 0.259 (0.027) 2,154 0.244 (0.013) 4,307
Household gender index 
(0–3)
1.953 (0.021) 2,317 1.955 (0.021) 2,278 1.954 (0.008) 4,595
Household gender 
equitable decision￾making index (0–2)
0.705 (0.036) 2,317 0.718 (0.030) 2,278 0.712 (0.010) 4,595
Household access to 
formal safety nets index 
(0–13)
3.047 (0.150) 2,317 3.178 (0.171) 2,278 3.113 (0.050) 4,595
Household local 
government 
responsiveness index 
(0–2)
0.064 (0.037) 2,317 0.068 (0.041) 2,278 0.066 (0.004) 4,595
One dimension of the transformative capacity index is the accessibility of infrastructure and services in 
the community. While few communities have key infrastructure, most have access to basic services. The 
access to infrastructure index28 reflects how many types of key infrastructure (electricity grid, piped 
water, mobile phone service) are available in the community. On average, households have access to 
0.96 out of three of these types of key infrastructure. The basic services index29 illustrates the number of 
services (police force, primary schools, health and financial services) that are available in a community. 
Households have access to 2.6 out of four of these services on average.
The next dimension of the transformative capacity index is the availability of economic institutions to 
support livelihoods. Access to these economic institutions is varied. The majority (90%) of households 
have access to markets.30 Only one-quarter (27%), however, report having access to agricultural 
extension services.
31 Few households have access to natural communal resources: on average 
households have access to only 0.48 out of four natural communal resources (communal grazing land, 
water source, firewood, and irrigation source).
 28 As access to roads is not available, this index was calculated using the other three service types instead.
29 This index does not include a measure of quality for each of the service types. Instead, it only captures whether or not the 
service exists. Health services reflect whether NGOs are currently conducting health activities and not whether local health 
institutions are available. 30 We assumed that any community in which a household reported selling to a local market had access to a market.
31 This index was calculated based on percentage using agricultural extension services versus those with access given available 
data.
IMPEL | Implementer-Led Evaluation and Learning
44 Findings
Another aspect of the transformative capacity index reflects the strength of households to support 
themselves through their networks. Overall, the ability for households to draw on their networks is low. 
In particular, the bridging social capital index reflects the number of types of individuals that households 
could draw on outside of their communities (out of six groups). On average, households feel able to 
draw on 1.5 of these types of individuals. The linking social capital index reflects how well-connected 
households are to government or NGO leaders and whether they can draw on them for help. 
Households score very low (0.04 out of 4) on this suggesting that the majority of households either do 
not know leaders or are not able to ask leaders for help. The social cohesion index32 illustrates how 
active households have been in various support groups in the village. On average, households report 
engaging in 0.11 out of three support groups reflecting that participation in support groups is not 
common. Finally, the local decision-making index reflects how actively households participate in groups 
in their communities. About one-quarter of households (24%) report active participation.
Another dimension captured by the transformative capacity index is the extent to which there are 
gender-related barriers in the community. Overall, there seem to be a moderate number of gender￾related barriers in the community. The gender index reflects constraints to gender-neutral behavior at 
the community level. On average, communities report 2.0 out of three gender-neutral behaviors are 
norms. The gender equitable decision-making index33 reflects how equitable decision-making is across 
male and female adults within the same household. On average, households score 0.71 out of 2 on this 
index reflecting that out of two key household decisions, on average 0.71 involve both male and female 
household members.
A final dimension of the transformative capacity index is how available and reliable external sources of 
support are. Overall, households have access to a low number of these external resources. The formal 
safety nets index reflects the number of external safety nets (e.g., emergency food or cash assistance, 
agricultural inputs) available in the community. Overall, households have access to 3.1 out of 13 formal 
safety nets. The government responsiveness index reflects whether households have access to a reliable 
police force and a peace committee.34 On average, households have access to 0.07 of two of these 
resources, indicating that very few households have access at all.
 
32 This index ranges from 0–3 instead of 0–4 as we do not have data on whether community members came together for social 
events.
33 This index does not include measures of equitable decision-making around nutrition and child health as well as around 
savings. Thus, the index ranges from 0–2 instead of 0–4. 34 This indicator also corresponds to BL 24% of households that believe local government will respond effectively to future 
shocks and stresses.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Comparison of Treatment and Control Groups 45
COMPARISON OF TREATMENT AND CONTROL GROUPS 
Based on the data collected in the BL round, the research can evaluate the overall comparability of the 
treatment and control groups. This is done by comparing the mean values of a range of demographic 
and household level characteristics and identifying any trends of statistically significant differences 
between the two. The table below illustrates the results of this exercise and confirm that the treatment 
and control group are, overall, balanced. There are no statistically significant differences in means 
between the two household groups at BL. Furthermore, a joint test of orthogonality demonstrates that 
the BL characteristics do not predict treatment status (p value = 0.365). Additional balance tables 
between treatment and control groups are in Annex A of this report.
