Impact Assessment Report DryDev Final

The Drylands Development Programme
(DRYDEV)
A Farmer-led Programme to Enhance Water Management, Food

Security, and Rural Economic Development in the Drylands of Burkina
Faso, Mali, Niger, Ethiopia, and Kenya
Impact Assessment Report
Prepared for: The Ministry of Foreign Affairs, the Netherlands

Prepared by: The World Agroforestry (ICRAF)
Submission date: September 2020
Table of Contents
Executive summary ....................................................................................................................................... 4
1. Introduction .......................................................................................................................................... 6
2. Impact Evaluation Strategy ...................................................................................................................... 8
3. Impact and Outcome Indicators ........................................................................................................ 10
4. Data Collection and Analysis.................................................................................................................. 12
4.1. Data collection and cleaning ............................................................................................................ 12
4.2. Data processing and construction of indicators ............................................................................... 15
4.2.1. Crop production and sales estimates ........................................................................................ 15
4.2.2. Dietary diversity index ............................................................................................................... 16
4.2.3. Women’s Empowerment in Agriculture Index (WEAI) .............................................................. 17
4.2.4. Household Consumption Expenditure ....................................................................................... 18
4.2.5. Household Asset Wealth ........................................................................................................... 19
4.2.6. Multidimensional Poverty Index ................................................................................................ 19
4.2.7. Resilience Index ......................................................................................................................... 21
5. Comparison of covariates and key outcome variables between intervention and comparison
groups at baseline ...................................................................................................................................... 22
5.1. Respondent and general household characteristics......................................................................... 22
5.2: Economic activities and farm characteristics.................................................................................... 25
5.3. Application of soil and water management options at baseline ....................................................... 27
5.4. Comparison of outcome variables at baseline .................................................................................. 29
6. Impact of DryDev on selected indicators .......................................................................................... 32
6.1. Impact on dietary diversity .......................................................................................................... 32
6.2. Impact on poverty reduction ....................................................................................................... 34
6.3. Impact on Women empowerment ............................................................................................... 37
6.4. Impact on household resilience ................................................................................................... 40
6.5. Probing the mechanisms .................................................................................................................. 41
6.5.1. Impact on local governance and farmer organisations’ functioning ....................................... 41
6.5.2 Impact on adoption of selected soil and water management options ...................................... 43
6.5.3. Impact on production, market integration and profitability .................................................... 47
7. Conclusions, recommendations and lessons learnt .......................................................................... 53
References................................................................................................................................................... 55
Annexes ....................................................................................................................................................... 56
1
Annex 1: Country Maps of Impact study sites.......................................................................................... 56
Table 1. 1: DryDev programme's logical framework and work packages .................................................... 6
Table 3. 1: Impact and outcome Indicators from Inception Report Versus Actual .................................... 11
Table 4.2. 1: Country specific prioritised crops for food and market ......................................................... 16
Table 5.1. 1: Comparison of respondent characteristics at baseline, by programme and country ........... 24
Table 5.1. 2: Comparison of general household characteristics at baseline, by programme and country 24
Table 5.2. 1:Comparison of households’ economic activities at baseline, by programme and country.... 26
Table 5.2. 2:Comparison of farm characteristics at baseline, by programme and country ....................... 26
Table 5.3. 1: Application of soil and water management options at baseline, by programme and country
.................................................................................................................................................................... 28
Table 5.4. 1: Comparison of production and market outcomes at baseline, by programme and country 30
Table 5.4. 2:Comparison of selected outcome variables at baseline, by programme and country ........... 31
Table 6.2. 1: Difference-in-differences results for three measures of poverty .......................................... 36
Table 6.3. 1: DiD results for women empowerment in agriculture index ................................................ 38
Table 6.5. 1: Change in average gross cash value of crops produced (US$), by programme and country 48
Table 6.5. 2: Change in average size of land under cultivation, by programme and country .................... 48
Table 6.5. 3: Change in average net cash value of crops produced (US$), by programme and country ... 49
Table 6.5. 4: Change in average gross cash value o crops sold (US$), by programme and country .......... 50
Table 6.5. 5: Change in net crop profit (gross margins), by programme and country ............................... 52
2
Figure 4.1. 1: Number of farmers interviewed during baseline by country ............................................... 12
Figure 4.1. 2:Number of farmers interviewed during endline survey by country ...................................... 13
Figure 4.2. 1: WEAI Dimensions and Indicators used for DryDev's Impact Evaluation .............................. 17
Figure 4.2. 2: Dimensions and Indicators of the Multi-dimensional poverty index used in DryDev's Impact
Evaluation ................................................................................................................................................... 20
Figure 4.2. 3: Dimensions and Indicators of Resilience Index..................................................................... 21
Figure 6.1. 1: Proportion of HHs consuming 5 or more food groups at baseline and endline ................... 33
Figure 6.1. 2: Proportion of women consuming 5 or more food groups at baseline and endline ............. 33
Figure 6.1. 3: Proportion of men consuming 5 or more food groups at baseline and endline .................. 33
Figure 6.2. 1: Proportion of multidimensionally poor households at baseline and endline ...................... 34
Figure 6.2. 2: Proportion of households above national poverty line at baseline and endline, by
programme and country ............................................................................................................................. 35
Figure 6.3. 2: Changes in dimensions of WEAI for women only sample, by programme and country ...... 39
Figure 6.3. 1: Dimensions of WEAI for the entire sample, by programme and country ............................ 39
Figure 6.4. 1: Comparison of resilient HHs at programme and country level between intervention and
comparison sites ......................................................................................................................................... 41
Figure 6.5. 1: Proportion of HHs receiving extension services at baseline and endline at programme and
country level ............................................................................................................................................... 42
Figure 6.5. 2: Proportion of farmers participating in farmer organisations and reporting benefits at
baseline and endline, by programme and country ..................................................................................... 43
Figure 6.5. 3: Proportion of HHs practicing integrated soil and water management options at baseline
and endline, by programme and country ................................................................................................... 44
Figure 6.5. 4: Proportion of the main cropping field under integrated SWM options at baseline and
endline, by programme and country .......................................................................................................... 45
Figure 6.5. 6:Proportion of HHs practicing soil fertility management at baseline and endline ................. 46
Figure 6.5. 5: Proportion of HHs practicing SWC at baseline and endline ................................................. 46
Figure 6.5. 7: Proportion of HHs practicing water harvesting at baseline and endline ............................. 46
Figure 6.5. 8: Proportion of main cropping field under irrigation at baseline and endline ........................ 47
Figure 6.5. 9: Proportion of HHs practicing irrigation at baseline and endline .......................................... 47
Figure 6.5. 10:Comparison of change in gross cash value of crops produced between intervention and
comparison HHs, by programme and country ............................................................................................ 49
Figure 6.5. 11: Comparison of change in gross cash value of crops sold between intervention and
comparison HHs, by programme and country ............................................................................................ 51
Figure 6.5. 12: Comparison of proportion of farmers selling crop produce at baseline and endline, by
programme and country ............................................................................................................................. 51
3
Executive summary
Over the past decades many development programmes have emerged to reduce poverty, food insecurity
and effects of climate variability in the drylands of sub-Saharan Africa. One such programme is the
Drylands Development Programme (DryDev), which was a six-year initiative (August 2013-July 2019)
funded by the Dutch government with substantial contribution from World Vision Australia to support the
implementation of contextually appropriate natural resource management interventions in the Sahel
(Burkina Faso, Mali and Niger) and the Horn Africa (Kenya and Ethiopia). The overarching goal of the
DryDev programme was to improve and sustain smallholder farmers’ food and water security, livelihoods
and resilience and empower women and other disadvantaged groups.
In this report, we present evidence of the impact of DryDev interventions across five countries in the Sahel
and the Horn of Africa. The objectives of the assessment are twofold: (i)demonstrate whether there is
impact of the DryDev interventions in terms of reduced poverty and food insecurity, empowerment of
women and improved livelihoods and resilience (ii) document mechanisms linking interventions to impact
if any. The assessment was conducted in sites where implementation began in earnest in late 2015 in
countries in the Horn of Africa and in late 2016 in countries in the Sahel.
Our evidence of impact is based on a systematic comparison of outcomes in randomly assigned DryDev
and non-DryDev sites spread across five countries using estimated quasi-experimental evaluation design
based on the difference-in-differences (DiD) method. Outcome data was collected in two time periods
before and after the intervention in 2015 and 2019, respectively over 5,276 randomly sampled households
in the five DryDev countries.
Our main findings indicate that there is mixed evidence of some impact, but evidence is not robust across
the countries. Overall, DryDev induced the adoption of improved soil and water management practices;
better local governance and capacity (local sustainable development plans; access to farmer-friendly and
demand-driven extension); improved household food security (dietary diversity) social inclusion and
women empowerment, with most of the impacts manifesting at the country level or within specific sub-
populations. Specifically, there is some evidence that DryDev improved local governance and institutional
functioning across the programme; led to diversified diets among women in Kenya; increased inclusion
and empowerment of women in Kenya and Ethiopia; and modestly enhanced resilience among
households in Ethiopia. When comparing DryDev(intervention) and non-DryDev(comparison) households,
the results were often inconclusive on production and market outcomes, and poverty outcomes, both at
the programme and country levels.
We postulate that given the short time since implementation of the DryDev interventions, the lack of
evidence of impacts on productivity, market and poverty outcomes were expected. It is highly unlikely
that the NRM interventions, which are mainly characterised by lagged impacts, would yield strong effects
on complex phenomenon like poverty, food security, household resilience after two to three cropping
cycles only. In countries and sub-populations where positive impacts are observed, our analysis could not
establish the mechanisms linking these impacts to outcomes as hypothesised in the theory of change.
Further probing of the data to understand the mechanisms that link the positive impacts observed on
dietary diversity among women in Kenya, women empowerment in Ethiopia and Kenya and household
4
resilience in Ethiopia is necessary to inform the design of future programmes with similar mandate as
DryDev.
One critical lesson arising from the analysis is that maximising productivity gains from natural resource
management interventions requires integration of the options and their adoption at scale. Likewise, in
degraded landscapes where sustenance seems to take precedence over commercialisation, integrating
farmers to markets may require much more efforts than capacity building on innovative marketing
approaches and creating linkages with the private sector actors. Efforts should be geared towards massive
transformation of the landscapes to induce productivity gains beyond what smallholder farmers require
for sustenance in order to generate surplus necessary for commercialisation.
5
1. Introduction
Drylands Development Programme (DryDev) was six-year initiative (August 2013 to July 2019) funded by
the Ministry of Foreign Affairs (MoFA) of the Netherlands, with a substantial contribution from World
Vision Australia (WVA) to cater for implementation of activities in in the Horn of Africa. The DryDev
programme was designed to provide relevant and contextually appropriate support to smallholder
farmers in selected dryland areas of the Sahel (Burkina Faso, Mali and Niger) and the Horn of Africa
(Ethiopia and Kenya). The World Agroforestry (ICRAF) was the overall implementing agency, working with
a consortium of 21 NGOs partners; five National Lead Organizations (NLOs) and 16 implementing partners
across the five countries. The programme sought to contribute to the realization of a vision where
households residing in the drylands would transition from subsistence farming and emergency aid to
sustainable rural development.
The DryDev’s implementation was guided by two complementary and overarching theories of change,
developed during the programme’s inception year (Inception Report). The programme had the ultimate
goal of realising sustainable water and food security, improving livelihoods and resilience and empowering
women and other disadvantaged groups on the one hand and scaling out of programme’s outcomes and
impacts to other dry lands on the other hand. The DryDev programme was implemented through eight
interrelated work packages (Table 1). The eight interrelated work packages adopted by the DryDev
programme, were expected to lead to improvements in short-term and intermediate outcomes and
ultimately contribute to the realisation of the broader goals.
Table 1. 1: DryDev programme's logical framework and work packages

