Mejoramiento Agrícola Sostenible 
MAS+
Impact Evaluation Baseline Report
November 2018
b
Prepared for TechnoServe by the International Food Policy Research Institute
Last revision: January 10, 2019
Contacts: 
Alan de Brauw
Markets, Trade and Institutions Division
International Food Policy Research Institute
1201 Eye Street NW, Washington, DC
Miguel Almanzar
Markets, Trade and Institutions Division
International Food Policy Research Institute
1201 Eye Street NW, Washington, DC
Impact Evaluation Baseline Report for FY 2017 Food 
for Progress Mejoramiento Agrícola Sostenible (MAS+) 
Project in Honduras
Program: Food for Progress
Agreement Number: FCC-522-2017/023-00
Funding Year: Fiscal Year 2017
Project Duration: 2017-2022
Implemented by: TechnoServe
i
Contents
List of Tables ...................................................................................................................................iii
List of Figures..................................................................................................................................vi
List of Acronyms.............................................................................................................................vii
Executive Summary...................................................................................................................... viii
Data Sources...............................................................................................................................ix
Coffee Farmers ...........................................................................................................................ix
Beans Farmers.............................................................................................................................x
Summary of Key Indicators..........................................................................................................x
1 Introduction.............................................................................................................................. 1
2 Background.............................................................................................................................. 1
Primary Research Questions .................................................................................................. 3
Theory of Change.................................................................................................................... 4
3 Impact Evaluation Methodology .............................................................................................. 5
Randomized Control Trial Methodology: Coffee Producers Impact Evaluation ..................... 6
Propensity Score Methodology: Beans Producers Impact Evaluation ................................... 7
4 Data Sources ........................................................................................................................... 8
4.1 Survey Design .................................................................................................................. 8
4.2 Data Collection: Training, Pilot and Fieldwork............................................................... 13
4.3 Statistical Power and Sample Selection ........................................................................ 14
Survey Response Rates........................................................................................................ 16
Effective Minimum Detectable Effects................................................................................... 17
5 General Characteristics of the Intervention Area.................................................................. 18
6 Analysis of Coffee Producers in MAS+ Evaluation ............................................................... 20
6.1 Demographic Characteristics......................................................................................... 20
6.2 Agricultural Outcomes.................................................................................................... 22
Size of Cultivated Land ......................................................................................................... 22
Coffee Production.................................................................................................................. 23
Coffee Yields.......................................................................................................................... 25
Coffee and Markets: Buyers, Sales and Prices .................................................................... 29
Evaluation Validity Tests at Baseline: Agricultural Productivity for Coffee............................ 32
ii
Inputs, Technologies and Management Practices ................................................................ 39
Coffee Extension ................................................................................................................... 44
Coffee Pests .......................................................................................................................... 45
Investments and Equipment.................................................................................................. 46
Group Membership and Participation ................................................................................... 47
Evaluation Validity Tests at Baseline: Inputs, Extension and Pests for Coffee .................... 48
7 Analysis of Beans Producers in MAS+ Evaluation ............................................................... 52
7.1 Demographic Characteristics......................................................................................... 52
7.2 Agricultural Outcomes.................................................................................................... 54
Size of Cultivated Land ......................................................................................................... 54
Beans Production .................................................................................................................. 55
Beans Food Security ............................................................................................................. 56
Beans Yields .......................................................................................................................... 56
Beans Sales........................................................................................................................... 60
Evaluation Validity Tests at Baseline: Agricultural Productivity - Beans............................... 61
Inputs ..................................................................................................................................... 67
Practices ................................................................................................................................ 68
Beans Extension.................................................................................................................... 70
Beans Pests........................................................................................................................... 71
Investments and Assets of Beans Farmers .......................................................................... 71
Group Membership and Participation ................................................................................... 73
Evaluation Validity Tests at Baseline: Inputs, Extension and Pests for Beans..................... 74
8 Discussion and Conclusions ................................................................................................. 79