Table 57. Household roster balance table
Control Treatment
Outcome Mean N Mean N Difference P￾value
Average age of people in the 
household roster
17.455 12,664 17.515 12,248 0.06 0.861
Percent of females in the household 
roster
52.409 12,665 52.488 12,248 0.079 0.916
Percent of farmers in the household 
roster
44.956 5,134 46.163 4,992 1.207 0.569
Percent of people with at least some 
schooling in the household roster
52.128 9,550 52.308 9,207 0.18 0.96
Percent of people who worked for cash
in the roster
41.334 6,245 41.586 6,112 0.252 0.909
Percent of households with adult male 
and female present in the roster
70.059 2,317 68.712 2,278 -1.347 0.489
Percent of households with adult 
female only present in the roster
25.828 2,317 28.209 2,278 2.381 0.212
Percent of households with adult male 
only present in the roster
4.113 2,317 3.078 2,278 -1.035 0.122
Household Head
Average age of heads of households in 
the roster
42.417 2,317 42.337 2,274 -0.081 0.923
Percent of female head of households 
in the roster
36.817 2,319 38.702 2,275 1.885 0.344
Percent that did not attend school in 
the roster 
68.922 2,319 67.894 2,275 -1.029 0.754
Percent of people in the roster with 
some schooling, less than primary
15.002 2,319 16.878 2,275 1.876 0.262
Percent of people in the roster that 
completed Primary or more
15.923 2,315 15.105 2,272 -0.818 0.755
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.365
IMPEL | Implementer-Led Evaluation and Learning
46 Conclusion
CONCLUSION
The indicators presented in this report from the BL survey for the Maharo RFSA reflect the severe food 
security crisis in southern Madagascar. Nearly 90% of the population is living on less than $1.90 per day. 
Nearly all households cite the ongoing drought as one of the major shocks the household has faced. 
When asked how their ability to meet food needs has changed in the last 12 months, 95% said their 
situation is worse today and 78% said they foresee things getting even worse in the coming year. FCSs 
and dietary diversity for women and children are all extremely low. Dietary diversity scores of zero, zero 
reported consumption, and zero agricultural production are outcomes that would be considered 
reporting errors in most surveys but were verified by supervisors in the field.35 Adaptive and absorptive 
resilience capacities to mitigate these shocks are indicated to be low.
The results underscore the need for the types of livelihood support being provided by Maharo’s 
activities. The survey design and results from the EL survey will provide the opportunity to assess the 
impact of additional activities beyond emergency aid. However, there are significant challenges ahead 
for the evaluation.
First, a significant number of households have migrated out of the area or have migrated internally and 
may continue to do so. This suggests that there will be significant attrition at EL. Furthermore, this type 
of attrition is particularly problematic because it is driven by the same factors the IE is tasked with 
evaluating.
Second, the treatment areas appear to be somewhat better off than control areas at BL. For example, 
expenditure is slightly higher, while depth of poverty and prevalence of underweight and severely 
underweight children are lower, and more children have a MAD in the treatment areas. However, as 
shown in the balance tables in the annex, most of these differences are not statistically significant when 
controlling for the clusters and matched pairs. Furthermore, the advantage of the RCT design with BL 
data is that we are able to control for these differences and they should not pose a problem for the 
evaluation.
Third, because of the ongoing crisis in the region, there are other organizations working in these areas 
providing emergency aid and other assistance. These other interventions may make it difficult to isolate 
the effect of the additional CRS programming in treatment areas.
Finally, the effect of agricultural interventions will be limited if the drought persists. Roughly 70% of 
farmers reported having no production and investments in the sector are unlikely to show positive 
results at EL unless conditions improve.