Work Package Sub-outcome Outcome Impact
1. Subcatchment level NRM 1. Appropriate subcatchment level 1. Increased water capture & soil
NRM initiatives undertaken conservation/fertility at
2. On-farm water & soil 2. Improved & climate smart on-farm subcatchment & farm levels 1. Sustained
management water & soil management improvements in
practiced 2. Increased production of food and water
3. Agricultural commodity 3. Improved, inclusive & climate- profitable, climate-smart security,
production smart production options pursued commodities & food crops livelihoods, and
4. Enhancing market access 4. Increased participation of male, resilience, and the
3. Increased sales of targeted empowerment of
female and disadvantaged farmers
value chain commodities by women and
in lucrative value chains
male, female, and vulnerable disadvantaged
5. Financial services linking 5. Increased numbers of famers farmers
linked to credit/financial services groups
6. Local governance & 6. Increased capacity of local duty- 4. Improved local governance &
institutional bearers and farmer organizations & farmer organization
strengthening ‘duty fulfillment’ pressure applied functioning
7. Planning, M&E, and 7. Key ‘scaling stakeholders’ 5. Critical mass of development
scaling of learning identified, find evidence & learning actors motivated, able, and
2. Programme
credible and relevant, and actively resourced to support/directly
outcomes and
promote their uptake implement evidenced options
impacts scaled
8. Policy analysis & 8. Awareness raised and attitudes 6. More supportive/appropriate
out to other dry
influencing improved among key policy makers policies & wider institutional
land areas
and other stakeholders, resulting in environment conducive for
their taking desired action wide uptake of evidence
6
Work packages 1, 2 and 3 focused on enhancing farm productivity sustainably by promoting climate smart
soil and water management options at the sub-watershed and farm-level as well as crops and livestock
production options at the farm level.
Work packages 4 and 5 aimed at expediting the transition of the rural economies in the five countries
from subsistence to commercially oriented agriculture by promoting innovative marketing and financial
deepening approaches including developing capacity of smallholder farmers and linking them with
financial institutions to enable them to invest in climate smart soil and water management options and
the selected lucrative value chains.
The third set of work packages: work packages 6, 7 and 8 were based on the premise that an enabling
environment is a prerequisite for the sustained uptake of productivity enhancing practices and integration
of farmers in lucrative markets and value chains. Thus, work packages 6 and 7 aimed at strengthening the
capacity of local institutions such as farmer organisations and national institutions, which were critical
programme delivery mechanisms, to enable farmers and local duty bearers to participate effectively in
and facilitate development programmes, respectively.
At a higher level, policy analysis and influencing were expected to raise awareness of policy makers of the
need to support development initiatives aimed at improving household and environmental resilience.
Using evidence generated from the DryDev programme to demonstrate the value of investing in natural
resource management interventions, policy makers and key government officials would be incentivised
to invest their time and resources in providing supportive policies and well-functioning institutions, which
are critical for the sustainability of the technologies and scaling approaches developed and promoted by
the DryDev programme.
To ensure that implementation of the programme was on course and that the programme was likely to
realise its intended long-term objectives, a monitoring and evaluation system was put in place. The system
involved periodic tracking of the programme’s outputs by the NLOs and assessment of changes in short-
term outcomes (sub-outcomes) through short annual surveys. Changes in the intermediate and higher-
level outcomes were to be measured through a longitudinal survey comprising two datasets collected
before and at the end of the programme as outlined in the Inception Report. This report provides a
detailed account of the quantitative impacts of DryDev in the five countries, focusing particularly on the
intermediate and long-term outcomes realised through direct support to smallholder farmers. The report
focuses only on geographical areas where the programme was implemented after the inception phase
from 2016 to 2019.
The rest of this report is organised as follows: section 2 presents information on impact evaluation
strategy and the evaluation design adopted in the study. In section 3, we provide a summary of impact
and outcome indicators on which impact estimation was based. In section 4, we present a brief description
of the type of data collected, how the data was collected including issues pertaining to sample attrition,
as well as how the data was analysed, and various impact indicators constructed. Section 5 compares
household socio-economic and demographic covariates and key outcome indicators between the
intervention and comparison households at baseline to ascertain if the balance in the covariates and
outcome indicators between the two groups was affected by sample attrition. In section 6, we present
and discuss the impacts of DryDev and explore the mechanisms that led to the impacts. The report ends
with conclusion, recommendations and lessons arising from the evaluation in section 7.
7
2. Impact Evaluation Strategy
DryDev sought to achieve impact through two mutually reinforcing approaches (a) directly supporting
smallholder farmers; and (b) promoting the uptake of programme-generated evidence and learning by
policy makers and other development actors for application elsewhere. This section outlines the strategy
adopted to estimate DryDev’s impact on the smallholder farmers targeted directly by the programme.
Estimating the directly facilitated outcomes and impacts involved comparing changes over the life of the
programme in the status of key impact and outcome variables (e.g. household income and food security)
experienced by smallholder farmers the DryDev programme directly supported with similar farmers in
neighbouring geographic areas that were not supported. DryDev’s strategy to evaluate the programme’s
overall direct impacts focused only on geographic areas the programme targeted following its Inception
Phase. The inception (quick win sites), were not included in the overall assessment because they had
benefitted from programme implementation prior to and during the inception phase1.
The changes (impacts) were estimated using quasi-experimental evaluation design based on the
difference-in-differences (DiD) method. As outlined in the Inception Report, several evaluation designs
including randomised controlled trial (RCT) design were considered and their applicability evaluated
before settling on the DiD method. RCT proved infeasible because of the relatively small number of sub-
watersheds, the unit of intervention, where the programme could expand to during its full
implementation phase. Random assignment ensures that the units (e.g. households) assigned to the
intervention and comparison groups are statistically similar in all respects including baseline status of the
outcome variable itself and observable and unobservable factors that may affect the evolution of this
variable over time. Randomisation, however, requires a large number of units to yield statistically valid
conclusions.
Difference-in-differences method
DiD method compares changes in outcomes over time between a population that is enrolled in a
programme (the treatment group) and a population that is not enrolled (the comparison group), assuming
equal trends over time for the two groups. The DiD method, which makes use of data collected in at least
two time periods, allows us to correct for differences between the treatment and comparison group that
is constant over time. Simply observing the before and after changes in outcome and impact indicators of
interest, say productivity or food security, for households enrolled in the programme does not capture
the programmes causal impact because there are many other confounding factors, which are likely to
influence productivity or food security over time. Similarly, comparing households in the programme and
non-programme sites is likely to lead to biased outcomes because there may be some unobserved reasons
that make the two groups inherently different and these reasons may be related to the outcome of
interest (e.g. productivity or food security).
1
An exception to this is Ethiopia where the two districts delineated for impact assessment were found not
to be representative of the four districts where the inception phase activities were implemented. As such,
data was collected from two more districts following a quasi-experimental design in which the sub-
watersheds in the intervention districts were matched with sub-watersheds located in the same district
but had not been exposed to similar NRM interventions. Impact evaluation for these districts are
presented in the country report.
8
DiD approach combines the two counterfeit estimates i.e. differences in the before-and-after outcomes
for the treatment group (T1 -T0), and the before and after change in outcomes for a group that did not
enrol in the programme (C1-C0) but was exposed to the same set of conditions as the treatment group.
The result of the two differences, (T1 -T0)- (C1-C0) is the causal impact of the programme as outlined in
Figure 2.1.
Prior to implementation of the DryDev programme in the impact sites, efforts were made to purposively
match each sub-watershed unit targeted for programme expansion to one comparison sub-watershed
unit that is (a) relatively nearby but not so close that it could be influenced by significant programme spill
over effects; (b) had similar biophysical characteristics (e.g. slope, rainfall patterns, and soil conditions)
and socioeconomic characteristics (e.g. poverty levels and land use patterns); and (c) was part of the same
larger watershed, yet independent from being affected by the programme’s NRM interventions from a
hydrogeological perspective. Detailed country maps showing the location of farmer fields that were
sampled during the baseline survey and by extension the location of the DryDev’s impacts study areas and
the sub-watersheds can be found in Annex 1 of the Baseline Survey Report together with a complete list
of the names of each of the sub-watershed units. The data collection and analysis processes used during
DryDev’s impact evaluation exercise are described in Section 4.
9
During DryDev’s endline survey, efforts were made to collect data from the same representative samples
of farmers and farmer fields that were surveyed during baseline, despite sample attrition of about 28%
on average experienced at the programme level. Although the plans were to use mixed methods approach
encompassing both quantitative and qualitatively approaches, the report is based on quantitative data.
The plan was to use qualitative methods to triangulate and add depth to quantitative results and
interrogate the mechanisms of how and why the changes expected from the programme did or did not
manifest. Thus, preliminary findings from the quantitative study was a precursor for the qualitative study.
The planned qualitative study did not materialise because data cleaning and analysis took considerable
amount of time than was anticipated in the budget. By the time the data cleaning and analysis was
completed, most of the staff on the ground had already left the DryDev programme.
3. Impact and Outcome Indicators

DryDev’s quantitative impact assessment strategy focused on assessing progress in relation to the
outcome and impact levels of the programme’s theory of change for its direct work with smallholder
farmers. Efforts were made during DryDev’s baseline and endline surveys to capture data on specific
indicators documented in the DryDev’s Inception Report for each corresponding impact and outcome
statement. The extent to which relevant data were successfully captured for each indicator is presented
in Table 3.1.
As shown in Table 3.1, Baseline and endline data were successfully collected for nearly all the indicators
except for crop water productivity. It proved technically difficult to capture reliable data on this specific
indicator in the household surveys due to the nature of the targeted mixed smallholder farming systems.
While there were plans to capture data for this indicator through remote sensing to inform DryDev’s
endline evaluation, this was not pursued further because of lack experienced field technicians. Capturing
accurate data requires intensive field measurement to monitor crop growth, water conditions and water
supply at the study sites. In addition, difficulties were encountered in differentiating food crops from
market crops, given that many households both consume and sell the former. Consequently, indicators
for (a) the cash value of food crops and (b) the cash value of marketed crops were lumped together. The
resulting indicator—the cash value of crops harvested during the previous farming season—estimates the
total monetary value (using farm gate prices) of the crops a household produced during a given period
(see Section 4).
10
Table 3. 1: Impact and outcome Indicators from Inception Report Versus Actual
Outcome level Indicator from Inception Relevant data captured Relevant data captured
Report in baseline survey? in endline survey?
Impact Level
Sustained % of senior men and senior women Yes
improvements in food in HH consuming 5 or more of the
Minimum Dietary Diversity food
Yes
and water security,
groups
livelihoods, and
% HHs predicted to be above the Yes, but World Bank poverty Yes
resilience, and the
national poverty line lines used
empowerment of
women and other % of HHs > median of comparison Yes
group on HH asset index
Yes
disadvantaged groups
% of HHs that are multi- Yes
dimensionally poor
Yes
Women Empowerment in Yes
Agricultural Index (WEAI): % of men
& women scoring positively on over
Yes
2/3 of weighted indicators
% of HHs scoring positively on over Yes
2/3 of weighted indicators of Yes
resilience index
Outcome Level
1. Increased water Predicted soil organic carbon Not estimated, but farm
capture & soil content (via remote sensing) Yes
polygons were collected
conservation/fertility Predicted erosion prevalence & rood Not estimated, but farm
at watershed & farm depth restriction (via remote Yes
levels
sensing)
Enhanced vegetation & herbaceous Not estimated, but farm
cover indices (via remote sensing) Yes
% of HH farm plot(s) serviced by Yes
Yes
irrigation in last growing season
2. Increased production Crop water productivity—yield Still being explored; indicator Not captured
of profitable, climate- (biomass)/ evapotranspiration (via technically challenging to
smart commodities & remote sensing) implement in mixed farming
food crops
systems
Estimated cash value of main food Yes
Yes, but cash value of crops
crops last main harvest
not differentiated between
Estimated cash value of main market Yes
crops last main harvest main food and market crops
3. Increased sales of Estimated cash value of agricultural Yes
targeted value chain commodities sold by male & female
commodities sold by HH members in last 12 months Yes
male, female, and
vulnerable farmers
4. Improved local # of M&F farmers reporting Yes
governance & farmer improved agricultural local gov. Yes
organization services
functioning # of M&F farmers participating in & Yes
reporting benefits from FOs
Yes, as part of WEAI
11
4. Data Collection and Analysis
4.1. Data collection and cleaning
As mentioned earlier in section 2, during DryDev’s endline survey, efforts were made to track the same
households who were interviewed during the baseline. Baseline survey in the impact study sites was
conducted between July 2015 and November 2015 while the endline survey was undertaken between
April 2019 and May 2019. To administer DryDev’s baseline survey, population statistics were first
compiled for each sub-watershed site and stratified proportionate sampling was used to calculate the
sample size required for each (see Baseline Survey Report). In this way, it was possible to obtain
representative country-level statistics without employing sampling weights. During baseline, a total of
7901 female and male farmers were interviewed, and data were captured via remote sensing from 7,887
randomly sampled fields. The number of farmers interviewed in each country during the baseline survey
by treatment group is presented in Figure 4.1.1
2,000
1,500 1,116
802 881 864 775
1,000
500 873
626 685 625 654
0
Kenya Ethiopia Mali Niger Burkina
Intervention Households Comparison Households
Figure 4.1. 1: Number of farmers interviewed during baseline by country
As is expected of any longitudinal study, not all farmers interviewed in the baseline could be tracked for
the endline survey. A total of 5276 farmers or 67% of the baseline sample consisting of 52% and 48%
female and male farmers, respectively were interviewed. High attrition rates were observed in the Sahel
countries, with rates as high as 35% and 41% in Niger and Mali, respectively. In Mali 15 % of the baseline
farmers could not be interviewed because of insecurity in Segou and Bankass regions, while 20% of the
baseline sample was not interviewed in Niger because one partner pulled out of the consortium midway
through the programme. Attrition was mainly due to natural causes such as death or relocation of
previously interviewed farmers from the study site or refusal of consent by the farmer who was
interviewed the during baseline. Figure 4.1.2 shows the number of farmers interviewed in each country
by treatment group during the endline survey. Detailed location of the farmers interviewed in each of the
sub-watersheds in each country is presented in Annex 1. While the baseline survey covered a total of 71
sub watersheds, the endline survey was conducted in 62 sub-watersheds. The endline survey could not
be conducted in 3 sub-watersheds in Mali because of insecurity while six sub-watersheds in Torodi
12
commune in Niger were dropped from the programme’s implementation sites when the implementing
partner pulled out of the consortium.
2000
1500
1000
642 691
575
504
500 446
510 569 449 512
378
0
Kenya Ethiopia Mali Niger Burkina
Intervention households Comparison households
Figure 4.1. 2:Number of farmers interviewed during endline survey by country
Survey instrument
The same survey instrument developed for the baseline survey to capture data on the indicators
presented in Section 3, as well as relevant farmer and household characteristics was used for the endline
survey. Nearly 90% of the questions in the baseline instrument were administered during endline survey,
and only time invariant household characteristics such as sex, age of the respondent among others were
dropped from the endline survey instrument. The instrument was reviewed, translated into the local
languages in Ethiopia and French for the Sahel countries and formatted in Excel for application in
SurveyCTO. The draft survey instrument was first tested and implemented in Kenya, then adapted for
each of the four other countries. Considering the linguistic context of each country, Tigrinya and Oromiffa
versions of the survey instrument were used in Ethiopia, an English version in Kenya, and French versions
in the three Sahelian countries.
Recruitment of enumerators in each of the countries followed the same procedure that was applied
during baseline. In each country, efforts were made to recruit enumerator teams comprising of an even
number of males and females with appropriate educational qualifications (e.g. preferably a university or
college degree) and relevant enumeration experience. Those initially recruited were then trained for a
period of two to three days, which included practice run at administering the survey instrument. Their
performance was critically reviewed during this time and corrective measures undertaken, including
disengaging several underperforming enumerator trainees.
Names of farmers who were interviewed in the baseline survey including the GPS coordinates of their
homes were extracted from the baseline database to aid in tracking them in preparation for endline
survey. The list was then uploaded on the surveyCTO server together with the instrument adapted for the
endline survey. The respondent’s critical details collected during baseline survey were programmed in the
endline survey instrument and pulled during the interviews to minimise errors in identifying the
respondents. During baseline, attempts were made to ensure that the sampling frame only included
smallholder farming households that farmed during the last farming season within the sub-watershed site
in question. During endline survey, however, this restriction was relaxed, and efforts were made to track
13
all the farmers in the baseline database irrespective of whether they farmed in the previous agricultural
season or not. As indicated earlier, 67% of the baseline respondents were tracked and interviewed.
Given that DryDev was particularly interested in evaluating whether the programme impacted male and
female smallholders differently, male and female famers were each asked questions pertaining to both
themselves and their households. Efforts were further made to ensure that (a) the same male and female
smallholder farmers interviewed at the baseline were tracked and interviewed; and (b) the gender of the
enumerator and each farmer they interviewed was the same. Because data was required from the same
farmer who was interviewed in the baseline, replacement was not done for the respondents who could
not be tracked, were not available due to death or relocated to other areas outside the study site or
declined to be interviewed.
The enumerator protocol involved interviewing the farmers privately for the personalized questions first
and then opening the interview process to include other household members for the questions focusing
on the broader household. One of the modifications made to the endline tool was the introduction of a
question to verify if the respondents’ main farm that was sampled during the baseline was still being
farmed by the respondent or the household. Where this farm was no longer available for the household,
another farm was randomly selected to replace the farm that was surveyed in the baseline. The end of
the interview involved visiting the newly sampled field or the same field that was surveyed during
baseline. The visit included taking perimeter of the field by walking around the field, as well as asking the
farmer questions about the field and making specific complementary observations. A total of 5276 farms
were georeferenced, with about 6% of them being newly sampled farms that had been farmed by the
respondents for between 1 and more than 5 years.
The polygons of the sampled fields were to be analysed at the ICRAF’s Geospatial Lab, for the purpose of
generating remote sensing-based estimates for the specific soil health indicators like those generated in
the baseline. Because the endline soil health indicators were not ready at the time of analysis, the weights
for one of the dimensions of resilience index indicator that required soil health indicators for estimation
was modified accordingly.
During the first few days of the survey, the completed questionnaires (in electronic form on the
enumerators’ smartphones or tablets) were checked on a daily basis by the assigned ICRAF scientist or
consultant in question, prior to being uploaded onto an internet server (https://ona.io/home/) and
(https://www.surveycto.com/) for baseline and endline surveys, respectively. After each country survey
exercise was completed, the data were then downloaded directly in Stata for cleaning. In Kenya, additional
quality checks were put in place such as backchecks. In this case, a set of questions were randomly
sampled and administered daily to a subset of farmers who had been interviewed the previous day. Both
the enumerators to be backchecked as well as the questions for backchecking were randomly sampled.
Administration of the backcheck questions was done by two independent enumerators who were not
involved in the actual survey. To make the exercise credible, the backchecking team was not informed of
the questions or the enumerators to be backchecked until the day of the exercise. A similar exercise could
not be undertaken in the four countries because of budget constraints.
Generally, the quality of data was reasonable, mainly due to the electronic data collection method used,
which made it possible to screen the data online and provide instant feedback to the field supervisors and
enumerators. As is expected of surveys that rely on recall, some mistakes were identified during data
cleaning. Like in the baseline survey, most errors were noted with quantitative variables such as those
pertaining to land size, crop output and expenditure on food consumption. To rid the data of extremely
14
outlying values, most of the quantitative variables were winsorised at 99%. Other errors noted during
data cleaning include:
• Food consumption errors. As part of the household food consumption module,