9 Recommendations for Monitoring ......................................................................................... 81
References.................................................................................................................................... 82
Appendix 1: Collective Action, Transfers, Credit, Shocks, Aspirations and ICTs......................... 83
Coffee Sample........................................................................................................................... 83
Beans Sample ........................................................................................................................... 90
iii
List of Tables
Table 1 Projected Baseline Sample Characteristics and Minimum Detectable Effect................. 15
Table 2 Coffee Sample Design ..................................................................................................... 16
Table 3 Beans Sample Design ..................................................................................................... 16
Table 4 Completed sample and Response Rates........................................................................ 17
Table 5 Minimum Detectable effects of Sample ........................................................................... 17
Table 6: Demographic Characteristics for Coffee Sample ........................................................... 20
Table 7: Demographic Characteristics.......................................................................................... 21
Table 8: Human Capital Characteristics, by Gender.................................................................... 21
Table 9: Agricultural and Non-Agricultural Income ....................................................................... 21
Table 10: Size of planted and cultivated area for Coffee Sample................................................ 23
Table 11: Number of total plots and owned plots for coffee sample ............................................ 23
Table 12 Units and Processing Levels selected by respondents................................................. 24
Table 13: Coffee production.......................................................................................................... 25
Table 14: Coffee Yields ................................................................................................................. 26
Table 15 Coffee Buyers and Quantity Sold .................................................................................. 29
Table 16 Coffee Buyers and Sale Prices...................................................................................... 30
Table 17: Coffee Sales: Value....................................................................................................... 31
Table 18 Coffee Sales: Quantity by Processing Level ................................................................. 32
Table 19: Balance Test for Coffee Area Size................................................................................ 33
Table 20: Balance Test for Coffee Production .............................................................................. 34
Table 21: Balance Test for Coffee Yields...................................................................................... 35
Table 22: Balance Test for Coffee Sold and Value of Sales......................................................... 37
Table 23: Balance Test for Quantity Sold by Buyers of Coffee .................................................... 38
Table 24: Share of farmers that applied coffee inputs.................................................................. 39
Table 25: Cost of inputs and labor costs ...................................................................................... 40
iv
Table 26: Coffee Practices............................................................................................................ 40
Table 27: Coffee Practices: Plant and tissue management ......................................................... 41
Table 28: Coffee Practices: Quality .............................................................................................. 41
Table 29: Coffee Practices: Soil Care........................................................................................... 42
Table 30: Coffee Practices: Post-Harvest..................................................................................... 43
Table 31: Coffee Practices: More on Quality................................................................................ 44
Table 32: Coffee Extension........................................................................................................... 45
Table 33: Pests in Coffee.............................................................................................................. 46
Table 34: Investments in the agricultural plots ............................................................................ 46
Table 35: Investment in Equipment .............................................................................................. 47
Table 36: Groups present in the community................................................................................. 48
Table 37: Balance Test for Share of Farmers that applied inputs ................................................ 49
Table 38: Balance Test for Cost of Inputs and Labor Costs......................................................... 49
Table 39: Balance Test for Coffee Extension................................................................................ 50
Table 40: Balance Test for Pests in Coffee................................................................................... 51
Table 41: Demographic Characteristics, Beans Sample.............................................................. 52
Table 42: Demographic Characteristics........................................................................................ 53
Table 43: Demographic Characteristics by Gender ..................................................................... 53
Table 44: Agricultural and Non-Agricultural Income..................................................................... 53
Table 45: Area Size of planted and harvested area with beans................................................... 54
Table 46: Beans Production.......................................................................................................... 55
Table 47: Beans Production Final use, Primera........................................................................... 55
Table 48: Beans Production Final Use, Postrera ......................................................................... 56
Table 49: Beans Food Security..................................................................................................... 56
Table 50: Beans Yields ................................................................................................................. 57
Table 51: Beans sales, quantity and value of sales ..................................................................... 60
v
Table 52: Balance Test for Beans Area Size Planted and Harvested .......................................... 62
Table 53: Balance Test for Beans Production .............................................................................. 63
Table 54: Balance Test for Beans Yields ...................................................................................... 64
Table 55: Balance Test for Beans Sales, quantity and value of sales ......................................... 66