 
35 A 24 hour period without reported food consumption is not surprising given the extreme drought conditions. 
This was further validated by the coping strategies people reported which includes begging and relying on charity. 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex A: Balance Tables 47 
ANNEX A: BALANCE TABLES
Table 58. Food security
Control Treatment
Outcome Mean N Mean N Difference P-value
Household FCS (0–112) 24.869 2,241 25.842 2,223 0.973 0.351
Household poor consumption score (<22) 47.175 2,241 43.076 2,223 -4.099 0.286
Household borderline consumption score (22–35) 35.781 2,241 36.645 2,223 0.864 0.793
Household acceptable consumption score (>35) 17.044 2,241 20.279 2,223 3.235 0.379
Households worried about not having enough 
food to eat because of a lack of money or other 
resources
99.637 2,241 99.07 2,223 -0.567 0.174
Households unable to eat healthy and nutritious 
food because of a lack of money or other 
resources
99.142 2,241 99.132 2,223 -0.009 0.99
Households that ate only a few kinds of foods 
because of a lack of money or other resources
99.309 2,241 99.106 2,223 -0.203 0.692
Households that skipped a meal because there 
was not enough money or other resources
94.012 2,241 92.735 2,223 -1.276 0.421
Households that ate less than you thought you 
should because of a lack of money or other 
resource
98.771 2,241 97.136 2,223 -1.635 0.06
Household that did not have food because of a 
lack of money or other resources
82.579 2,241 80.194 2,223 -2.385 0.423
Households that are hungry but did not eat 
because there was not enough money or other 
resource
83.54 2,241 80.423 2,223 -3.117 0.331
Households that went without eating for a whole 
day because of a lack of money or other resource
79.202 2,241 75.13 2,223 -4.072 0.269
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.567
Table 59. Child nutrition and health
Control Treatment
Outcome Mean N Mean N Difference P-value
MDD children ages 6–23 months 0.024 814 0.038 823 0.015 0.404
MAD children ages 6–23 months 0.018 814 0.028 823 0.011 0.44
Exclusively breastfed under 6 months of age 0.389 222 0.384 200 -0.005 0.948
Children under 5 children who had diarrhea 0.347 2,682 0.341 2,468 -0.006 0.831
Children under 5 who had diarrhea and were 
given ORS
0.145 875 0.133 817 -0.012 0.734
IMPEL | Implementer-Led Evaluation and Learning
48 Annex A: Balance Tables
Table 60. Anthropometry
Control Treatment
Outcome Mean N Mean N Difference P-value
Weight-for-age Z-score -1.642 2,900 -1.589 2,848 0.053 0.225
Children underweight (ages 
0–59 months)
37.686 2,900 34.932 2,848 -2.754 0.078
Children severely underweight
(ages 0–59 months)
15.552 2,900 14.152 2,848 -1.4 0.221
Table 61. Women's health, maternal nutrition, and reproductive health
Control Treatment
Outcome Mean N Mean N Difference P-value
Women of reproductive age (15–49) with MDD 0.015 2,084 0.019 2,024 0.004 0.677
Women in a union using birth control 0.116 913 0.138 871 0.021 0.484
Women of reproductive age who had a live 
birth during the last five years that received
ANC during last pregnancy
0.66 1,057 0.637 1,072 -0.024 0.572
Women of reproductive age in a union who 
have knowledge of modern family planning 
methods
0.884 2,274 0.89 2,213 0.006 0.77
Women of reproductive age in a union who use 
a modern family planning method in the last 12 
months who made decisions about modern 
family planning methods in the past 12 months
0.896 124 0.895 134 -0.001 0.978
Table 62. WASH
Control Treatment
Outcome Mean N Mean N Difference P￾value
Households with access to basic drinking 
water services
9.612 2,312 15.71 2,272 6.098 0.174
Percent of households with improved water 
source
42.141 2,313 42.905 2,274 0.765 0.923
Water source within 30 minutes per 
household
28.971 2,315 35.886 2,272 6.916 0.266
Water available year-round per household 84.314 2,316 80.36 2,274 -3.954 0.124
Handwashing available per household 4.875 2,039 3.638 2,038 -1.236 0.374
Households that treat their water 43.119 2,316 39.889 2,274 -3.23 0.571
Households that treat water by adding bleach 
or chlorine before drinking
5.116 2,316 4.687 2,274 -0.428 0.737
Households that treat water by flocculation 
before drinking
27.215 2,316 24.574 2,274 -2.641 0.497
Households that treat water by filtration 
before drinking
3.455 2,316 3.256 2,274 -0.199 0.832
Households that treat water by solar 
disinfection
13.522 2,316 13.484 2,274 -0.038 0.988
Households that treat water by boiling before 
drinking
0.313 2,316 0.42 2,274 0.106 0.628
Households practicing open defecation 55.84 2,316 57.956 2,274 2.117 0.661