respondents were asked how many units (e.g. kilograms) of specific food items (e.g.
maize flour) their households consumed during the previous week. In several cases,
the number of consumed units was excessively high and implausible in the baseline
data. Given that such errors were only confined to one or two of the many food
items, correcting such excessively high values involved replacing the erroneous
values with the median value of the other households in the country dataset in
question. In the endline survey module, however, such errors were minimised
considerably by precoding of the units of measurement.
• Abnormally high monthly savings rates. In the regular monthly expenditure module,
the respondents were asked whether their household had saved any money during
the previous four weeks and, if so, the amount. Savings, is seen as a type of
expenditure, particularly given that household consumption expenditure is being
used as a proxy for household income (see Section 4.2). However, particularly in
Kenya, the amount indicated for several households was very high, i.e. the
equivalent of over USD $1,000. While this is possible, it is unlikely that such high
rates of savings would take place on a monthly basis. Albeit arbitrary, the maximum
amount allowable for household monthly savings was capped at USD $200 both in
the baseline and endline data.
• Likelihood of spill over effects, particularly in Kenya. As explained in Section 2, the

intention was to select intervention and comparison sub-watershed units separated
by considerable distance (e.g. 5 kilometres) or, otherwise, by natural barrier, such as
a mountain. However, in the case of Kenya, this proved difficult in negotiations with
the implementing partners. Consequently, the intervention and comparison sites
were selected from neighbouring sub-locations. Efforts were made to assess and
control for potential spill-over effects during the endline survey and data analysis.
4.2. Data processing and construction of indicators

4.2.1. Crop production and sales estimates
As mentioned earlier in section 3, one of DryDev’s key outcomes was to increase the sustained production
and marketing of selected crops for which each site had comparative advantage. Each country prioritised
a set of crops, which were promoted for food as well as income during the DryDev programme’s
implementation (see Table 4.2.1.). Aside from the prioritised crops, which benefitted from value chain
strengthening interventions, other major staple crops which are not listed in Table 4.2.1 but were
cultivated by majority of the households benefitted greatly from productivity enhancement interventions
implemented in the DryDev programme. Therefore, to compute production and market outcomes, all the
crops cultivated by the households in the previous agricultural year (2014 and 2018 for baseline and
endline, respectively) were aggregated using the prevailing farm gate prices reported by the farmers.
15
Table 4.2. 1: Country specific prioritised crops for food and market
Burkina Faso Ethiopia Kenya Mali Niger
• Groundnut • Milk • Mango • Millet • Groundnut,
• Millet • Haricot bean • Onion • Sorghum • Cow peas
• Onion • Onion • Pulses (green • Fonio • Sesame
• Cow peas • Potato gram, pigeon, • Groundnuts
• Sorghum • Sorghum cow peas) • Sesame
• Tomato • Spinach • Soya beans
• Wheat • Tomato • Onion
• Garlic • Watermelon
To capture data on crop production levels at baseline and endline, the interviewed farmers were asked:
(a) the specific crops their households grew in 2014 and 2018, respectively; (b) the quantify of these crops
harvested and their cash value at the time of harvest (i.e. farm gate prices); (c) the expenses incurred in
producing and/or marketing each crop: and, finally, (d) the quantity sold, if any.
Rather than analysing results for each specific crop, the data were aggregated for all crops to generate
four variables: (1) gross harvest value, i.e. the total cash value of all crops each household harvested in
2014 and 2018 at the time of their harvest; (2) net harvest value, i.e. the total cash value of all crops
harvested in 2014 and 2018 minus all relevant expenses (e.g. fertilizer, pesticides, labour, and transport);
(3) gross crop sales, i.e. the gross amount received from all crops harvested and sold in 2014 and 2018;
and (4) net crop sales, i.e. cash received from all crops harvested and sold in 2014 and 2018 minus all
expenses incurred to produce and/or market them.
4.2.2. Dietary diversity index

Another key aim of the DryDev programme was to improve household food security. Capturing and
analysing data on anthropometric measures, such as weight-for-height, height-for-age, and weight-for-
age of under-five children and body mass index (BMI) for adults, is recognized as a valid approach for
assessing the nutritional and food security status of a population.2 However, such measures are costly and
time consuming to administer, and, consequently, a number of alternative measures have been proposed.
The Women’s Dietary Diversity Project (WDDP) was launched to address the need for simple, yet valid,
measure of women’s diet quality, with a specific focus on micronutrient adequacy. This culminated in the
development and ‘adoption’ of the Minimum Dietary Diversity–Women (MDD–W) indicator.3
To capture data on this indicator during DryDev’s baseline and endline survey, both the female and male
respondents were asked whether they consumed particular food items during the previous day, which fall
under the following 10 categories: (1) starchy staple foods (2) pulses (beans and peas) (3) nuts and seeds
(4) dairy products (5) fleshy foods (meat, fish, poultry) (6) eggs (7) vitamin A-rich dark green leafy
vegetables (8) other vitamin A- rich vegetables and fruits (9) other vegetables, and (10) other fruits. While
this indicator is specifically designed for women, the corresponding survey module was also administered
to men. A binary indicator was constructed for the MDD-W indicator by assigning a value of 1 if a
respondent reported that she/he had consumed at least five of the 10 food groups during the previous
2
See, for example: WFP. (2005) A Manual: Measuring and Interpreting Malnutrition and Mortality. Nutrition
Service, World Food Programme, Rome.
3
See: http://www.fao.org/3/a-i5486e.pdf
16
day and 0 if otherwise.
4.2.3. Women’s Empowerment in Agriculture Index (WEAI)
An adapted version of the Women’s Empowerment in Agricultural Index (WEAI) was administered to both
the female and male respondents in the baseline and endline surveys. The WEAI was developed by the
Oxford Poverty and Human Development Initiative (OPHI), the United States Agency for International
Development (USAID), and the International Food Policy Research Institute (IFPRI). It is designed to
measure the empowerment, agency, and the inclusion of women in the agricultural sector.4
Figure 4.2. 1: WEAI Dimensions and Indicators used for DryDev's Impact Evaluation
The WEAI measures women’s empowerment in five dimensions or areas: production, resources, income,
leadership, and time use. The original WEAI comprises 10 indicators that fall under these five dimensions.
The original framework was followed largely, albeit with several adaptations. The adaptions made in the
DryDev instrument to capture dimensions of WEAI together with other details concerning WEAI are
explained in the Baseline Survey Report.
The WEAI is a weighted index that ranges from 0 to 1. Each of its constitutive indicators is binary, taking a
value of 1 if the respondent scores positively in relation to it and 0 if otherwise. These binary cut-offs are
arbitrary but reflect the fact that the respondent must reach a minimum level in order to be given a
positive score. The 11 indicators that make up DryDev’s adapted version of the WEAI were all weighted
equally within each of the five dimensions, and each of these dimensions was also weighted equally.
Annex 3 of the Baseline Survey Report presents details on how data for each of the 11 indicators was
4
See: http://www.ophi.org.uk/policy/national-policy/the-womens-empowerment-in-agriculture-index/
17
obtained from the interviewed respondents, as well as the specific binary cut-offs that were used and the
specific weights applied.
4.2.4. Household Consumption Expenditure
Household consumption expenditure is one popular proxy measure that has been used by World Bank
and other international institutions in low- and middle-income countries where majority of the population
is unemployed or self-employed, making it impossible to obtain accurate data on income. Hence
estimation of the percentages of households living below and above the poverty line is usually based on
household consumption and expenditure data.
To capture these data, several modules were incorporated into the baseline and endline household survey
instrument. The respondents were asked what types of food they consumed over the previous seven-day
period, as well as the respective quantities. The quantities of each food item consumed were then
converted into a monetary value. This was done by asking the respondent how much was paid for the
food item in question or, if the food item was sourced through the household’s own production, how
much it would have cost if it was purchased from the local market. The respondents were also asked how
much they spent on non-food items and services from a detailed list, such as soap, toothpaste, and
minibus fares, over the past four weeks (regular non-food expenditure). Finally, they were asked about
particular “big ticket” expenditures over the previous 12 months from another pre-defined list, such as
school and healthcare expenses, clothes, and home repair, over the last 12 months (non-regular non-food
expenditure).
The basic per capita measure was calculated as follows for each household:
• The weekly cash value of each food item consumed during the past seven days were
added together and divided by seven, thereby estimating the daily cash value of
food consumed by the household.
• Household expenditure on items from both the regular monthly non-food
expenditure list and annual non-food expenditure list were added together and
divided by 30 and 365, respectively, thereby estimating the household’s average
daily expenditure on regular and non-regular non-food items.
• The daily consumption expenditure estimated for food and the regular and non-
regular non-food items were then added together and converted into US dollars,
while adjusting for purchase power parity (PPP).5
• Finally, to derive each household’s per capita consumption expenditure, its PPP
adjusted dollar value was divided by the number of its members (household size),
with another adjustment being made for assumed lower consumption among
children and economies of scale.6
5
Adjusting to PPP was undertaken to take into account each country’s idiosyncratic purchasing power, i.e. the
quantity of currency required to purchase a given basket of goods and services. The PPP conversion rates used were
taken from the World Bank’s website: http://data.worldbank.org/indicator/PA.NUS.PPPC.RF
6
While dividing the above by household size as the overall denominator is recommended in the literature, it is
considered important to avoid underestimating expenditure for larger sized households relative to their smaller
counterparts. A recommended formula for computing household size for this purpose is: HH size = (𝐴 + 𝛼𝐾)𝜃 where
A is number of adults in the household; K is the number of children; 𝛼 is the cost of a child relative to an adult; and 𝜃
adjusts for economies of scale. For low income countries, it is recommended that 𝛼 be set at .25 or .33 and 𝜃 be set at
.9 (ibid).
18
4.2.5. Household Asset Wealth
An alternative way of measuring a household’s wealth status in low-and middle-income countries is by