Table 56: Share of farmers that used beans inputs ..................................................................... 67
Table 57: Total costs of Inputs and Labor Costs .......................................................................... 67
Table 58: Practices Beans: Access to Markets, Varieties and Selling Points .............................. 68
Table 59 Sales of Bean Production by buyers ............................................................................. 68
Table 60: Practices in Bean parcels ............................................................................................. 69
Table 61: Beans Farming Practices.............................................................................................. 70
Table 62: Extension received for beans sample........................................................................... 70
Table 63: Pests that affected Beans crops................................................................................... 71
Table 64: Investments and Asset Counts ..................................................................................... 72
Table 65: Eigenvalues for first component of PCA for Investments............................................. 72
Table 66: Eigenvalues for first component of PCA for Equipment............................................... 73
Table 67: Groups in the Community, beans sample .................................................................... 73
Table 68: Balance Test for Shared of Beans Farmers that applied inputs................................... 75
Table 69: Balance Test for Input Costs and Labor Costs ............................................................. 75
Table 70: Balance Test for Beans Extension................................................................................ 77
Table 71: Balance Test for Pests in Beans................................................................................... 78
Table 72 Summary Baseline Measures: Coffee Key Variables.................................................... 80
Table 73 Summary Baseline Measures: Beans Key Variables .................................................... 81
Table 74: Coffee Practices: Access to Market and Expectations................................................. 83
Table 75......................................................................................................................................... 84
Table 76: Remittances and transfers received by coffee sample ................................................ 86
Table 77: Credit for coffee sample................................................................................................ 87
vi
Table 78: Shocks experienced in the last 12 months................................................................... 88
Table 79: Sources of received agricultural information ................................................................ 88
Table 80: Internet and Cellular Use .............................................................................................. 89
Table 81: Collective Action............................................................................................................ 90
Table 82: Remittances and Transfers received by beans sample ............................................... 92
Table 83: Credit received by beans sample ................................................................................. 93
Table 84: Shocks experienced during the last 12 months, beans sample................................... 94
Table 85: Sources of agricultural information ............................................................................... 94
Table 86: Internet and cellphone use ........................................................................................... 95
List of Figures
Figure 1: Coffee and Beans Sample Geographic Distribution ..................................................... 18
Figure 2 Distribution of coffee productivity ................................................................................... 27
Figure 3: Coffee area and coffee production, total and for commercial production .................... 28
Figure 4 Distribution of Prices of Coffee: Calculated by Processing Level ................................. 31
Figure 5 Distribution of coffee sales ............................................................................................. 32
Figure 6 Area Planted Distribution across Treatment Assignment .............................................. 34
Figure 7 Yield Distribution across Treatment Assignment ........................................................... 36
Figure 8 Quantity sold across treatment assignment................................................................... 38
Figure 9 Distribution of Beans productivity by season................................................................. 57
Figure 10: Beans Area and Beans Production in Primera and Postrera ..................................... 59
Figure 11 Distribution of Beans Sales by season......................................................................... 61
Figure 12 Distribution of Area Planted by Treatment Assignment ............................................... 63
Figure 13 Distribution of Yields by Treatment Assignment .......................................................... 65
Figure 14 Distribution of Beans Quantity Sold by Treatment Assignment................................... 66
Figure 15: Aspirations, locus of control and self-esteem perceptions ......................................... 85
Figure 16: Aspirations, locus of control and self-esteem perceptions ......................................... 91
vii
List of Acronyms
ANCOVA Analysis of Covariance
CAPI Computer-Assisted Personal Interviews
COPECO Permanent Emergency Contingency Committee
CRT Cluster Randomized Controlled Trial
CSA Canadian Standard Association
CSB Community Seed Bank
ES Executive Summary
EV Eigen-Values
ESA ESA Consultores. Economics, Society, Environment and Engineering
GPS Geographic Positioning System
ICT Information and Communication Technologies
IHCAFE Honduran Institute for Coffee
IFPRI International Food Policy Research Institute
MAS Mejoramiento Agrícola Sostenible - Sustainable Agricultural Improvement
MDE Minimum Detectable Effects
MDG Millennium Development Goals
MT Megagram
MT Metric Ton
MSU Michigan State University
PCA Principal Component Analysis
POs Producer Organizations
PPP Purchasing Power Parity
PSM Propensity Score Matching
RCT Randomized Controlled Trial
SAG Honduran Ministry of Agriculture
SD Standard Deviation
SE Standard Error
USAID United States Agency for International Development
USD United States Dollar
USDA United States Department of Agriculture
USG United States Government
WP Wet parchment
viii
Executive Summary
The project Mejoramiento Agrícola Sostenible (henceforth referred to as ‘MAS+’) is a five-year 
project funded by the United States Department of Agriculture (USDA) and implemented by 
TechnoServe in Honduras. The MAS+ project is a continuation and expansion of proven 
interventions first implemented in the FY2012 MAS project1
 . It seeks to work with 32,000 
smallholder coffee and bean producers across nine departments in Honduras. The MAS+ 
project’s integrated approach is based on seven discrete but interrelated activities that aim to 
increase coffee and beans producers’ incomes through: training to improved crop productivity and 
quality, expansion of financial services and inputs, capacity building for producer organizations, 
and marketing initiatives aimed at expanding the coffee and dry-beans value chains.