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex A: Balance Tables 49
Control Treatment
Outcome Mean N Mean N Difference P￾value
Households using improved sanitation 
facilities
14.14 2,316 16.266 2,274 2.127 0.461
Household water use per capita (liters) 7.471 1,086 8.276 1,016 0.805 0.182
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.940
Table 63. Agriculture—cassava, sorghum, and cowpea
Control Treatment
Outcome Mean N Mean N Difference
P￾value
Farmers using at least one practice for 
cassava
60.538 1,536 57.536 1,509 -3.002 0.518
Farmers sowing after significant rain for 
cassava
27.739 1,564 27.291 1,524 -0.449 0.915
Farmers interplanting for cassava 36.569 1,564 34.844 1,524 -1.725 0.72
Farmers wind break for cassava 16.067 1,564 13.352 1,524 -2.715 0.439
Farmers not using improved practices for 
cassava
40.149 1,564 42.555 1,524 2.406 0.598
Farmers using at least one practice for 
sorghum
85.262 192 72.338 198 -12.924 0.056
Farmers sowing after significant rain for 
sorghum
49.809 193 44.076 200 -5.733 0.594
Farmers interplanting for sorghum 70.847 193 59.609 200 -11.239 0.177
Farmers using crop rotation for sorghum 30.178 193 22.021 200 -8.157 0.471
Farmers using wind break for sorghum 23.813 193 20.556 200 -3.257 0.685
Farmers not using improved practices for 
sorghum
14.618 193 28.084 200 13.465 0.046
Farmers using at least one practice for 
cowpea
64.701 1,525 64.347 1,602 -0.354 0.931
Farmers sowing after significant rain for 
cowpea
34.454 1,532 34.169 1,613 -0.285 0.941
Farmers interplanting for cowpea 46.403 1,532 43.453 1,613 -2.95 0.556
Farmers using crop rotation for cowpea 6.95 1,532 5.365 1,613 -1.585 0.553
Farmers using wind break for cowpea 16.342 1,532 15.7 1,613 -0.642 0.863
Farmers not using improved practices for 
cowpea
35.85 1,532 35.91 1,613 0.06 0.988
Farmers using agricultural credit 4.825 2,167 4.575 2,187 -0.25 0.842
Farmers who saved 3.573 2,196 2.517 2,208 -1.057 0.243
Farmers using insurance 0.067 2,196 0.095 2,208 0.029 0.806
Farmers reporting at least one value chain 
activity
25.674 1,460 24.113 1,448 -1.561 0.72
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.000
IMPEL | Implementer-Led Evaluation and Learning
50 Annex A: Balance Tables
Table 64. Agriculture–yield
Control Treatment
Outcome Mean N Mean N Difference P-value
Farmers reporting zero cassava 
production
66.744 1,548 67.515 1,522 0.771 0.71
Farmers reporting cassava yield 1,401.987 458 1,269.36 409 -132.627 0.327
Farmers reporting zero cowpea 
production
73.841 1,529 75.727 1,611 1.886 0.319
Farmers reporting cowpea yield 386.71 304 336.364 320 -50.346 0.284
Farmers reporting zero sorghum 
production
69.395 193 73.051 211 3.656 0.474
Farmers reporting sorghum yield 364.615 37 334.596 28 -30.018 0.807
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.068
Table 65. Poverty measurements
Control Treatment
Outcome Mean N Mean N Difference P-value
Poor per household 89.896 2,317 89.739 2,278 -0.157 0.918
Daily per capita expenditure (Ariary) per 
household
1,243.118 2,317 1,302.017 2,278 58.899 0.356
Daily per capita expenditure (Dollars)
per household
0.967 2,317 1.013 2,278 0.046 0.356
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.325
Table 66. Use of financial resources
Control Treatment
Outcome Mean N Mean N Difference P-value
Women and men in a union who earned cash 
in the past 12 months
0.518 2,607 0.522 2,543 0.004 0.924
Women in a union earning cash who report 
participation in decision about the use of 
self-earned cash
0.811 456 0.828 416 0.017 0.753
Men in a union earning cash who report 
spouse/partner participation in decisions 
about the use of self-earned cash
0.736 913 0.753 929 0.017 0.665
Table 67. Credit
Control Treatment
Outcome Mean N Mean N Difference P-value
Men and women in a union who have access to 
credit
0.368 2,356 0.357 2,328 -0.01 0.791
Men and Women in a union who report making 
the borrowing decision
0.662 887 0.654 821 -0.008 0.731
Table 68. Female household input
Control Treatment
Outcome Mean N Mean N Difference P-value
All women in a union 0.477 1,250 0.519 1,181 0.042 0.299
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.299
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex A: Balance Tables 51
Table 69. Resilience
Control Treatment
Outcome Mean N Mean N Difference
P￾value
Household adaptive capacity index (0–100) 39.604 1,946 39.765 1,959 0.161 0.899
Household bridging Social Capital index (0–6) 1.488 2,317 1.498 2,278 0.011 0.926
Household linking social capital (0–4) 0.035 2,317 0.036 2,278 0.001 0.937