analysing the assets a household owns and other wealth indicators, e.g. the material of a household’s roof
and/or floor. Consumption expenditure data are generally recognised as being more sensitive to recent
changes in household income, while asset based measures typically reflect a household’s more
established wealth status.7 This is because increases in household income generally need to be sustained
for some time before they translate into significant increases in household asset ownership. In other
words, after a household’s income has been sustainably increased, it will take time for it to accumulate
assets, make improvements to the home, and so forth.
During the surveys, respondents were asked whether they or anyone in their household currently owns
particular assets and other wealth indicators from a long list of over 90 items. The inter-item correlation
among these assets was subsequently tested.8 Assets that were not sufficiently correlated with the main
pool of assets were dropped, leaving streamlined set of household wealth indicators.
A wealth index for each country (or district for Ethiopia) was constructed from a tetrachoric matrix, which
is used to measure the degree to which binary variables are statistically associated. Principal component
factor analysis was then run on this matrix. This is a data reduction technique that was used to narrow in
on the variation in household asset ownership, which is assumed to represent wealth status. The more an
asset is correlated with this variation, the more weight it is given. Hence, each household’s weighted index
score is determined by both (a) the number of assets it owns; and (b) the particular weight assigned to
each asset. This enables the relative wealth status between defined groups of households to be compared.
Note that the index is a relative (rather than absolute) measure, so comparisons are only valid between
the intervention and comparison groups, rather than across countries.
4.2.6. Multidimensional Poverty Index
Also developed by the Oxford Poverty and Human Development Initiative (OPHI), the global
Multidimensional Poverty Index (MPI) is designed to complement income-based measures of poverty by
simultaneously capturing several of its other dimensions.9 Drawing on existing sources of country-level
data, it has been applied to over 100 low and middle income countries.
7
Moser, C. F. 2007. The Construction of an Asset Index: Measuring Asset Accumultion in Ecuador. Washington: The Brookings
Institution.
8
When items are used in a scale or index, they should all measure the same underlying latent construct (e.g.
household wealth status). The items, therefore, must be significantly correlated with one another. Cronbach’s
alpha is a measure of this inter-item correlation. The more the variables are correlated, the greater is the sum of
the common variation they share. If all items are perfectly correlated, alpha would be 1 and 0 if they all were
independent from one another. For comparing groups, an alpha of 0.7 or 0. 8 is considered satisfactory. See:
Bland, M. J. & Altman, D. G. 1997. Statistics notes: Cronbach's alpha. BMJ, 314, 572.
9
See: http://www.ophi.org.uk/multidimensional-poverty-index/
19
Figure 4.2. 2: Dimensions and Indicators of the Multi-dimensional poverty index used in DryDev's Impact
Evaluation
Figure 4.2.2 presents the three dimensions and 10 specific indicators for a slightly adapted version of the
global MPI that was used for DryDev’s impact evaluation study. Like the WEAI, each of the indicators
falling under each of the three dimensions are weighted equally, as are each of these three dimensions.
For the purposes of DryDev’s impact evaluation, there was only one significant modification from the
original global MPI. For this global measure, women’s body mass index (BMI) is used to inform its nutrition
indication. Given that such anthropometric baseline data were not captured by DryDev’s baseline and
endline surveys, the MDD-W binary indicator was used as a substitute. Annex 4 of the Baseline Survey
Report presents specific details on how each of the above indicators was measured, as well as the specific
binary cut-offs and weights associated with each. As per the Alkire-Foster method, calculating the MPI
involves “discounting”. In particular, if an individual or household is deprived on one-third or more of the
weighted indicators, the weighted indicator score is replaced with 1, given that the multi-dimensional
poverty threshold has been surpassed.
20
4.2.7. Resilience Index
Whereas DryDev’s ultimate goal was to improve household income and food security, as well as empower
women and disadvantaged groups, the interventions were carefully selected, and their implementation
guided by the programme’s seven principles outlined in the Inception report to ensure that the outcomes
were sustained for the long-term. In the face of climate change and external drivers of change affecting
the drylands, the programme promoted resilience building among the households and communities it was
targeting. However, given that resilience is another complex multidimensional concept, measuring it is
particularly challenging.
Figure 4.2. 3: Dimensions and Indicators of Resilience Index
DryDev adopted following Oxfam GB’s approach for measuring resilience.10 One of the complexities
associated with measuring resilience is that the success of a household, community, organization, or any
system for that matter in bouncing back from a shock or adapting to change more generally can only be
ascertained after the fact.11 Characteristic approaches for measuring resilience, such as Oxfam GB’s,
attempts to overcome this by hypothesizing that there are particular characteristics of households (and
10
Hughes, K. and Bushell, H. (2013), A Multidimensional Approach for Measuring Resilience. Oxfam GB Working
Paper, Oxford.
11
Dodman, D., Ayers, J. and Huq, S. (2009), ‘Building Resilience’, Chapter 5, in World Watch Institute (ed), ‘2009
State of the World: Into a Warming World’, Washington D.C: World Watch Institute, pp. 151-168.
21
systems more generally) that affect how well they are able to cope with shocks and positively adapt to
change. The dimensions and specific characteristics (indicators) that make up DryDev’s adapted version
of the resilience index are presented in Figure 4.2.4. Detailed information on the four dimensions and
specific indicators under each dimension is provided in the Baseline Survey Report.
As is the case for the WEAI and MPI, Annex 5.4 of the Baseline Survey Report presents specific details on
how each of the above indicators was measured, as well as the specific binary cut-offs and weights
associated with each. DryDev’s resilience index also makes use of the Alkire-Foster method. Here, the
index is set on a scale ranging from 0 (very low resilience) to 1 (significantly high resilience), and each
indicator is weighted equally under each of the four dimensions, which are also equally weighted.
Moreover, each weighted indictor score was replaced with the maximum possible score of 1 if the
household in question scored positively in relation to at least two-thirds of the weighted indicators.
5. Comparison of covariates and key outcome

variables between intervention and comparison
groups at baseline
Whereas the treatment and comparison groups were balanced on most of the outcome variables and
other respondent and household characteristics at baseline (see Baseline Survey Report), the high
attrition rate observed across the programme in general and in Mali and Niger in the endline survey, in
particular, justifies the need to assess whether the balance on key outcomes and farmer characteristics is
still maintained between the two groups. In this section, farmer characteristics and baseline outcome
status associated with each of the treatment groups are compared to assess the extent to which the
remaining sample is balanced across the five countries, as well as at the programme level.
5.1. Respondent and general household characteristics

Table 5.1.1 presents baseline characteristics of the respondents in the intervention and comparison
groups at the programme and country level. At the programme level, respondents in the intervention and
comparison groups appear similar in most of the assessed covariates, with the exception of one indicator
where significantly more respondents in the comparison sites than in the intervention sites are of good
health. Ethiopia is the only country where all the treatment and comparison groups are balanced on
respondent characteristics while in Kenya a significantly higher number of the respondents in the
intervention sites were spouses of the household. Niger seems to have lost significantly more young, male
respondents who were also heads of the households in the intervention sites due to attrition. In Mali,
however, the remaining sample appears to be balanced on respondent’s characteristics except on health
status, and so is Burkina Faso, where significantly more respondents in the comparison sites appear to be
of good health compared to those in the comparison sites. Respondents in the intervention site in Burkina
Faso are more literate and have more years of schooling than those in the comparison sites. In sum, Niger
and Burkina Faso seem to have been affected by the attrition resulting in more unbalanced respondent
characteristics compared to the rest of the countries.
22
General household characteristics presented in Table 5.1.2 show that the intervention and comparison
groups are balanced on all household covariates at programme level, as well as in Ethiopia and Kenya.
Burkina Faso has the highest number of unbalanced general household characteristics followed by Mali
and Niger with three covariates. All the unbalanced covariates in Burkina Faso relate to household
composition, and households in the intervention sites have significantly larger household sizes, more
children and adult members and more productive adult household members. For Mali, the unbalanced
household covariates relate to education of the household head and other adult household members, as
well as age category of adult household members. Households in intervention sites in Mali have
significantly relatively more educated household heads and adult members of the households compared
to those in the comparison sites while adult household members in the comparison sites are more likely
to be older than 59 years. Households in the intervention sites in Niger have significantly more adult
household members and the household members are significantly older than 60 years. On the other hand,
households in the comparison sites have significantly more literate female adult members. Striking from
the table is the high proportion of female headed households in Kenya compared to the rest of the DryDev
countries. The figure is twice as high as the pooled programme average.
23
Table 5.1. 1: Comparison of respondent characteristics at baseline, by programme and country
Programme Ethiopia Kenya Burkina Faso Mali Niger
Dif. Dif. Dif. Dif. Dif. Dif.

Inter. Com. Inter. Com. Inter. Com. Inter. Com. Inter. Com. Inter. Com.
Female 0.52 0.51 0.009 0.52 0.53 -0.03 0.52 0.53 -0.02 0.49 0.49 0.019 0.47 0.48 -0.011 0.58 0.53 0.14***
Over 60 0.10 0.10 0.043 0.03 0.03 -0.05 0.21 0.20 0.061 0.12 0.11 0.045 0.11 0.11 -0.052 0.04 0.03 0.18***
Non-dominant religion 0.15 0.16 -0.12 0.004 0 0(.) 0.01 0.02 -0.18 0.25 0.34 -0.28 0.59 0.59 0.16 0.00 0.00
Ethnic minority 0.04 0.03 0.05 0 0.003 0(.) 0.00 0.01 0 (.) 0.05 0.05 -0.032 0.15 0.09 0.4 0.02 0.02 0.059
Married & living with
0.88 0.87 0.009 0.89 0.89 0.007 0.72 0.70 0.051 0.91 0.90 0.052 0.95 0.97 -0.17 0.95 0.95 -0.1
spouse
Head of HH 0.50 0.52 -0.04 0.52 0.51 0.023 0.54 0.56 -0.052 0.52 0.52 -0.018 0.48 0.48 -0.062 0.44 0.50 -0.17***
Wife of HH head 0.34 0.33 0.04 0.42 0.43 -0.02 0.39 0.36 0.066*** 0.18 0.19 -0.002 0.29 0.29 0.007 0.40 0.33 0.20***
Literate 0.32 0.32 0.018 0.22 0.23 -0.07 0.85 0.85 0.015 0.15 0.13 0.11** 0.18 0.17 0.018 0.15 0.14 0.072
> 7 years of schooling 0.12 0.13 -0.03 0.03 0.04 -0.12 0.47 0.48 -0.011 0.02 0.03 -0.23*** 0.03 0.03 0.11 0.02 0.02 0.14
Good health 0.97 0.98 -0.18* 0.97 0.98 -0.33 0.98 0.99 -0.099 0.97 0.98 -0.23* 0.94 0.97 -0.30* 0.97 0.96 -0.012
occupation=farmer 0.78 0.80 -0.05 0.96 0.98 -0.19 0.69 0.69 0.014 0.82 0.82 -0.024 0.63 0.75 -0.33 0.75 0.70 0.048
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
* p<0.1, ** p<0.05, *** p<0.01; Inter. = Respondent comes from intervention, Com. = Respondent comes from comparison site, Programme=Pooled sample
Table 5.1. 2: Comparison of general household characteristics at baseline, by programme and country
Inter. Com. Dif. Inter. Com. Dif. Inter. Com. Dif. Inter. Com. Dif. Inter. Com. Dif. Inter. Com. Dif.
Female headed HH 0.10 0.10 -0.03 0.08 0.09 -0.06 0.2 0.22 -0.053 0.09 0.08 0.07 0.06 0.05 -0.04 0.04 0.06 -0.13
Household size 7.36 7.18 0.11 5.9 5.75 0.19 5.81 5.97 -0.17 10.85 9.9 0.97*** 6.09 6.43 -0.29 8.03 8.25 -0.34
No. of children in HH 3.96 3.78 0.13 3.29 3.2 0.13 2.61 2.73 -0.13 6.31 5.58 0.73** 2.33 2.42 -0.3 5.01 5.06 -0.19
No. of adults in HH 3.40 3.40 -0.014 2.62 2.55 0.062 3.19 3.24 -0.045 4.55 4.32 0.24** 3.75 4.01 0.009 3.01 3.19 -0.15**
No. of productive adults in HH 3.18 3.22 -0.042 2.47 2.46 -0 2.93 3.01 -0.087 4.25 4.02 0.23*** 3.49 3.81 -0.04 2.9 3.04 -0.007
Years of education head 2.27 2.24 0.12 1.12 1.16 -0.05 7.03 6.86 0.17 0.61 0.74 -0.11 1.48 0.72 0.92** 0.86 0.88 0.003
Highest years of education of any adult in HH 4.62 4.75 -0.029 3.31 3.34 -0.14 9.73 9.58 0.14 3.61 3.88 -0.31 3.93 3.68 0.51** 2.21 2.42 -0.11
All adults in HH over 59 0.01 0.01 0.04 0.005 0.001 0.43 0.03 0.03 0.14 0.008 0.007 0.051 0.005 0.011 -0.40** 0.002 0.006 -0.38
Head is 60 or older 0.22 0.23 -0.025 0.13 0.11 0.1 0.33 0.32 0.049 0.262 0.247 0.048 0.238 0.316 -0.2 0.16 0.20 -0.12***
Literate adult in HH 0.52 0.52 0.027 0.52 0.51 -0.007 0.98 0.97 0.025 0.256 0.242 0.053 0.458 0.433 0.033 0.33 0.34 -0.009
Female literate adult in HH 0.30 0.30 0.019 0.20 0.22 -0.13 0.86 0.87 -0.038 0.141 0.125 0.073 0.159 0.128 0.15 0.076 0.06 0.18**
Head is literate 0.36 0.36 0.051 0.31 0.28 0.05 0.85 0.83 0.084 0.137 0.134 0.029 0.265 0.238 0.042 0.2 0.2 0.02
Head is productive 0.95 0.96 -0.027 0.96 0.97 -0.12 0.97 0.97 0.091 0.932 0.939 -0.065 0.907 0.91 0.028 0.98 0.98 -0.076
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
24
5.2: Economic activities and farm characteristics
The baseline status of economic activities undertaken by the sample households is summarised in Table
5.2.1. Burkina Faso and Niger stand out as the two countries with the highest number of unbalanced
covariates. Significantly more households in the intervention sites in Burkina Faso produce livestock
products and engage in casual and unskilled formal paid work while significantly more households in the
comparison sites engage in formal paid work compared to those in the intervention sites. In Niger,
significantly more households in the intervention than in the comparison sites process crops or natural
products and produce livestock products while more households in the comparison than in the
intervention sites operate off-farm businesses. In Ethiopia more households in the intervention sites
produce livestock products, although the difference is marginally significant, while in Mali significantly
more households in the comparison than in the intervention sites engage in formal skilled work.
Farm characteristics presented in Table 5.2.2 show that overall, all the covariates are balanced. However,
some variations exist at the country level, with Niger having the highest number of unbalanced covariates
where households in the intervention sites have significantly larger land sizes and tend to put more land
into cultivation compared to those in the comparison sites. In Kenya intervention households have more
land under irrigation compared to those in the comparison sites while the converse is true in Mali.
Households in Mali had relatively larger land sizes at the baseline compared to those in the rest of the
DryDev countries, while households in Ethiopia seem to be severely land constrained with average land
sizes per household that are slightly more than half of an hectare. As a result, nearly all the land owned
by the households in both the comparison and intervention sites was under cultivation. Likewise,
households in Ethiopia had more land under irrigation compared to the rest of the four countries.
25
Table 5.2. 1:Comparison of households’ economic activities at baseline, by programme and country
Farming 1.00 1.00 -0.051 1.00 1.00 0 0.99 0.99 0.11 1.00 1.00 - 1.00 1.00 - 0.99 0.99 0.1
Processing of crops or natural
0.15 0.15 0.016 0.01 0.01 0.25 0.08 0.07 0.077 0.15 0.12 0.15 0.55 0.65 -0.3 0.05 0.03 0.39***
products
Rearing livestock 0.71 0.71 0.027 0.66 0.61 0.078 0.90 0.91 -0.075 0.78 0.74 0.099 0.65 0.73 -0.27 0.55 0.52 0.11
Producing livestock products 0.17 0.16 0.13 0.15 0.10 0.26* 0.52 0.52 -0.006 0.06 0.02 0.62*** 0.06 0.06 -0.061 0.03 0.01 0.38***
Running off-farm business 0.19 0.20 -0.061 0.10 0.10 -0.004 0.28 0.28 0.002 0.16 0.17 -0.016 0.16 0.21 -0.07 0.24 0.29 -0.14**
Casual work 0.17 0.16 0.1 0.27 0.22 0.13 0.39 0.39 0.018 0.01 0.00 0.43* 0.03 0.03 0.082 0.07 0.06 0.087
Unskilled formal paid work 0.11 0.10 0.1 0.02 0.02 -0.037 0.28 0.26 0.039 0.17 0.11 0.24* 0.02 0.02 0.12 0.04 0.04 -0.055
Skilled formal paid work 0.04 0.04 0.049 0.01 0.01 0.13 0.12 0.13 -0.062 0.00 0.01 -0.53*** 0.04 0.01 0.46** 0.01 0.00 0
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
Table 5.2. 2:Comparison of farm characteristics at baseline, by programme and country