This report describes the impact evaluation approach being applied to the MAS+ project, as well 
as the results of the impact evaluation baseline survey conducted in 2018. The impact evaluation 
described in this report is an important component of the overall monitoring and evaluation plan 
for MAS+. The goal of the impact evaluation is to provide rigorous evidence about how specific 
activities lead to improved agricultural productivity related to coffee and beans production. The 
impact evaluation will measure intermediate outcomes along the hypothesized causal chain to 
understand how impacts may have come about; furthermore, by measuring against a control 
group, it will help us understand what gains in production and/or productivity can be attributed to
the MAS+ project.
The impact evaluation is focusing on three MAS+ activities: training, capacity building, and 
financial services (for the coffee value chain). It seeks to answer three primary research questions: 
1) What is the effect of training on agricultural productivity? 2) What is the additional effect of 
finance on agricultural productivity? and 3) What practices are more likely to be adopted by 
smallholder farmers, and why? 
The impact evaluation takes slightly different approaches to answering questions 1 and 3 for both 
value chains and all three questions for the coffee value chain. For the coffee value chain, the 
impact evaluation methodology will be a cluster randomized controlled trial. After a sample of 
potential participant villages was established, half were randomly chosen to be in the treatment 
group and the other half in the control group. The impact evaluation in the beans sample is being 
conducted with non-experimental methods. To further improve the design, beneficiary groups 
were initially chosen, and then non-participants were selected to be similar to the participants at 
the start of the project by initially pairing the prospective groups with similar ones. 
 
1 This includes leveraging the private and public sector partnerships developed during the MAS Program and expanding 
on that project’s technical assistance model, focus on higher quality markets and use of proper inputs, access to finance 
model, and efforts to strengthen producer organizations and incorporate the participation of youth in the program’s 
activities.
ix
Data Sources
The baseline survey was jointly designed by IFPRI, TechnoServe, and ESA Consultores, the local 
firm that conducted the survey. It was conducted using computer assisted personal interviews 
software called Survey Solutions. The content of the survey focused on three themes: agricultural 
production and commercialization, extension and technology practices, and producer association 
and credit access, which mirrors the research questions noted above. It was conducted between 
July and August of 2018. IFPRI researchers, jointly with ESA Consultores, conducted a 14-day 
training between June 4th to June 17th: 7 days with field managers and groups’ supervisors and 
the remaining 7 days with the whole team of enumerators to review agricultural concepts 
facilitated by TechnoServe. 
We designed the sample to target an increase in coffee yields of 24% (40% among participants, 
at a 60% participation rate) and bean yields of 21% at the 5 percent significance level and 80 
percent power, through the baseline and endline surveys; the goal was to survey 1000 households 
for each arm. We ended up with a sample of 961 coffee producers and 968 beans producers. 
With the realized results, the minimum detectable effects are estimated as 25% of the mean for 
the beans yields and 29% of the mean for the coffee yields, at the same significance and power 
levels. If covariates can help explain yields at endline, these minimum detectable effects sizes 
will decrease somewhat.
Coffee Farmers
Among coffee producers, we find that coffee income represents about 69 percent of total income 
on average. Coffee farmers produce for commercial purposes and usually sell in wet parchment 
presentation. Coffee farmers mainly sell to an intermediary and very few to exporters. Yields are 
around 70 quintales per hectare in wet parchment on average and 34 quintales per hectare in wet 
parchment at the median, highlighting the wide range in coffee farmer productivity found in the 
sample. Not all coffee farmers use inputs, the most common inputs used are chemical fertilizers
(63 percent) and herbicide (36 percent). Few coffee farmers, 27 percent, use foliar fertilizer. 