Household social network index (0–6) 1.758 2,317 1.81 2,278 0.052 0.747
Household education/training index (0–7) 0.75 2,308 0.726 2,271 -0.025 0.753
Household asset ownership index—total 
type (0–31)
5.793 2,290 5.747 2,246 -0.046 0.8
Household access to financial resources 
(0–2)
0.799 2,317 0.804 2,278 0.005 0.968
Household livelihood diversification index 
(0–20)
1.958 2,317 2.033 2,278 0.075 0.407
Household adoption of improved practices 
index (0–1)
0.11 2,176 0.123 2,141 0.013 0.673
Household exposure to information index 
(0–19)
2.436 2,317 2.485 2,278 0.049 0.88
Household aspirations/confidence to 
adapt index (0–16)
8.773 2,082 8.815 2,084 0.041 0.84
Household absorptive capacity index (0–100) 33.626 2,130 33.496 2,091 -0.13 0.917
Household access to cash savings index (0–1) 0.034 2,313 0.029 2,277 -0.005 0.723
Household asset ownership index—total 
type (0–31)
5.793 2,290 5.747 2,246 -0.046 0.8
Household remittances index (0–1) 0.022 2,272 0.015 2,214 -0.007 0.165
Household access to insurance index (0–1) 0.001 2,176 0.001 2,141 0 0.804
Household bonding Social Capital index (0–6) 1.751 2,317 1.752 2,278 0.001 0.991
Household access to informal safety nets 
index (0–6)
2.494 2,317 2.484 2,278 -0.01 0.976
Household shock preparedness & 
responsiveness index (0–3)
0.792 2,317 0.794 2,278 0.002 0.985
Household access to humanitarian 
assistance index (0–1)
0.813 2,317 0.826 2,278 0.014 0.76
Household transformative capacity index 
(0–100)
66.712 2,153 67.294 2,154 0.582 0.93
Household access to formal safety nets index 
(0–13)
3.047 2,317 3.178 2,278 0.131 0.668
Household access to markets index (0–1) 0.91 2,317 0.899 2,278 -0.012 0.833
Household access to communal natural 
resources index (0–4)
0.486 2,317 0.48 2,278 -0.006 0.969
Household basic services index (0–4) 2.542 2,317 2.581 2,278 0.039 0.803
Household access to infrastructure index 
(0–3)
0.958 2,317 0.953 2,278 -0.006 0.961
Household access to agricultural services 
index (0–1)
0.257 2,317 0.287 2,278 0.03 0.777
IMPEL | Implementer-Led Evaluation and Learning
52 Annex A: Balance Tables
Control Treatment
Outcome Mean N Mean N Difference
P￾value
Household bridging Social Capital index (0–6) 1.488 2,317 1.498 2,278 0.011 0.926
Household linking social capital (0–4) 0.035 2,317 0.036 2,278 0.001 0.937
Household social cohesion index (0–3) 0.109 2,317 0.12 2,278 0.011 0.868
Household gender equitable decision￾making index (0–2)
0.705 2,317 0.718 2,278 0.013 0.832
Household local decision-making index (0–1) 0.228 2,153 0.259 2,154 0.031 0.512
Household local government 
responsiveness index (0–2)
0.064 2,317 0.068 2,278 0.005 0.951
Household gender index (0–3) 28.922 2,317 28.673 2,278 -0.249 0.788
Household ability to recover from shocks & 
stresses index
2.178 2,012 2.237 1,989 0.059 0.521
Household index of social capital at 
household level (0-100)
40.357 2,317 40.507 2,278 0.15 0.952
*Denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. The p-value for a joint test of orthogonality is 0.703
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex B: Risk Mitigation Plan 53 
ANNEX B: RISK MITIGATION PLAN
Project Title
Maharo Impact Evaluation
Principal Investigators
Dr. Christine Moser
Christine.moser@wmich.edu
Reimar Macaranas
Causal Design Inc.
Reimar@causaldesign.com
Research Assistant
Asabea Amaniampong
Causal Design Inc.
Asabea.amaniampong@causaldesign.com
Version Date
This protocol was developed on December 30, 2020 and updated on June 22, 2021
Context
Per the goals and objectives of improving food security and promoting well-being and welfare among 
beneficiary communities, the Maharo RFSA activities aim to provide a range of support services to a 
large population of households in Southern Madagascar. The Maharo RFSA activities will include but are 
not limited to the provision of food assistance as well as maternal health support for mothers and their 
children, particularly in the early stages of childhood development. In addition to these services, the 
RFSA will also include a range of tailored livelihood support activities. While general RFSA support will be 
delivered unconditionally in the region of focus, livelihood activities will be designed to be targeted to 
specific communities. The delineation between communities that receive a RFSA package activities with 
and without the additional livelihood support will allow for the ability to design an experimental impact 
evaluation to estimate the marginal benefit of these livelihood activities. Experimental impact 
evaluation to estimate the marginal effect of the "full" RFSA package in this context.