HH owns land 0.96 0.96 0.002 0.98 0.97 0.22 0.99 0.98 0.12 0.93 0.93 0.02 0.91 0.95 -0.46 0.99 0.99 0.25
Total size of land owned(ha) 4.08 3.89 0.17 0.59 0.55 0.01 4.40 4.29 0.062 3.66 3.96 -0.34 10.17 8.98 -0.004 3.95 3.66 0.57***
Size of land under cultivation
3.05 2.78 0.25 0.58 0.54 0.01 1.34 1.32 0.002 3.31 3.13 0.15 8.59 7.48 0.25 3.71 3.51 0.42***
(ha)
HH irrigated land 0.12 0.14 -0.062 0.24 0.27 -0.086 0.07 0.07 -0.048 0.07 0.07 0.038 0.08 0.09 -0.14 0.13 0.16 0.078
Size of land irrigated (ha) 0.13 0.13 -0.001 0.04 0.05 -0.002 0.10 0.07 0.033** 0.08 0.07 0.007 0.23 0.34 -0.16** 0.23 0.22 0.068
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
26
5.3. Application of soil and water management options at baseline
Table 5.3.1 shows the extent to which the intervention and comparison households are balanced in
relation to application of soil and water management options at baseline. In the baseline survey, farmers
were asked if they used a set of soil and water management options on their main cropping field, the
duration they had used such practices and the extent to which they had applied them on the main farm.
The practices were classified into three categories depending on their main function: practices that aim
at (i) reducing erosion and conserving soil moisture —soil and water conservation options (ii) improving
soil fertility — soil fertility management option, and (iii) harvesting water for irrigation —water harvesting
options. The practices on which data were collected during baseline survey include soil and water
conservation options like terraces, zai pits, tree planting and/or management, mulching, trenches, check
dams/rock dams, half-moons, cover cropping, minimum tillage; water harvesting options like wells, farm
ponds, and soil fertility management options such as fertilizer application, manuring , improved fallows,
fertilizer trees among others. It is important to note that some practices are likely to perform more than
one function, for example zai pits, minimum tillage and tree planting and/or management. However, the
classification was based on the main function performed by each option. An integrated soil and water
management option entails a household combining either soil and water conservation option with soil
fertility management option or water harvesting and soil fertility management options.
Overall, all the assessed covariates are balanced at the programme level. In Kenya and Mali significantly
more households in the intervention sites than in the comparison sites practiced water harvesting and
soil and water conservation, respectively. Ethiopia, Niger and Burkina Faso have the highest number of
unbalanced covariates. Significantly more households in the intervention sites than in the comparison
sites practice soil and water conservation and have significantly higher proportions of their farm under
integrated soil and water management options. on the contrary, significantly more households in
Ethiopia’s comparison sites than in the intervention sites practice water harvesting. Similarly, significantly
more intervention than comparison households practice water harvesting while more households in the
comparison sites than in the comparison sites practice integrated soil and water management and have a
significantly higher proportion of the main farm under integrated soil and water management options. In
Niger, significantly more households in the intervention than in the comparison sites practice integrated
soil and water management and soil fertility management and have significantly higher proportion of the
main farm under integrated soil and water management options.
27
Table 5.3. 1: Application of soil and water management options at baseline, by programme and country
HH practiced integrated SWM 0.67 0.64 0.10 0.78 0.72 0.19 0.49 0.47 0.06 0.73 0.76 -0.12* 0.92 0.83 0.28 0.46 0.42 0.15***
HH practiced SWC 0.9 0.87 -0.01 0.79 0.71 0.26** 0.89 0.9 -0.05 0.94 0.95 -0.14 0.97 0.9 0.43*** 0.97 0.94 0.17
HH practiced SFM 0.73 0.72 0.16 0.94 0.94 0.03 0.52 0.5 0.04 0.77 0.79 -0.071 0.94 0.9 0.09 0.46 0.44 0.12**
HH practiced water harvesting 0.17 0.14 0.04 0.17 0.23 -0.17** 0.09 0.05 0.31** 0.16 0.1 0.29*** 0.32 0.22 0.23 0.13 0.14 0.13
HH practiced at least 1 SWM 0.97 0.97 0.11 0.98 0.98 -0.04 0.91 0.93 -0.11 0.99 0.99 0.029 0.99 0.98 0.14 0.98 0.98 0.014
% of main farm under integrated SWM
37.73 34.07 3.72 46.18 37.91 7.45** 26.3 25.16 1.23 31.91 35.48 -3.34** 66.58 56.45 5.25 22.4 18.76 3.71**
options
% of main farm under at least 1 SWM
68.79 66.91 1.84 73.31 70.68 1.25 58.8 59.43 -0.55 59.99 63.75 -3.17 85.76 78.07 6.05 70.2 65.02 2.34
option
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
SWM=soil and water management options; SWC =soil and water conservation( practices such as trenches, terraces, soil, earth or stone bunds/belts, zai pits/planting basins, tree planting and/or management, pasture reseeding,
minimum tillage, check dams, planting pits, half-moons, contour bunds, cover cropping, controlled grazing); SFM=soil fertility management options (manuring, composting, microdosing, crop rotation, intercropping, fertilizer
application, fertilizer trees or improved fallows, fertility trenches); water harvesting options (farm ponds, roof catchment, road run off/flood diversion, canal, retention ditches/basins/cut off drains, micro basins, hand dug wells,
water harvesting check dams, on-farm water pans) ; Integrated SWM options= at least 1 SWC option and SFM option or water harvesting option + SFM option
28
5.4. Comparison of outcome variables at baseline
Aside from respondent and household covariates, we tested if the intervention and comparison groups
were still balanced on key outcome variables. Tables 5.4.1 and 5.4.2 present the results of balance test
performed on the selected baseline outcome variables. Production and market outcome variables are
presented in Table 5.4.1 where the results show that the variables are balanced at the programme and
across all the five countries, except for Burkina Faso where the gross and net cash value of crops produced
by intervention households are significantly higher than the values in the comparison sites.
The results in Table 5.4.2 show that nearly all the baseline outcomes are balanced at programme level. At
the country level, however, there are significant differences between the intervention and comparison
households on between 2 and 7 of the 16 outcome variables assessed. Ethiopia has the highest number
of unbalanced outcome variables, with varying degrees of significance. In nearly all the outcome variables,
households in the intervention sites in Ethiopia are significantly better off than those in the comparison
sites, except for one outcome where significantly more men in the comparison households consumed
more diverse food groups compared to households in the intervention sites. Households in the
intervention sites in Ethiopia were better off with regards to outcomes relating to local governance and
institutional functioning, women empowerment, food diversity among women and household resilience.
In Niger and Burkina Faso, households in the intervention sites seemed to have started at a significantly
better level than households in the comparison sites in relation to outcomes for household resilience and
food diversity. Similarly, households in the intervention sites in Burkina Faso are significantly relatively
wealthier than those in the comparison sites while the converse is true for Niger. In Kenya, significant
differences between the intervention and comparison households is observed in access to extension and
empowerment of women among men only samples. For the Mali sample, households in the intervention
and comparison sites are almost comparable on all the 16 baseline outcome variables, except for two
outcomes on poverty status and women empowerment where small significant differences are noted.
In general, there is still good balance between the two groups in relation to baseline covariates and
outcomes at programme level, even after a reduction in the original sample size by about 33%, perhaps
due to the large number of clusters. However, at country level, the results are mixed with Burkina Faso
and Niger in the Sahel and Ethiopia in the Horn of Africa having the highest number of unbalanced
covariates and outcome variables. Whereas some significant differences between the groups were noted
even at the baseline in each of the five countries, sample attrition may have contributed to a greater
extent to the unbalanced covariates, especially in Niger. However, the difference-in-differences
evaluation method adopted in this study permits comparison in outcomes before and after an
intervention between groups by controlling for bias from unobserved variables that remain fixed over
time. This is contrary to other evaluation methods such as randomized control trial (RCT) designs, which
require comparable groups that are well balanced on multiple factors which may influence the outcome
in order to evaluate treatment effects of an intervention.
29
Table 5.4. 1: Comparison of production and market outcomes at baseline, by programme and country
Gross cash value of crops
570.2 491.4 72.4 317.3 302.2 0.16 412.8 414.7 -1.26 466.9 396.5 56.7** 1539.1 1127.6 145.6 393.4 429.5 44.9
produced (US$)
Net cash value of crops
464.6 405.6 48.5 245.9 242.5 -8.13 323.8 341.6 -17.3 406.4 338.6 55.4* 1227.8 900.8 100.2 341.7 376.1 33.4
produced (US$)
Gross cash value of crops sold
226.6 179.8 44.7 66.4 88.6 -23.6 250.8 250.3 -0.23 80.8 65.3 14.4 786.9 436.5 115.0 106.6 129.6 17.0
(US$)
Net crop profits (US$) 185.4 151.8 32.4 56.5 78.8 -23.1 235.1 237.4 -3.17 69.7 55.3 13.4 589.9 318.9 109.4 89.3 110.7 12.0
Sample size 2400 2833 5233 564 690 1254 509 642 1151 512 575 1087 366 423 789 449 503 952
30
Table 5.4. 2:Comparison of selected outcome variables at baseline, by programme and country
Burkina
Programme Ethiopia Kenya Mali Niger
Faso
Local governance and institutional functioning
Farmers participating in farmer

organisations and reporting 0.33 0.31 0.10 0.38 0.28 0.27*** 0.60 0.60 0.02 0.30 0.27 0.06 0.23 0.20 0.10 0.10 0.12 -0.10
benefits
HH receiving extension more
0.29 0.27 0.08 0.70 0.60 0.26** 0.13 0.09 0.25** 0.12 0.16 -0.20 0.24 0.24 -0.16 0.18 0.20 -0.03
than 1 times in 12 months
Poverty status
HHs above national poverty line 0.62 0.60 0.08 0.75 0.74 0.01 0.85 0.82 0.15 0.39 0.36 0.03 0.68 0.56 0.31* 0.42 0.46 0.00
Wealth index 0.49 0.49 0.00 0.63 0.62 0.00 0.43 0.40 0.03 0.51 0.44 0.05** 0.66 0.76 -0.14 0.24 0.28 -0.04**
HH multi-dimensionally poor 0.88 0.89 -0.089 0.98 0.99 -0.12 0.61 0.63 -0.06 0.92 0.94 -0.22** 0.92 0.95 -0.27 0.97 0.96 0.15
Multidimensional index (MPI) 0.90 0.91 -0.01 0.98 0.99 0.00 0.69 0.71 -0.02 0.94 0.96 -0.02 0.93 0.96 -0.03 0.98 0.97 0.01
Food security (Minimum dietary diversity-MDD)
HHs consuming 5 or more food

0.19 0.18 0.04 0.01 0.01 0.00 0.29 0.32 -0.10 0.13 0.11 0.17** 0.19 0.16 0.16 0.38 0.35 0.13***
groups
Women consuming 5 or more
0.21 0.19 0.04 0.02 0.00 0.68** 0.26 0.30 -0.11 0.14 0.10 0.20 0.18 0.16 -0.01 0.46 0.45 0.03
food groups
Men consuming 5 or more food
0.18 0.17 0.04 0.01 0.02 -0.51** 0.31 0.35 -0.10 0.13 0.11 0.15 0.21 0.15 0.27 0.23 0.27 0.23***
groups
Women empowerment
Women scoring positively on
0.09 0.10 0.02 0.04 -0.01 0.34 0.33 0.04 0.05 0.03 0.02 0.02 0.02
2/3 of weighted indicators 0.04 0.06 0.06 0.13 -0.17
Men scoring positively on 2/3 of
0.37 0.34 0.29 0.21 0.42 0.40 0.54 0.52 0.20 0.15 0.35 0.39
weighted indicators 0.15*** 0.29*** 0.10 0.11 0.18* 0.07
WEAI -Entire sample 0.50 0.49 0.01 0.46 0.43 0.02 0.64 0.63 0.01 0.54 0.53 0.00 0.43 0.40 0.03 0.38 0.43 -0.03
WEAI -women 0.37 0.38 0.00 0.35 0.35 0.001 0.60 0.62 -0.02 0.34 0.33 0.01 0.30 0.30 -0.01 0.24 0.26 -0.01
WEAI-men 0.62 0.59 0.03 0.56 0.51 0.034* 0.69 0.64 0.05** 0.73 0.71 0.00 0.54 0.48 0.06 0.58 0.61 0.00
Household resilience
Resilient HHs 0.05 0.04 0.17 0.10 0.07 0.23* 0.09 0.08 0.11 0.00 0.01 -0.35 0.02 0.01 0.32 0.01 0.00 0.59*
Resilience index 0.44 0.43 0.01 0.47 0.43 0.03 0.51 0.51 0.00 0.38 0.35 0.02** 0.44 0.43 0.003 0.38 0.39 0.00
Sample size 2418 2858 5276 569 691 1260 510 642 1152 512 575 1087 378 446 824 449 504 953
31
6. Impact of DryDev on selected indicators
As mentioned earlier, the DryDev programme’s overall goal was to improve and sustain smallholder
farmers’ food and water security, livelihoods and resilience and empower women and other
disadvantaged groups in the drylands of the Horn of Africa and the Sahel. Implementation of DryDev set
in motion a chain of events through 8 interrelated work packages guided by two overarching and
reinforcing theories of change. We therefore distinguish between intervention short-term outcomes
(changes in behaviour of farmers and institutions resulting from the DryDev activities), intermediate
outcomes (changes in productivity and market integration) and impacts (effect on poverty, food security,
women empowerment and household resilience resulting from the outcomes). We begin by presenting
impacts, then return to intervention outcomes in the subsequent sections to identify the mechanisms
linking the DryDev interventions to impact.
6.1. Impact on dietary diversity
Impact on food security: Indicator Initial Impact DryDev Impact (DiD)