Further, less than 10 percent have received extension in the past 12 months. Forty percent of 
coffee farmers report having problem with pests, particularly coffee rust and broca. 
Given that the intervention will be cluster randomized, not surprisingly very few differences were 
found between the treatment and control groups at conventional significance levels (specifically, 
due to farmers’ coffee being affected by the minador pest). Overall, the analysis shows that the 
randomization was properly implemented and that the treatment and control groups are 
comparable at baseline across all the key outcomes of the evaluation and project.
The distribution of prices across different coffee processing levels had a bi-modal shape after 
standardizing the quantities and processing levels. This indicates that there are two different 
pricing regimes for coffee; which price a farmer obtains depends on other attributes of their 
production. In the endline, we will characterize these attributes, to help explain differences in 
prices within each processing level. By helping to explain these differences, the power to detect 
differences will increase.
x
Beans Farmers
Among beans producers, we find that some farmers primarily plant in the primera season, while 
others in the postrera. This tendency is mainly due to inherent characteristics of the bean crop 
cycle and the primera season’s more favorable weather conditions, including temperature, rainfall,
and soil drainage capabilities, among others. For this reason, primera season generally yields 
higher volumes than postrera. Thus, households producing in the primera tend to sell more of 
their harvest than those in the postrera, and many produce for home consumption. Households
that sell beans typically do so to an intermediary. Yields are around 888 kilograms per hectare in 
the primera and 670 in the postrera. While not all households use inputs, the most common inputs 
used are herbicides and fertilizer. Few households use any inoculant when planting, which boosts 
yields; very few farmers (17 percent of the sample) have received any extension in the past 12 
months either. Some farmers report having problems with pests; therefore it will be important to 
monitor whether pest exposure differs between treatment and control areas. 
Despite the fact that the beans intervention is not randomized, we found very few statistically 
significant differences between the treatment and control groups. The main difference we found 
is that the beans treatment group has received more extension in the past twelve months than 
the control group. We hypothesize that some farmers may have received extension with similar 
themes as MAS+ in the past 12 months; however, it is a relatively small number of households. 
We will control for this exposure in the endline, and plan to tailor the survey form to ensure that 
we can directly capture the types of extension that are provided by MAS+ partners. We also find 
less exposure to pests, specifically losses, among the treatment group relative to the control 
group. However, the magnitude is small. For this second difference, we will keep in mind at endline 
to learn whether pests are more endemic to control areas, so that we can control for this difference 
if necessary as well.
Summary of Key Indicators 
We summarize the baseline findings for important agricultural productivity outcomes—the 
production, yields, volume sold, and the total reported value of sales- in Table ES-1 for coffee and 
Table ES-2 for beans. Farmers in the sample have a wide spread in their agricultural productivity, 
which creates an opportunity for MAS+ to provide a diverse set of training that can respond to 
different characteristics of farmers. Namely, for farmers that are below the median of the yield 
distribution, training aimed at increasing productivity can bring about large impact across these 
outcomes. In the case of those farmers that are above the median in the productivity distribution, 
an intervention aimed at increasing their access to better markets (including export), credit, 
certifications, etc. can have a large impact on the total income they derive from agriculture.
xi
TABLE ES-1 SUMMARY BASELINE MEASURES: COFFEE KEY VARIABLES
Mean SD Median Observations
Coffee production in qq WP, total 70 125 34 925
Total yields (qq/ha WP) 30.9 30.1 21.0 925
Quantity of coffee sold in Total MT WP 4.08 12.95 1.41 898
Total Sale value (USD) 3,220 6,887 1,500 873
Notes: Sample for beans consists of 961 farmers. Difference in sample sizes is due to farmers that do not have positive quantities for 
the indicators.
TABLE ES-2 SUMMARY BASELINE MEASURES: BEANS KEY VARIABLES
Mean SD Median Observations
Production, Primera (kg) 862 943 544 657
Production, Postrera (kg) 645 761 454 738
Yields, Primera (kg/ha) 888 596 782 657
Yields, Postrera (kg(ha) 670 440 586 738
Volume Sold, Primera (MT) 0.81 0.92 0.54 535
Value of Sales (USD), Primera 631 724 369 487
Volume Sold, Postrera (MT) 0.61 0.76 0.36 541
Value of Sales, Postrera (USD) 1311 6637 277 491
Beans Production Home Consumption kg in Primera 140 192 91 633
Beans Production Home Consumption kg in Postrera 144 219 91 708
Notes: Sample for beans consists of 968 farmers. Difference in sample sizes is due to farmers that do not have positive quantities for the 
indicators.