Evaluation Design
The evaluation team will conduct a clustered RCT to estimate the marginal impact of the Maharo RFSA 
intervention on food security and other well-being indicators in selected communities. The approach will 
use statistical analysis to estimate the direct impact of livelihood support activities on communities 
receiving RFSA support using standard BHA food security and health indicators36. The design will 
maximize the ability for the research to measure direct and attributional impacts and will employ 
 
36 https://www.usaid.gov/food-assistance/partner-with-us/implementation-and-reporting 
IMPEL | Implementer-Led Evaluation and Learning
54 Annex B: Risk Mitigation Plan
statistical tools and methodologies, specifically randomization, ex-ante matching, and regression 
analysis.
Identification Strategy
Implementer-defined clusters: Following an in-country field visit and evaluability assessment in January 
2020, village or fokontany-level randomization was deemed infeasible because (1) villages or fokontany 
within the target area are too small to practically implement within, and (2) the definitions of fokontany 
in this area are dynamic and highly subject to change. The commune is the next highest level of 
administration. However, there are only 20 communes in the Maharo intervention areas—fewer than 
would be required for statistical power—ruling this out as an enumeration area for the purposes of 
randomization. CRS proposed defining “implementation clusters” composed of groups of fokontany that 
could be served by a CRS team. CRS identified 218 of these clusters in the target area, each with an 
average of 3 fokontany per cluster, to be used as the sampling frame. The “implementation clusters”
approach should ensure that a statistically adequate number of treatment and control clusters can be 
assigned.
Randomization Strategy 
CD will use a two-stage randomization strategy to select final treatment and control communities and 
households. On the cluster level, CD will use a matched pair randomization approach (rather than 
stratified random sampling) to ensure better balance prior to BL data collection. In the matched pair 
randomization, units will first be matched based on variables related to outcomes. In other words, 
clusters that look “similar” based on available data will be paired together. One unit from each pair will 
then be randomly assigned to treatment and the other to control. This process will then be used to 
generate matched pairs that will be utilized in the randomization exercise. CRS has provided information 
on the estimated number of households, number of potential beneficiaries, access to a river, access to a 
market, proximity to the coast, and presence of a clinic for each of the 217 clusters. Based on these data 
we formed 98 pairs of clusters for the matched-pair randomization. This has resulted results in 98 
treatment areas and 98 control areas with 21 unassigned areas where CRS is free to implement as it 
sees fit. Prior to assignment, the CRS team verified that the matches were reasonable based on 
knowledge of the areas.
The second stage of randomization will involve selecting fokontany within clusters and household within 
each fokontany. CRS conducted a household census in its intervention area in August 2020. Using this 
data, we will randomly sample 1-4 fokontany per cluster, based on the total population of each 
intervention cluster. We will then sample 12 households per fokontany.
Inclusion and Exclusion Criteria 
Inclusion criteria will be solely based on residence in Maharo RFSA program’s intervention communities 
that have been designated as treatment or control areas. The primary desired respondent for the 
quantitative surveys will be adult (ages 18 and older) financial decision makers of the home, particularly 
individuals most responsible for purchasing or approving the purchase of home or livelihood assets. 
Where household heads are not available to participate in the survey, enumerators will be instructed to 
consult the spouse or secondary member most responsible for purchasing assets for the home or 
primary livelihood.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex B: Risk Mitigation Plan 55
The project will not include the following special populations:
• Adults unable to consent.
• Individuals who are not yet adults (minors): infants, children, teenagers
• Prisoners or other detained individuals.
As the study and data collection activity pose little physical risk to adult participants, the research 
reserves the right to interview consenting pregnant women should they fit the inclusion criteria 
described above.
Apart from the vulnerable populations highlighted above, the project will not impose specific exclusion 
criteria. All beneficiaries identified as eligible for compensation; a bar of soap will be offered.
Community and Government Coordination
The unique competent authority to issue a license to conduct surveys and research to subjects in 
Madagascar is the Committee of Ethics of Biomedical Research at the Ministry of Public Health (CERBM). 
After presenting the study to this Committee by submitting all the necessary documents (including 
questionnaires) and after analyzing all the documents submitted, the committee concluded that it is not 
biomedical research and does not require a special authorization to conduct it.