% senior men and senior target
women in the households
+1%
(HH) consuming 5 or more of
>7 % average all sites for men/women
the Minimum Dietary Diversity

Above comparison
+15%
food groups Women in Kenya
The impact of the interventions on food security, as measured by the proportion of farmers consuming
more than four food groups among households in the intervention and comparison group is mixed and
varies by country and gender category. Whereas there was a positive change in the proportion of farmers
consuming more than 4 food groups at the programme and country level, the change was significantly
higher among women in the intervention sites in Kenya only and no significant improvement was
observed in the entire sample at the programme and country level. In Kenya, being in the intervention
sites increased the likelihood of consuming more than four food groups among women by 15 percentage
points.
Unlike in Kenya, where dietary diversity scores were relatively higher at the baseline, the remaining four
countries had extremely low dietary diversity scores, except in Niger where the proportion of farmers
consuming more than 4 food groups was relatively high in the intervention and comparison sites (Figures
6.1.1, 6.1.2, and 6.1.3). Where food availability and access are a major concern as was the case in the rest
of the countries, farmers tend to put more efforts on meeting calorific needs as opposed to diversifying
their diets (Colen et al. 2018). In addition to meeting dietary needs from own production, studies have
shown that increased access to food through production is not sufficient to induce changes in dietary
diversity(Koppmair et al. 2017). It is recommended that interventions aiming at improving nutrition
security incorporate specific nutrition education component as part of the interventions.
32
100% 100%
90% 90%
80% 80%
70% 70%
% of households
% of women
60% 60%
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
Programme Burkina Faso Ethiopia Kenya Mali Niger Programme Burkina Faso Ethiopia Kenya Mali Niger
Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1 Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1
Figure 6.1. 1: Proportion of HHs consuming 5 or more food groups at baseline and endline Figure 6.1. 2: Proportion of women consuming 5 or more food groups at baseline and
endline
100%
90%
80%
70%
60%
% of men
50%
40%
30%
20%
10%
0%
-10% Programme Burkina Faso Ethiopia Kenya Mali Niger
Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1
Figure 6.1. 3: Proportion of men consuming 5 or more food groups at baseline and endline
33
6.2. Impact on poverty reduction
Impact on poverty Indicators Initial Impact DryDev impact

reduction: targets (DiD)
% HH above national
poverty line 5% above
comparison No significant
differences at
% HH above median programme level
asset index comparison 5% above M
group comparison MPI 5% below for
Burkina Faso
% HH multidimensional 5% below
poor comparison
DryDev’s impact evaluation on poverty reduction was based on three poverty indices: headcount ratio
(percentage of households above national poverty line of US$1.9 per capita per day), multidimensional
poverty index (MPI)12 and household assets index. DryDev has operated in very poor regions, as data
shows that 9 people out of ten living in DryDev’s areas of intervention were multidimensionally poor at
the start of the programme (Figure 6.2.1).
100%
90%
% of dimensionally poor HHs
80%
70%
60%
50%
40%
30%
20%
10%
0%
Programme Burkina Faso Ethiopia Kenya Mali Niger
Intervention-T0 Comparison-T1 Intervention-C0 Comparison-C1
Figure 6.2. 1: Proportion of multidimensionally poor households at baseline and endline
Overall, the proportion of households living above the national poverty line declined by 10% and 11%
respectively in the intervention and comparison sites (Figure 6.2.2). In some countries, however, like Mali
12
Multidimensional poverty index (MPI) is a compilation of ten weighted indicators, showing how people may be
left behind in terms of health, education and standard of living. Persons that are deprived in a third or more of these
weighted indicators are considered poor. More about MPI on UNDP Human Development Report, 2019.
34
and Kenya, an impoverishment was observed. In Mali, the decrease was even sharper, about 53% and
49% in the intervention and comparison sites, respectively. This could be attributed to period, high levels
of insecurity were witnessed in most parts of Mali during the implementation, which could have negated
returns from the interventions by preventing farmers from engaging in economic activities. In Kenya, the
proportion of dimensionally poor households increased slightly in both the intervention and comparison
sites. In 2019, more Kenyan households countrywide faced economic hardships due to loss of job
opportunities as many private investors went under because of the high cost of doing business in the
country. This could perhaps explain the general increase in the number of poor households in both the
sites.
100%
90%
80%
% of households
70%
60%
50%
40%
30%
20%
10%
0%
Figure 6.2. 2: Proportion of households above national poverty line at baseline and endline, by programme and
country
A closer look at household asset index shows that both the intervention and comparison households
accumulated more assets over time, albeit with varying magnitude except in Ethiopia where the index
declined in both sites. Likewise, the change in percentage of households with asset index above the
median for comparison group was higher by 7% in the intervention than in the comparison group,
although the difference was not statistically significant.
The DiD results for the three indices are consistent across the countries and at programme level and show
that there was no significant effect on poverty reduction at the programme and country levels (Table
6.2.1). The results suggest that the positive change in the proportion of households living below national
poverty line in the DryDev sites at the programme level, as well as in Burkina Faso and Ethiopia is not
significantly larger than that in the comparison sites. Similarly, changes in multidimensional poverty index
and asset accumulation observed in the intervention sites are only marginally different from those in the
comparison sites.
35
Table 6.2. 1: Difference-in-differences results for three measures of poverty
Headcount ratio Multidimensional poverty index Asset index
Std.
DiD error DiD Std. error DiD Std. error
Programme 0.056
0.003 -0.001 0.026 0.042 0.060
Burkina Faso 0.104 0.097 0.014 0.036 -0.032 0.091
Ethiopia 0.045 0.057 0.012 0.008 0.061 0.037
Kenya -0.022 0.034 -0.004 0.034 0.022 0.023
Mali -0.150 0.101 -0.015 0.018 0.19 0.13
Niger -0.025 0.063 0.0003 0.013 -0.021 0.091
* p<0.1, ** p<0.05, *** p<0.01; standard errors are clustered at sub-watershed level
It is evident from the results that macroeconomic and security context can have adverse effects on
poverty reduction programmes. Aside from increased productivity, public effort in the form of
reinforcement of education and social infrastructure (e.g. safety net programmes) is required to realise
poverty reduction goals in the marginal areas of sub-Saharan Africa where chronic poverty is most
concentrated. Besides, development programmes focusing on farm productivity gains require sufficient
time for productivity gains to materialise and be translated to reduced poverty. For instance, a 42%
decline in the percentage of poor households was reported in Kaartenga village, Sanmatenga province in
Burkina Faso where soil and water conservation practices were implemented over a period of 20 years
between 1980 and 2001(Reij, 2005).This is particularly true for natural resource management (NRM)
interventions that tend to have lagged impacts, which could be a disincentive for the resource constrained
target group to engage. As demonstrated in (), uptake of practices such as agroforestry among poor
smallholder farmers remains low because such farmers’ have the priority of getting food and cannot
afford to invest their time and labour in new technologies, which have uncertain benefits. In the short-
term, there may be increases in the number of households undertaking soil and water conservation
practices including soil fertility management, but time may be needed to increase yields to a sustainable
level. Therefore, while participatory approaches are critical when designing long-term NRM programmes
that require buy-in from policy makers, there is need to strike a balance between the planning and the
actual implementation phase.
Their findings showed that agroforestry fails to be taken up by the ‘poorest of the poor’, whose main
priority is to get food on the table and who cannot afford taking risks by investing time and labour in new
technologies which have uncertain benefits in the long term
36
6.3. Impact on Women empowerment
Impact on women Indicator Initial Impact DryDev impact (DiD)

empowerment target
% men and women Not significant
scoring positively on at programme level
2/3 of weighed
indicators of the
>10%
above
Significant increase
Women
comparison for Ethiopia (men/women)
Empowerment in and Kenya (women only)
group
Agriculture Index
(WEAI)
As part of DryDev social inclusion strategy, efforts were made to improve women’s involvement in
agriculture including empowering them to make decisions regarding production, control over income and
productive resources as well as their general wellbeing. The five countries used different approaches that
were suited to their context including reviewing existing laws and regulations that govern land tenure,
identifying and developing women targeted value chains, as well as providing priority slots to women in
any capacity building events.
In nearly all the countries, 68% of the participants had to be women. In some countries such as Kenya,
gender transformative approaches that emphasise the benefits of intra-household equity were developed
and piloted. All members of the household were taken through a training that enabled them to appreciate
the role played by each gender (e.g. men, women, youth). We assessed the extent to which the
interventions empowered women to make decisions regarding production, take up leadership positions
and own and control resources including income using the women empowerment in agriculture index
(WEAI)13. The index ranges from 0 to 1, with values tending towards zero indicating high levels of
disempowerment and vice versa.
13
The WEAI looks at five domains of empowerment for women : (i)production (autonomy in production and
production decision-making), (ii) resources (ownership and control over assets, and access to credit, (iii)
income(control over income),(vi) leadership (self-efficacy, confidence in public speaking, active group member) (v)
leisure and work (low stress and leisure time). More information on WEAI website http://weai.ifpri.info/
37
Figure 6.3.1 and 6.3.2 show the contribution of each dimension of the WEAI to the overall index at the
baseline and endline for the entire sample and women only sample. The results suggest that there was an
improvement in the overall empowerment index at the programme and country level. The DiD results on
women empowerment are mixed, with significant impacts being reported at the country level among
women in Kenya and Ethiopia. The findings are inconclusive for the entire sample at the programme and
country level, except for Ethiopia where the results show a weak significant impact (Table 6.3.1)
Empowerment index increased by 5% points for women in the intervention sites in Kenya and Ethiopia.
Table 6.3. 1: DiD results for women empowerment in agriculture index

Programme Burkina Ethiopia Kenya Mali Niger
WEIA-men&women 0.02(0.02) 0.02(0.05) 0.04*(0.02) 0.01(0.02) 0.03(0.05) 0.04(0.04)
Sample size 5276 1076 1255 1142 807 920
WEIA-women only 0.03(0.021) 0.01(0.06) 0.05**(0.02) 0.05**(0.02) 0.021(0.03) 0.040(0.04)
Sample size 2702 524 662 603 379 503
WEIA-men only 0.02(0.03) 0.03(0.09) 0.03(0.03) -0.03(0.03) 0.031(0.07) 0.03(0.07)
Sample size 2578 552 593 539 428 417
A closer examination of the five dimensions revealed that in Kenya and Ethiopia the greatest change
among women was a better control over income (Figure 6.3.2). Although more empirical evidence is
required to corroborate this proposition, it appears that the gender transformative approach piloted in
Kenya coupled with involvement of more women in the programme as trainers of trainees could have
resulted in more women being able to make decisions regarding income generated from productive
activities. Similarly, more women were involved in the saving schemes in both Ethiopia and Kenya,
which could have contributed to their financial independence.
38
Niger(C1) Niger(C1)
Niger(C0) Niger(C0)
Niger(T1) Niger(T1)
Niger(T0) Niger(T0)
Mali(C1) Mali(C1)
Mali(C0) Mali(C0)
Mali(T1) Mali(T1)
Mali(T0) Mali(T0)
Kenya (C1) Kenya (C1)
Kenya (C0) Kenya (C0)
Kenya (T1) Kenya (T1)
Kenya (T0) Kenya (T0)
Ethiopia (C1) Ethiopia (C1)
Ethiopia (C0) Ethiopia (C0)
Ethiopia (T1) Ethiopia (T1)
Ethiopia (T0) Ethiopia (T0)
Burkina Faso(C1) Burkina Faso(C1)
Burkina Faso (C0) Burkina Faso (C0)
Burkina Faso (T1) Burkina Faso (T1)
Burkina Faso (T0) Burkina Faso (T0)
Programme(C1) Programme(C1)
Programme(C0) Programme(C0)
Programme (T1) Programme (T1)
Programme(T0 ) Programme(T0 )
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Production Resource Income Leadership Leisure&low stress

Production Resource Income Leadership Leisure&low stress
Figure 6.3. 2: Dimensions of WEAI for the entire sample, by programme and country Figure 6.3. 1: Changes in dimensions of WEAI for women only sample, by programme and
country
39
6.4. Impact on household resilience
Impact on household Indicator Initial Impact target DryDev Impact (DiD)

resilience:
% households scoring
positively on 2/3 of >10% +3%
weighed indicators of above comparison group
the Resilience index
DryDev promoted climate smart technologies and practices that were appropriate for the local context in
order to improve household livelihood resilience as well as the local environment. We assessed whether
the interventions resulted in an improvement in household resilience through an index consisting of five
dimensions: livelihood viability, innovative potential, access to contingency resources and support and
integrity of natural and built environment.
Like the WEAI, resilience index values range between 0 and 1, with the values tending to zero indicating
less resilience and vice versa. Households less resilient are more vulnerable to shocks because of less
diversified livelihood strategies, low innovative potential, limited access to productive resources and
support and/or being in more fragile environments that cannot sustain production. The results suggest
that there was positive, albeit marginal change in resilience index at the programme level as well as in all
the countries, except in Ethiopia, where the decline in the proportion of resilient households in the
comparison sites was significantly larger than in the intervention sites. Households in the intervention
sites were exposed to more diversified livelihood strategies, as well as more sustainable production
practices, including low cost water harvesting techniques, which cushioned them from adverse shocks
unlike their counterparts in the comparison sites. At the programme level, the percentage of households
scoring positively for two-thirds of the index increased by 4% compared to the one percent increase
observed in the comparison sites. A similar result is depicted at the country level except in Ethiopia, where
there was a decline, albeit less sharp in the DryDev sites by 6%.
40
18%
16%
% of HHs scoring positively
14%
12%
10%
8%
6%
4%
2%
0%
-2% Programme Burkina Faso Ethiopia Kenya Mali Niger
Figure 6.4. 1: Comparison of resilient HHs at programme and country level between intervention and
comparison sites
Despite the positive change in the percentage of resilient households in the intervention sites, the number
is very low across the programme. The highest share of resilient households, in Ethiopia, is less than 15%
which is consistent with the widespread poverty situation. DryDev target populations, like many families
living in the drylands, live in constant hardship and are exposed almost every year to agro climatic and
social shocks. It reinforces the major importance of development interventions such as DryDev.
6.5. Probing the mechanisms