1
1 Introduction
The project Mejoramiento Agrícola Sostenible (henceforth referred to as ‘MAS+’) is a five-year 
project funded by the United States Department of Agriculture (USDA) taking place in Honduras, 
aimed at increasing coffee and beans producers’ incomes through training to improved crop 
productivity and quality, expansion of financial services and inputs, capacity building for producer 
organizations, and marketing initiatives aimed at expanding the coffee and dry-beans value 
chains.
The impact evaluation described in this report is an important component of the overall monitoring 
and evaluation plan for MAS+. The first part will monitor activities throughout the project to provide 
evidence about changes in agricultural productivity and expanded sales of agricultural products 
taking place among beneficiaries. The impact evaluation will specifically attempt to measure the 
increase in productivity among beneficiaries relative to a control group that will not be participating 
in project activities. The goal of the impact evaluation, then, is to provide rigorous evidence about 
how specific activities lead to improved agricultural productivity, related specifically to coffee and 
bean production. The impact evaluation will also measure intermediate outcomes along the 
hypothesized causal chain to understand how impacts may have come about; furthermore, by 
measuring against a control group, it will help us understand what gains in production and/or 
productivity can be attributed to the MAS+ project.
This report presents the results of the baseline survey conducted in July-August 2018 to measure 
the key indicators and outcomes of the project and evaluate the impact of the activities under 
MAS+. The report describes the survey design, baseline estimates for key indicators, and 
validation of the proposed impact evaluation design and the comparability between prospective 
beneficiaries and non-beneficiaries. 
2 Background
The MAS+ project, led by TechnoServe, is a continuation and expansion of proven interventions 
first implemented in the 2012 MAS project to work with 32,000 smallholder coffee and bean 
producers in Honduras. The integrated approach is based on seven discrete but interrelated 
activities. MAS+ will expand its successful comprehensive training curriculum for farmers and 
producer organizations (POs), handing over implementation to private and public-sector
extension workers by Year 4. MAS+ will also provide advisory services to firms supplying 
improved agricultural inputs and climate smart agriculture equipment and services to increase 
access. It will promote innovative financing mechanisms to increase access to finance for farmers 
and POs and develop opportunities to link strengthen these relationships. Finally, it will provide 
capacity building to government institutions to address key issues in the policy environment 
inhibiting trade and productivity in the agricultural sector. When key project initiatives have 
transitioned to market actors in Year 5, project goals include training targeted farmers to adopt a 
minimum of three good agricultural practices that will increase their yields by 30 percent and 
decreasing costs by 10 percent on average.
2
The MAS+ project has seven main activities:
 Activity 1: Training: Facilitate Improved Farmer Productivity. MAS+ will implement a 
suite of on-farm training activities tailored to the specific needs of targeted farmers to 
facilitate improved productivity of coffee and bean parcels. TechnoServe will also support 
targeted producers to produce secondary horticultural crops for additional income.
 Activity 2: Capacity Building: Producer Groups and Cooperatives. MAS+ will build 
the capacity of 435 POs to provide effective marketing, financial, and technical assistance 
to farmers. In addition, we will help POs implement value-added services.
 Activity 3: Inputs: Develop Input, Service, and Equipment Providers. Market-oriented 
inputs and equipment suppliers will understand and supply the input and infrastructure 
needs of farmers and POs. MAS+ will provide business advisory services to existing 
agricultural input, equipment and service providers, with a particular focus on those 
providing CSA products for water-harvesting, irrigation, and soil analysis. TechnoServe
and its partner MSU will continue to scale its community seed bank (CSB) model with 
additional refinements. 
 Activity 4: Capacity Building: Agricultural Extension Agents/Services. To ensure the 
sustainability of MAS+, TechnoServe will help public sector, private sector, and civil society 
organizations develop their own agricultural extension services and take over 
implementation of Activity 1. 
 Activity 5: Financial Services: Facilitate Agricultural and SME Lending. MAS+ will 
strengthen Cajas Rurales, rural savings and loans groups, to facilitate access to finance 
at the PO level. The program will continue to promote innovative agreement mechanisms 
that can successfully offset risk to exporters, such as factoring, input-supplier loans and 
harvest advances. 