Data Protection Processes
Protection of privacy and household information are key concerns for the both the implementation 
team and the associated researcher. Currently all data collection done in the service of monitoring and 
implementing the existing program is subject to USAID security protocols. To minimize any vulnerability 
and risks of data breach, the research team will work to ensure that shared datasets are fully 
anonymized and cannot be linked directly to participating households.
Confidentiality and Data Management
The research team will rely on existing protocols governing data privacy and confidentiality. All 
monitoring and survey data will continue to be housed on password protected and encrypted servers 
using the Survey CTO platform.
Provisions to Protect and Privacy Interest of Participants
When transferring data from field surveys to the Principal Investigators for ensuing analysis, no names 
or means for direct identification will be included. Instead, randomized household ID numbers will be 
assigned to participants and. As a result, no researcher outside the original research staff will have the 
ability to directly link data and information to households in the study. 
While the questionnaire and data collection activities do not pose any substantial physical risk or collect
overly sensitive data, numerous steps will be taken to ensure to participants are protected and not 
exposed to any undue risk from data collection activities or enumerators. Data collection enumerators 
will be required to participate in training exercises to ensure compliance with research protocols and 
observance of risk mitigation processes. Additionally, supervisors will be dispatched with each team of 
enumerators to ensure adherence to data collection protocols. Spot checks will also be conducted on 
anonymized datasets by the research team at the outset of each of the data collection periods to help 
ensure data validity.
IMPEL | Implementer-Led Evaluation and Learning
56 Annex B: Risk Mitigation Plan
Quantitative interviews are estimated to take about one hour to complete, while interviews are 
scheduled to last between 45 minutes to 1.5 hours. Data will be analyzed through the scheduled end of 
the project. 
Quantitative data collection will be conducted using electronic tablets using questionnaires designed 
through ODK (Open Data Kit) and collected using the Survey CTO platform.
Process to Document Consent (Waiver Requested)
The research conducted will obtain verbal consent from clients before undertaking research activities 
and including the respondent’s answer in the study. The research team requests a waiver to collecting 
signatures from agreeing participants for the following reasons.
A waiver of informed consent is being requested due to the nature of the research. The research team 
will be analyzing data with no confidential or identifying information. Furthermore, the research activity 
involves little to no risk to beneficiaries as it does little to change the way that normal program. 15 
implementations would be done, apart from introducing different schedules for receive cash benefits. 
Finally, documentation of informed consent, if required, would be the only information linking the 
participant to the research and would introduce the only means for potentially compromising 
beneficiary privacy. Further explanation and rationale is given in an attached document requested by 
Solutions IRB. By collecting verbal consent, beneficiaries are shielded from all vulnerability of exposure 
posed by the research linking specific households to the study. Security and data management protocols 
are already in place and mandated by donor agencies for regular monitoring purposes and data is stored 
in password protected and encrypted servers using the Survey CTO platform. By anonymizing data 
before it is given to the research team, privacy concerns and risks are minimized further.
The secondary reason is that a substantial number of beneficiaries are potentially illiterate. While the 
project would stress that receipt of financial support would be in no way linked to participation in the 
study, administering a written form for beneficiaries to sign would create undue suspicion surrounding 
their receipt of funds. 
The actions of the research team amount to researchers effectively working with an existing 
administrative dataset. While identifying data will be collected for basic monitoring purposes, that 
information would be collected regardless of the research and is governed by standing protocols and 
agreements with the United States Agency for International Development (USAID). Data shared with the 
research team will be fully anonymized meaning that there would effectively be little to no risk to 
beneficiary confidentiality as a result of participation in the study.
The following language will be conveyed to all research participants at all rounds of data collection:
Title of Research: IMPEL Maharo Impact Evaluation 
Investigators: Christine Moser, Reimar Macaranas
Purpose of the Research
Hello, my name is [ENUMERATOR]. I am working with Consultant Associates in Madagascar on 
behalf of Causal Design, a US company that does research on programs that support 
communities in Madagascar. We are speaking with households in your area to learn more about 
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex B: Risk Mitigation Plan 57
how families and communities support themselves and respond to challenges. Your household 
was selected to be interviewed to learn how you have been affected by different challenges and 
how your family responded to them. Households like yours were selected because they live in 
an area where Catholic Relief Services, CRS, is delivering the Maharo program and its activities. 
Explanation of Procedures
The interview may take up between 1 hour and 2 hours to finish. We will ask to interview 
different members of your household. We will ask many things about your daily life in the 
survey and hope that none of the questions make you or your family members uncomfortable. If 
I ask you any questions you or your family members don't want to answer, let me know and I 
will go on to the next question. You can also stop the interview at any time. If you decide not to 
participate in this survey or if you withdraw from participating at any time, you will not be 
punished in any way. Your privacy is important to us. We hope you will agree to answer the 
questions since your views and experiences are important. Choosing to participate will not do 
anything to change the services and support you receive from the Maharo program. 