6.5.1. Impact on local governance and farmer organisations’ functioning
Strong local institutions and well-functioning governance structures are prerequisite for enhanced uptake
of NRM interventions and may act as an incentive for farmers to invest in lucrative value chains. Well-
functioning institutions and governance systems restore confidence in economic agents, thereby enabling
businesses to thrive( Shiferaw et al. 2009). Economies of scale, which is a necessary but not sufficient
condition for consolidating market share and bargaining power among producers requires the existence
of strong farmer organisations (Shiferaw et al. 2011). The DryDev programme focused on developing the
capacity of duty bearers in local institutions such as farmer organisations, sub-watershed committees,
government staff (e.g. extension agents) and any other grass root organisation that worked directly with
the farmers. In other areas grass root organisations, such as farmer organisations, were established and
strengthened (see End of Programme Report). The interventions, which varied across the five countries
were expected to improve access to and quality of services rendered to the farmers to enable them to
take up improved soil and water management options and invest in the priority value chains. This section
aims at assessing the impact of the interventions on the functioning of farmer organisations and access
to extension services.
41
The DiD results suggest that access to extension services improved significantly in all the five countries
except in Niger, where the frequency with which farmers accessed extension services in the intervention
sites improving by 12 percentage points at the programme level. The magnitude of the effect was greater
in Mali and Burkina Faso where being enrolled in the DryDev programme increased the probability of
accessing extension services by 18 and 17 percentage points, respectively. In Ethiopia and Kenya, being in
the programme sites increased the likelihood of accessing extension services by 10 and 7 percentage
points, respectively. Until 2017, farmers were not satisfied with the availability of duty bearers and the
quality of extension services. Ethiopia was the exception as the country has a strong government
extension services and DryDev’s programme collaborated closely with the public extension institutions
for the implementation of trainings and co-learning. As shown in Figure 6.5.1, the proportion of farmers
accessing extension services in the intervention sites in Ethiopia was slightly more than twice as high as
the programme average at baseline. After the intervention, the proportion of farmers receiving extension
services declined in both the intervention and comparison sites in Ethiopia. In contrast, the proportion of
farmers receiving extension services is relatively small despite significant improvement in access to
extension services in Kenya and Burkina Faso.
100%
90%
80%
70%
% of households
60%
50%
40%
30%
20%
10%
0%
Figure 6.5. 1: Proportion of HHs receiving extension services at baseline and endline at programme and
country level
DryDev opted for farmer-centered, scalable extension systems in the other four countries. In Kenya where
extension responsibilities had just been transferred to the county level without adequate resources,
farmer-to-farmer extension approach was adopted, and joint trainings were organised with lead farmers
and the county government extension staff. In Burkina Faso, collaboration between the farmer network
organisation and local extension agents was enhanced through joint capacity building and co-learning
events. In Niger, where the partners adopted innovation platforms as a delivery strategy, more effort and
42
time was required to rethink the methodology as the lack of farmer organisations meant farmers did not
benefit fully from the existing top-down public extension system.
Farmer organisations (FO) were the backbone of DryDev’s programme delivery mechanism. One difficulty
when working with existing FOs who did not initially share DryDev’s vision, was resistance to change and
adopt to new ways. The 2017 uptake survey results indicated that a small percentage of the farmers were
active members of FOs. Thereafter, each country supported more inclusive farmer groups, investing in
developing new visions, business plans and undergoing training. In sites where there were no farmer
organisations, farmers were encouraged and guided to form and register interest groups, in order to have
a legal persona that can transact with other legal entities like commercial banks and buyers.
There was an increase in the number of farmers joining active groups in both the comparison and
intervention sites in all the countries but no significant difference between intervention and comparison
sites (Figure 6.5.2). While there are good success stories like Ngengi self-help group in Kenya (see End of
Programme Report), it takes considerable time to build effective capacity of FOs and trust among farmers
through demonstrable success, benefits, and accountable leadership. Therefore, not every farmer was
motivated to join farmer groups.
100%
90%
80%
70%
% of households
60%
50%
40%
30%
20%
10%
0%
Figure 6.5. 2: Proportion of farmers participating in farmer organisations and reporting benefits at baseline and
endline, by programme and country
6.5.2 Impact on adoption of selected soil and water management options
A cursory look at adoption indicators suggests that improved access to extension services may have
resulted in a significant increase in adoption of soil and water management options at the programme
level as well as in Ethiopia, while no significant statistical difference was observed in Kenya, Burkina Faso,
43
Mali and Niger. The results suggest that being in the intervention sites increases the likelihood of adopting
any soil and water management option by 2 percentage points at the programme level and 7 percentage
points in Ethiopia. The planned comparisons fostered co-learning among farmers which may have led to
a significant, yet small increase in adoption of soil and water management options in some intervention
sites like in Ethiopia. Exposure to a wide range of contextually appropriate options is not enough to trigger
sustained change in soil and water management as there are other prerequisites for uptake like market
linkages for inputs and outputs and acquisition of technical skills.
Scaling of natural resource management practices promoted in the DryDev Programme was informed by
three approaches, namely bottom-up, integrated and inclusiveness. The scaling of soil and water
management options was premised on an integrated approach in which a household was presented with
three categories of options: soil and water conservation, soil fertility management practices and water
harvesting options. The difference-in-differences results indicate that despite the increase in access to
extension, there was no significant impact on the uptake of integrated soil and water management
options at the programme and country level. Instead, a decline in the proportion of households practicing
integrated SWM options (Figure 6.5.3), as well as the proportion of land under integrated SWM options
was observed in the intervention sites (Figure 6.5.4).
100%
90%
% of households practising
80%
70%
60%
50%
40%
30%
20%
10%
0%
Figure 6.5. 3: Proportion of HHs practicing integrated soil and water management options at baseline and
endline, by programme and country
44
100
% of main cropping field 90
80
70
60
50
40
30
20
10
0
Figure 6.5. 4: Proportion of the main cropping field under integrated SWM options at baseline and endline, by
programme and country
The results appear to suggest that the concept of integrated soil and water management did not fit all
households. Regardless of the observed impact of integrated soil and water management options on soil
productivity during the field trials, farmers tend to favour practices that suit their resource needs.
Households may favour options that are less labour and capital intensive as the return on investment for
any change of farming practice is still a risky gamble for any vulnerable smallholder family. Follow up
qualitative research on the reasons for disadoption of integrated soil and water management option
would be highly valuable to inform the design of future land restoration programmes.
A more detailed analysis of uptake data of soil and water management practices suggest a greater uptake
for low tech soil and water conservation options like terracing and trenches (Figure 6.5.5), compared to
more capital-intensive options such as soil fertility management (Figure 6.5.6) and water harvesting
structures (Figure 6.5.7). The results partly explain the marginal, albeit insignificant increase of 4% in the
proportion of households practising irrigation (Figure 6.5.8) above the change observed in the comparison
group and the insignificant change in the proportion of land under irrigation. Instead, a decline in the
proportion of land under irrigation is observed in both the comparison and intervention sites (Figure
6.5.9).
45
100% 100%
90% 90%
80% 80%
70% 70%
60% 60%
50% 50%
40%
40%
30%
30% 20%
20% 10%
10% 0%
0% Programme Burkina Ethiopia Kenya Mali Niger
Programme Burkina Faso Ethiopia Kenya Mali Niger Faso
Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1 Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1
Figure 6.5. 5:Proportion of HHs practicing soil fertility management at baseline and
Figure 6.5. 6: Proportion of HHs practicing SWC at baseline and endline
endline
100%
90%
80%
% of households
70%
60%
50%
40%
30%
20%
10%
0%
Programme Burkina Ethiopia Kenya Mali Niger
Faso
Figure 6.5. 7: Proportion of HHs practicing water harvesting at baseline and endline
46
50% 14%
% of households
% of main farm irrigated

40% 12%
10%
30% 8%
20% 6%
4%
10% 2%
0% 0%
Intervention-T0 Intervention-T1 Intervention-T0 Intervention-T1