 Activity 6: Market Access: Develop Buyer and Seller Relationships. MAS+ will 
promote marketing contracts among farmers and anchor firms. Furthermore, our partners 
will coordinate efforts with IHCOFFEE and the private sector to elevate international 
recognition of Honduran coffee quality. 
 Activity 7: Capacity Building: Government Institutions and Policy Regulatory 
Framework. TechnoServe will provide technical assistance to the Honduran Ministry of 
Agriculture (SAG) to validate bean varieties that can help farmers improve yields. 
TechnoServe will collaborate with the Permanent Emergency Contingency Committee 
(COPECO) and other actors to train targeted local and national government personnel to 
strengthen areas that enable smallholder producers to respond more effectively to climatic 
crises, including drought and rain excess. 
Not all the activities are appropriate for an impact evaluation, as a counterfactual is not available 
to measure the impacts of some of these activities. Specifically, the impact evaluation will focus 
on activities 1, 2, and 5, as training and cooperative capacity are considered as constraining 
productivity in both the coffee and bean sectors. For activity 5, the impact evaluation will 
specifically focus on the coffee value chain, as availability of finance to the smallholder is not 
thought to be a constraining factor in increasing bean yields.
3
Clearly, the other 4 activities may also lead to increased impacts on smallholder agricultural 
productivity. Rather than trying to isolate each activity on its own, we consider the impact 
estimates that will be generated by the impact evaluation as inclusive of those four activities. We 
study these activities through project monitoring and a mid-term process evaluation. 
Primary Research Questions
The impact evaluation then will focus on three primary research questions, all of which relate to 
the USDA Food for Progress Learning Agenda. The key evaluation questions from the 
perspective of the quantitative impact evaluation are:
1. What is the effect of agricultural training on productivity? 
Two activities of MAS+ relate directly to the provision of agricultural extension. Therefore, the first 
primary research question is what the impact of training is on productivity. As trainings will take 
place through groups of farmers, this question relates to priority learning Question 15 in the 
Learning Agenda, “Do cooperatives, associations, federations, or collectives impact producers’ 
abilities to optimize sales to markets at the local, regional, or international level? What services 
provided by cooperatives lead to results? What types of cultural environments are best for these 
different models to thrive?” 
2. What is the additional effect of access to finance on productivity?
In the coffee value chain, the additional effect of access to finance may be important to increasing 
productivity. We will randomize this component in the coffee value chain in addition to provision 
of extension. This question relates to priority learning Question 14 in the Learning Agenda: “What 
are the best linkage models to help small and medium sized producers, traders, and post-harvest 
market actors, who frequently lack collateral, registration, and credit history to access loans or 
other financial instruments to effectively expand their businesses?”
3. What practices are more likely to be adopted by smallholder farmers, and why?
Within the training, farmers will be exposed to new or different practices than they typically use, 
with the goal of increasing productivity and ensuring the quality and marketability of their produce. 
Both through process evaluation and impact evaluation, we plan to study which practices are 
more likely to be adopted by smallholders. This goal indirectly speaks to priority learning Question 
21: “In what context is it profitable for agricultural actors, particularly producers and processors, 
to adopt higher product quality standards for sales in higher value markets, including international 
markets?”
In this report, we discuss baseline measure of the concepts underlying these three research 
questions. Namely, to answer question one we are particularly interested in the indicators that 
describe coffee and bean productivity (e.g. production over cultivated area). To answer question 
two, we are interested in understanding how much access to finance farmers have, primarily 
access to credit, particularly among coffee farmers. To answer question three, we describe the
practices farmers are already implementing that should help them increase agriculture
4
productivity, as well as the techniques they have learned from extension workers in the past and 
whether farmers have adopted them or not.
Theory of Change
To answer these research questions, it is valuable to have in mind a theory about how changes 
among producers will occur. The primary outcome of interest is in increasing productivity within 
the crop (coffee, beans) that each farmer is growing. In this subsection we describe our theory, 
and assumptions behind it, about how productivity at the farm level would increase.