Risks
There is little risk to you and your household for sharing your experiences. We have taken care 
to protect you and your household against the COVID-19 virus. We will wear masks during the 
interview and maintain a 2-meter distance from you and the people in your household. We will 
be weighing children during this survey and to make sure that we keep you and your household 
safe we have cleaned the scales, washed our hands with disinfectant, and we will still have our 
masks on. Your responses will help organizations like CRS better understand how to improve 
programs and provide support in the future.
Benefits
There is no direct benefit to your household for sharing your experiences. Your responses will 
help organizations like CRS better understand how to improve programs and provide support in 
the future. At the end of the interview, we will provide a small gift, a small packet of soap, for 
your participation. 
Confidentiality
If you agree to participate, the information will be entered into a database that will be used to 
understand how people in your community live their lives and do daily activities. Some of the 
information you provide will be available on a public website that researchers and others will be 
able to access, but all personal information will be kept confidential and not made public. There 
will be no audio or video recording of this interview. We will ask you for personal information 
now to make sure that we can visit you at the end of the project about 5 years from now to ask 
you the same types of questions but will remove all personal identifying information after we 
have finished collecting your responses at the end of the research. Any data attached to your 
personal information will be stored in a password protected electronic format. 
IMPEL | Implementer-Led Evaluation and Learning
58 Annex B: Risk Mitigation Plan
Questions
Do you have any questions about the survey or what I have said? If in the future you have any 
questions regarding the survey or the interview, or concerns or complaints, we welcome you to 
contact Consultant Associates or Causal Design. You can reach us at info@causaldesign.com or 
(720) 260-4837.
If you have questions about your rights as a research participant or concerns or complaints 
about the research, you may contact the Solutions IRB at (855) 226-4472. This organization is 
based in the United States. Regular hours for Solutions IRB are 8:00 a.m. to 5:00 p.m. CT, 
Monday through Friday. You can also email them at participants@solutionsirb.com
We will leave a copy of this statement and our organization’s complete contact information with 
you so that you may contact us at any time. Do you agree to participate in the survey?
Incomplete Disclosure or Deception
The study will not attempt to use any means of incomplete disclosure or deception in its data collection 
efforts.
Recruitment
Participants will be recruited based on availability and satisfaction of eligibility criteria.
Risk to Participants
The study team foresees that the overall risks to participants of the research study is low. There is little 
to no danger of physical strain given that data collection is primarily through self-reported answers. The 
bulk of questions deal with common behaviors around household decision making, interactions with 
other community members, and information around livelihood decisions. Participation is not required 
and households that refuse to take part in the research will still be allowed to receive humanitarian and 
livelihood services. There is the potential risk that for some, discussions around group dynamics may be 
uncomfortable. The participant will be reminded that no answer is required and that he/she is free to 
not give a response to any specific question.
Should a participant desire to withdraw from the interview, enumerators will be instructed to end the 
discussion, thank the participant for their time and exit the household. Data collected from withdrawn 
interviews will be excluded from analysis in the study.
Additionally, any critical findings relevant to individual well-being and/or health will immediately be 
shared with local authorities and with households.
COVID-19 Considerations
Any unexpected vulnerability or dangers that the research may have exposed participants to at any 
point during the study, including incidental exposure to the coronavirus will be communicated through 
local channels. A detailed COVID-19 safety protocol has also been developed and subsequently reviewed 
and approved by USAID staff to minimize the risk to households.
Potential Benefits to Participants
The research provides no foreseeable channels for providing additional benefits to respondents or their 
families outside of original intervention.
Baseline Study of the Maharo RFSA in Madagascar (Vol. I)
Annex B: Risk Mitigation Plan 59 
Financial Compensation
Survey respondents will not receive financial compensation in exchange for providing survey responses. 
The data collection team is authorized, however, to provide a nominal gift (approx. value 1-2 USD) to 
recognize the time given by the household to the research.
Qualifications to Conduct Research and Resources Available 
Causal Design is experienced in research methods and protocols and has completed learning modules on 
ethics and proper principles around human subject testing, research, and confidentiality. Additionally, all 
Principal Investigators associated with the project have similar certification. These are in line with 
current HIPAA standards and was administered by the Collaborative Institutional Training Initiative 
(CITI). Human Research Completion Reports for all Principal Investigators will be provided to the IRB for 
verification and can be provided upon request.