Comparison-C0 Comparison-C1 Comparison-C0 Comparison-C1
Figure 6.5. 9: Proportion of HHs practicing irrigation at baseline Figure 6.5. 8: Proportion of main cropping field under irrigation at
and endline baseline and endline
One assumption of the DryDev interventions was that creating farmers’ awareness on promising soil and
water management options and equipping them with the necessary skills to implement SWM practices of
their choice would lead to adoption. However, other adoption barriers related to economic access may
prevail. For instance, water harvesting options like farm ponds were identified as very promising in the
DryDev countries, but resource poor smallholder farmers need financial support to invest in this
technology. Microfinance institutions (MFI) were interested but needed evidence of return on investment
before engaging in this new business. DryDev developed a farm pond decision-support tool to provide
such information but the programme ran out of time to influence investors like the County Government.
It is recommended to pursue this dialogue between farmers, private sector and local government to
create the necessary financial linkages for the uptake of such effective technology.
6.5.3. Impact on production, market integration and profitability
The DryDev’s theory of change envisioned an increase in the production of profitable and climate smart
crops through integrated pathways. Farmers would be informed, trained and convinced to adopt
improved technologies like better seeds, good soil fertility management options and water conservation
and harvesting techniques. The potential of their farmland would then improve. This pathway was critical
for Ethiopia and Kenya where sizes of crop land per farm are relatively smaller and hence enhancing
productivity by putting more land into cultivation was not feasible. In the Sahel countries, however, farm
sizes were relatively larger, and extensively degraded, necessitating restoration to increase productivity
and generate marketable surplus. DryDev also linked farmers to inputs and outputs markets, as well as
financial services, and the farmer groups were trained in financial literacy to improve their repayment
capacity. Existing village savings and loans associations were strengthened to act as sources of capital to
farmers who could not access credit through the formal financial markets.
6.5.3.1. Impact on production
Overall, the impact assessment results suggest that there was no significant change in gross cash value of
crops harvested at the programme and country level (Table 6.5.1). The results, however, show that gross
value of crops harvested in Ethiopia and Kenya increased on average in both the comparison and
interventions sites, while decreasing in the Sahel. There was a sustained declined in land under cultivation
in nearly all the countries, except in Ethiopia, with the largest decline of about 50% being recorded in Mali
in both the intervention and comparisons households. This could perhaps explain the dramatic decline in
production in some countries like Mali where land under cultivation reduced by slightly more than a half
in the intervention and comparison sites. While the reduction in land sizes under cultivation in Mali could
be an indication of land sub-division due to population growth, it is highly probable that high incidences
of insecurity experienced in most parts of the country could be the main cause. In some countries, the
decline in land under cultivation could have been due to natural causes such as sub-division in response
to population growth.
Table 6.5. 1: Change in average gross cash value of crops produced (US$), by programme and country
Intervention Comparison Change
Baseline (T0) Endline (T1) Baseline (C0) Endline (C1) Inter. Comp. diff-in-diff
Programme 570.22 460.09 491.4 339.35 -109.00 -154.10 61.5(71.8)
Burkina Faso 466.9 193.95 396.51 164.98 -271.87 -231.24 -47.6(58.9)
Ethiopia 317.29 447.12 302.15 365.34 83.92 48.24 66.3(100.8)
Kenya 412.79 291.8 414.74 264.54 -122.60 -146.85 36(54)
Mali 1539.05 1239.34 1127.57 639.64 -214.82 -462.77 330.6(384.5)
Niger 393.36 323.32 429.52 335.43 -54.24 -89.40 41.4(143.3)
Table 6.5. 2: Change in average size of land under cultivation, by programme and country
Intervention Comparison % Change
Baseline (T0) Endline (T1) Baseline (C0) Endline (C1) Inter. Comp.
Programme 3.06 2.33 2.79 2.04 24% 27%
Burkina Faso 3.33 3.05 3.12 2.58 8% 17%
Ethiopia 0.57 0.74 0.53 0.62 -30% -17%
Kenya 1.33 0.87 1.33 0.91 35% 32%
Mali 8.58 4.12 7.47 3.44 52% 54%
Niger 3.75 3.74 3.58 3.6 0% -1%
The box plots shown in Figure 6.5.10 suggest that while about half of the farmers did not realise any
change in crop production, some farmers recorded an increase in the value of crop output of more than
48
US$ 500 and more of such farmers were in the intervention sites than in the comparison sites. The
boxplots show that variations exist across the countries. Malian farmers recorded the highest gross cash
values, three times over the programme average, as they had larger crop land size on average compared
to the other countries (Table 6.5.2). Their gross cash values decreased at the end too, although much
less comparatively in the intervention sites.
Figure 6.5. 10:Comparison of change in gross cash value of crops produced between intervention and
comparison HHs, by programme and country
We also looked at net value of crops produced to assess whether farmers were able to cover variable
costs of production. Table 6.5.3 shows baseline and endline net cash value of crops produced across the
five countries, as well as at programme level. The results indicate that there was a general decline in the
net value of crops produced in the intervention and comparison sites in all the countries, with endline
values remaining positive for all the countries except for Burkina Faso in both the intervention and
comparison households and in Mali in the intervention households. The decline in net cash value is,
however, not surprising given the decline in the gross value of crops produced observed earlier. The
decline was greater in the comparison than in the intervention sites in Niger and Kenya, resulting in a
positive albeit insignificant difference-in-difference net cash value of US$ 72 and US$25, respectively.
Table 6.5. 3: Change in average net cash value of crops produced (US$), by programme and country
Programme 464.64 21.84 405.64 49.22 -353.35 -347.35 -67.4(95.6)
Burkina Faso 406.41 -19.95 338.57 -24.66 -426.11 -362.33 -80(49.5)
Ethiopia 245.89 32.97 242.47 30.34 -213.75 -211.00 -1.6(39.8)
Kenya 323.83 36.9 341.64 33.89 -288.69 -304.60 24.8(59.3)
Mali 1227.8 -56.76 900.84 157.73 -739.74 -723.13 -419.2(467.6)
Niger 341.66 107.21 376.1 84.74 -207.23 -257.91 71.6(102.2)
49
6.5.3.2 Impact on market integration
Market integration is often considered as a key ingredient of a successful intensification strategy. Farmer
linkage to output market was facilitated for selected crops in the DryDev countries (see Table 4.2.1).
Interventions to improve smallholder farmer’s integration into output markets centred on enhancing the
producers’ capacity to use innovative marketing approaches to increase market share for their products
sustainably, thereby improving returns on investment. Producers were exposed to different strategies
some of which aimed at differentiating their products and segmenting the market, with a view to
increasing the market share of the commodities. Such strategies included sorting and grading, packaging,
and processing. Other strategies focused on consolidating the producers’ market share such as
amalgamation of producer groups into umbrella organisations or cooperatives. Yet other strategies aimed
at improving the producers’ bargaining power such as establishment of multistakeholder forums to foster
dialogue between value chain actors, as well as reducing transaction costs e.g. contracting, collective
selling of outputs and buying of inputs, bulking/aggregation of products and direct linking of producers
with buyers such as cooperatives, commodity exchange markets, traders and processers.
As a proxy of whether the DryDev interventions were able to promote market integration and improve
access to output markets, we compare the differences in gross and net cash value of crops sold. The
results for intervention and comparison households, summarised in Table 6.5.4 depict a consistent picture
across the countries and treatment groups. While we document that the value of crops sold declined for
all the countries and treatment groups, the magnitude varies across the board, with Niger and Ethiopia
experiencing the smallest decline in both the groups. The largest, albeit insignificant, difference-in-
difference in gross value of crops sold is observed in Mali where the decline in the value of crops sold was
greater in the comparison than in the intervention households. The boxplots shown in Figure 6.5.11
suggest that despite most of the farmers reporting no change in gross value of crops sold, a few recorded
positive change of more than US$ 500 particularly in Mali. The impact of the decline in production is
reinforced by Figure 6.5.12 where the proportion of farmers selling crop produce declined across the
countries and treatment groups, except in Niger where there was marginal increase in both the groups.
Table 6.5. 4: Change in average gross cash value o crops sold (US$), by programme and country
Programme 226.61 135.63 179.77 83.11 -77.76 -94.17 17.2(26.7)
Burkina Faso 80.83 22.72 65.33 18.39 -57.748 -46.77 -17.2(19.6)
Ethiopia 66.36 38.44 88.63 47.21 -27.71 -39.79 13.7(11.4)
Kenya 250.78 69.22 250.3 75 -183.256 -172.36 0.39(25.4)
Mali 786.88 579.24 436.48 211.09 -108.442 -203.55 91.8(107.1)
Niger 106.61 93.92 129.62 104.37 -15.306 -29.44 26.8(37.6)
50
Figure 6.5. 11: Comparison of change in gross cash value of crops sold between intervention and comparison HHs, by programme
and country
100%
90%
80%
% of households selling
70%
60%
50%
40%
30%
20%
10%
0%
Figure 6.5. 12: Comparison of proportion of farmers selling crop produce at baseline and endline, by programme
and country
6.5.3.3. Impact on crop profits
With enhanced market access, farmers are expected to switch from the production of less market
oriented to more market-oriented commodities, increase investment in such commodities, thereby
producing more marketable surplus and adding value to the commodities to diversify their marketing
channels and increase their profit margins. The results, however, show that the pathway described did
not materialise in the programme sites. Instead, there was a decline in profit margins, which is consistent
with the decline observed in gross value of crops produced and sold.
51
Table 6.5. 5: Change in net crop profit (gross margins), by programme and country
Programme 185.4 116.19 151.77 71.41 -69.196 -80.846 19.4(71.8)
Burkina Faso 69.69 20.17 55.27 16.68 -49.17 -38.432 -15.8(15.8)
Ethiopia 56.53 33.87 78.76 41.5 -22.615 -35.843 13.9(10.5)
Kenya 235.05 58.79 237.41 62.82 -177.911 -171.853 4.58(22)
Mali 589.89 495.6 318.91 182.22 -82.694 -135.929 97.7(105.8)
Niger 89.34 79.5 110.67 88.7 -12.429 -25.874 23.7(33.2)
While further interrogation of the mechanisms behind changes in production is required to unravel the
underlying reasons for the consistent decline in production across the countries, the positive change in
gross value of crops produced observed in Ethiopia and Kenya suggests that there are likely to be
productivity gains from natural resource management interventions, although more time may be required
for the impacts to manifest. Factors of production, particularly land, are faced with the problem of asset
fixity and seasonality of agricultural production. Therefore, the magnitude of the impact and duration it
takes to manifest depends on the rate at which households can switch factors of production from one
enterprise to another. NRM investments entail sunk costs, which reduce responsiveness to incentives
(Barrett, 2002). Studies have shown that farmers are slower to undertake profitable investments or divest
from low return strategies (Chavas, 1994;Davis, 1996).
52
7. Conclusions, recommendations and lessons learnt
Overall, impact assessment results are mixed and can be interpreted in different ways. Our main result is
that DryDev has “impact” in the sense that it is associated with improved household food security (dietary
diversity), inclusion and women empowerment and household resilience. However, these positive results
do not materialize across all countries and sub-populations. Specifically, there is some evidence that
DryDev improved local governance and institutional functioning across the programme; led to diversified
diets among women in Kenya; increased inclusion and empowerment of women in Kenya and Ethiopia;
and modestly enhanced resilience among households in Ethiopia. In fact these need to improve to drive
the rest of the changes on and off farm.
Our analysis has not been able to identify the mechanisms linking interventions to impact beyond short-
term outcomes —adoption of soil and water management options and improvement in local governance
and institutional functioning. When comparing intervention and comparison groups, the results are often
inconclusive on production and market outcomes, which were hypothesised as the pathways through
which impacts would manifest. The results point to the need to probe further into the data for possible
mechanisms other than the pathways proposed in the theory of change.
To maximise gains from NRM interventions, particularly in severely degraded landscapes such as the
DryDev sites, interventions must be integrated and adopted at scale. It is apparent from the results that
successful integration of smallholder farmers into markets depends largely on the extent to which farmers
can take advantage of the promoted climate smart production practices by learning about them, adapting
them and finally deciding to shift the available productive resources into the more lucrative value chains
to generate surplus for the market. Where productivity gains are not commensurate with the cost of
production, creating market linkages may not provide the necessary impetus for market integration
among smallholder farmers whose main objective is to produce for sustenance. Therefore, inducing
massive improvement in production beyond what the farmers require for consumption is a prerequisite
for future projects aiming to commercialise smallholder agriculture especially in the degraded landscapes
of the Sahel and the Horn of Africa. Only then can market linkages and capacity building on innovative
marketing approaches provide the much-needed incentive for farmers to integrate into agricultural
markets.
DryDev interventions were mainly NRM related and hence characterised by lagged impacts — thus could
not have had much impact in a span of less than three years since the first two year were dedicated to
programme design. Whereas implementation began in late 2015 after baseline data collection in some
countries, others, particularly those in the Sahel, did not begin in earnest until late 2016. Instead these
countries spent considerable amount of time trying to work out the right mix of partnership that could
deliver the impacts. Therefore, to expect strong impacts after only 2 years of effective implementation in
such countries where watershed approach was fairly a new concept would be overly optimistic. For
instance, while institutional strengthening was expected to expedite technology uptake among the
farmers, even receptive farmers require time to learn, internalise and embrace certain complex aspects
of NRM related interventions (Barrett et al. 2002). To the extent that integrated soil and water
management was a new concept to majority of the farmers, changes in productivity would be contingent
on the ability of the farmers to synthesise the concept, adapt it to the environment and then implement
53
it. Similarly, some of the production factors are fixed in the short term, and their use may be
predetermined. Economic incentives inform of tangible benefits emanating from the NRM interventions
that are necessary to induce enterprise switching or reallocation of resources to the supported value
chains may not materialise within a span of two years.
Studies have shown that where nutrition education is low or absent, as was the case in the DryDev sites,
increased productivity is necessary but not sufficient to improve food and nutrition security among
smallholder farmers(Murendo et al. 2018). Therefore, future programmes that aim at improving food and
nutrition security among smallholder farmers residing in marginal or degraded landscapes through natural
resource management interventions need to consider going beyond improving production and market
outcomes to incorporate nutrition education into such programmes.
54
References
Barrett C.B., Place F. and Brown D.R. 2002. The challenge of stimulating adoption of improved natural
resource management practices in African agriculture. In: Barrett C.B., Place F. and Aboud A.A.(Eds.),
Natural Resource Management in African Agriculture: Understanding and Improving Current Practices
Chavas JP. 1994. Production and Investment Decisions Under Sunk Cost and Temporal Uncertainty.
American Journal of Agricultural Economics 76:114-127
Davis, Graham. 1996. “Investment under Uncertainty: Avinash K Dixit and Robert S Pindyck Princeton
University Press, Princeton, NJ, 1994, Xiv + 468 Pp (Hardcover), ISBN 0-691-03410-9.” Resources
Policy 22 (3): 218–218.
Jerneck, A. and Olsson, L., 2014. Food first! Theorising assets and actors in agroforestry: risk evaders,
opportunity seekers and ‘the food imperative’in sub-Saharan Africa. International journal of agricultural
sustainability, 12 (1), 1– 22
Koppmair, Stefan, Menale Kassie, and Matin Qaim. 2017. “Farm Production, Market Access and Dietary
Diversity in Malawi.” Public Health Nutrition 20 (2): 325–35.
https://doi.org/10.1017/S1368980016002135.
Murendo, Conrad, Brighton Nhau, Kizito Mazvimavi, Thamsanqa Khanye, and Simon Gwara. 2018.
“Nutrition Education, Farm Production Diversity, and Commercialization on Household and
Individual Dietary Diversity in Zimbabwe.” Food & Nutrition Research 62 (May).
https://doi.org/10.29219/fnr.v62.1276.
Reij, C. 2005. Investing in Africa’s drylands: Impacts on agriculture, environment and poverty reduction.
In: Ruben, R., and Bart de Steenhuijsen Piters (Eds. ), Rural Development in Sub-Saharan Africa:
Policy Perspectives for Agriculture, Sustainable Resource Management and Poverty Reduction.
Bulletin / Royal Tropical Institute 370. Amsterdam: KIT Publ.
Shiferaw, Bekele A., Julius Okello, and Ratna V. Reddy. 2009. “Adoption and Adaptation of Natural
Resource Management Innovations in Smallholder Agriculture: Reflections on Key Lessons and
Best Practices.” Environment, Development and Sustainability 11 (3): 601–19.
https://doi.org/10.1007/s10668-007-9132-1.
Shiferaw, Bekele, Jon Hellin, and Geoffrey Muricho. 2011. “Improving Market Access and Agricultural
Productivity Growth in Africa: What Role for Producer Organizations and Collective Action
Institutions?” Food Security 3 (4): 475–89. https://doi.org/10.1007/s12571-011-0153-0.
55
Annexes
Annex 1: Country Maps of Impact study sites
Annex 1.1 Surveyed Households in Kenya at baseline and endline, by County and sub-watershed
Surveyed households in Kenya
Kitui County
Machakos County
Makueni County
56
Annex 1.2 Surveyed Households in Kenya at baseline and endline, by woreda and sub-watershed
Surveyed households in Ethiopia Samre Woreda
Jarso Woreda
57
Annex 1.3 Surveyed Households in Burkina at baseline and endline, by Province
58
Annex 1.4 Surveyed Households in Mali at baseline and endline, by sub-watershed
59
Annex 1.5 Surveyed Households in Niger at baseline and endline, by Region
60
Supported by:
Led by
Implementation partners:
in ETHIOPIA
Country Lead,
Implementer
Tigray region EOC/DICAC
Implementer Tigray region Implementer Rift Valley
in KENYA
Country Lead Value chain lead & Lead NRM in Lead NRM in
Implementer in all 3 counties Kitui county Makueni county
BURKINA FASO
Technical Lead Technical Lead

Agri value chains non-timber forest
Country Lead
products value chains
MALI
Country Lead
Implementer in Mopti Lead Value chains & Implementer AMEPPE
region Saving 4 Change Sikasso region Implementer
Segou region
NIGER
Country Lead Networking, Capacity Implementer in Toridi Implementer

Building, Value chains in Douroun/Zinder
Policy influencing
Implementer in AREN CRESA

Aguie/Maradi Implementer in Aguie Lead Innovation
Platform
61
62

Impact Assessment Report DryDev Final

Uploaded by

Copyright:

Available Formats

Impact Assessment Report DryDev Final

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Impact Assessment Report DryDev Final

Uploaded by

Copyright:

Available Formats

The Drylands Development Programme

A Farmer-led Programme to Enhance Water Management, Food

Impact Assessment Report

Prepared for: The Ministry of Foreign Affairs, the Netherlands

Table 1. 1: DryDev programme's logical framework and work packages .................................................... 6

Table 6.2. 1: Difference-in-differences results for three measures of poverty .......................................... 36

Table 1. 1: DryDev programme's logical framework and work packages

3. Impact and Outcome Indicators

4.1. Data collection and cleaning

Intervention Households Comparison Households

Figure 4.1. 1: Number of farmers interviewed during baseline by country

Intervention households Comparison households

Figure 4.1. 2:Number of farmers interviewed during endline survey by country

• Food consumption errors. As part of the household food consumption module,

• Likelihood of spill over effects, particularly in Kenya. As explained in Section 2, the

4.2. Data processing and construction of indicators

4.2.2. Dietary diversity index

4.2.3. Women’s Empowerment in Agriculture Index (WEAI)

4.2.4. Household Consumption Expenditure

An alternative way of measuring a household’s wealth status in low-and middle-income countries is by

4.2.6. Multidimensional Poverty Index

Figure 4.2. 3: Dimensions and Indicators of Resilience Index

5. Comparison of covariates and key outcome

5.1. Respondent and general household characteristics

Dif. Dif. Dif. Dif. Dif. Dif.

Table 5.2. 2:Comparison of farm characteristics at baseline, by programme and country

Local governance and institutional functioning

Farmers participating in farmer

Food security (Minimum dietary diversity-MDD)

HHs consuming 5 or more food

6.1. Impact on dietary diversity

Impact on food security: Indicator Initial Impact DryDev Impact (DiD)

the Minimum Dietary Diversity

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1 Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Impact on poverty Indicators Initial Impact DryDev impact

Intervention-T0 Comparison-T1 Intervention-C0 Comparison-C1

Figure 6.2. 1: Proportion of multidimensionally poor households at baseline and endline

Impact on women Indicator Initial Impact DryDev impact (DiD)

Table 6.3. 1: DiD results for women empowerment in agriculture index

Production Resource Income Leadership Leisure&low stress

Impact on household Indicator Initial Impact target DryDev Impact (DiD)

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

6.5. Probing the mechanisms

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

6.5.2 Impact on adoption of selected soil and water management options

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1 Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

% of main farm irrigated

Intervention-T0 Intervention-T1 Intervention-T0 Intervention-T1

6.5.3. Impact on production, market integration and profitability

6.5.3.1. Impact on production

Intervention-T0 Intervention-T1 Comparison-C0 Comparison-C1

6.5.3.3. Impact on crop profits

Surveyed households in Kenya

Surveyed households in Ethiopia Samre Woreda

Technical Lead Technical Lead