Recall that there are two mechanisms by which the MAS+ will interact directly with farmers that 
are described above—first, through agricultural training, and second, through finance in the 
context of the coffee intervention. So we first consider how agricultural training can influence 
productivity. A first way that trainings can increase productivity is by raising yields. To do so, we 
first assume that farmers are not using optimal techniques or practices on their coffee and/or 
beans; hence, it is possible for them to raise productivity by adopting techniques. Second, farmers 
must either attend trainings on those techniques, or learn about the techniques from their 
neighbors or peer farmers, who learn about them in the trainings conducted by MAS+. Third, 
farmers must either test adoption or adopt the techniques, and fourth, they must find the 
techniques both effective at making them more productive (e.g. in terms of yields) holding effort 
constant, or by making their farms more profitable, by reducing the amount of labor or effort it 
takes to produce the same or larger amounts. 
For coffee, it is not easy to adjust the amount of land used in the short term, but it is for beans. 
Therefore, bean farmers may also adjust the amount of land they use for beans; they may expand 
it if growing beans becomes more profitable relative to other crops using new techniques or 
practices. If not, farmers may continue to farm the same amount of land in beans or reduce the 
amount of land they use for beans.
The second pathway through which trainings can boost productivity is by improving quality, which 
should increase the prices received by farmers for their products. Some of the assumptions here 
are similar; for example, that farmers are not optimizing the quality of their produce, and that by 
adopting new techniques they can produce higher quality, which can be differentiated by the 
market, leading to higher prices. Farmers must again both learn about and then adopt such 
techniques, and the techniques must work (meaning lead to higher prices) so that farmers 
continue to use them. If market power on the buyer side of the market exists (monopsony or 
oligopsony power), it could dampen returns to adopting such techniques, which would lessen the 
incentives for adoption.
Throughout the theory so far, we have assumed that farmers are interested only in raising actual 
yields or the net value of production. Note that farmers are likely risk averse and may also worry 
about whether new techniques would increase the variability of yields, particularly in the context 
of climate change, which is raising temperatures and changing rainfall patterns. If so, then we 
must consider “risk adjusted” returns to the techniques. If farmers believe techniques they learn 
would both increase their incomes but also increase the variance of their incomes, the increase 
5
in income must be more valuable than the increase in variance. Alternatively, the techniques 
taught to farmers may be designed specifically to reduce variance of returns. For example, 
farmers may be encouraged to use bean seeds that are more adapted to higher temperatures or 
less predictable rainfall. If so, then farmers may adopt even if they do not perceive that yield 
increases would occur from adoption.
The second component to the theory of change is how added finance could increase coffee 
production. Here, we assume that finance implies credit (or liquidity); farmers do not have cash 
on hand to pay for either material or labor inputs when they are most needed to increase 
production (or the value of their crop). By making agricultural loans available, farmers can then 
take loans out and pay for missing inputs, increasing their productivity. We therefore are assuming 
that farmers would take out the loans and understand loan terms if made available, and would be 
able to obtain the needed inputs, hence increasing their productivity.
This discussion informs the selection of key indicators that we discuss in section 4.4. Before 
discussing key indicators, though, it is important to review the impact evaluation methodology and 
surveys to be conducted to measure outcomes.
3 Impact Evaluation Methodology
Impact evaluations are best defined as the response to question: “how is the outcome different 
than it would have been if the project had not been implemented?” However, we never observe a 
given beneficiary in both situations – with the project and without the project. Good impact 
evaluations resolve this problem through the use of a counterfactual. The counterfactual identifies 
what would have happened to the beneficiaries in the absence of the program. Understanding 
and accurately estimating this “parallel world” is a key challenge of impact evaluation, and is 
critical in understanding whether or not the intervention in question is effective and should be 
continued or expanded. A rigorous impact evaluation is one that combines the most appropriate 
and feasible evaluation design with the use of carefully selected indicators to document the impact 
pathway.
There are several methods for conducting impact evaluations, with the use of random assignment 
to create treatment and control groups producing the most rigorous results. Using random 
assignment, the control group will have – on average – the same characteristics as the treatment 
group. Thus, the only difference between the two groups is the program, which allows evaluators 
to measure program impact and attribute the results to the project. Random assignment is a 
preferred impact evaluation methodology and, in the case of coffee, we use a cluster 
randomized controlled trial (CRT) to identify the impacts of specific activities of MAS+. 
Random assignment is not always possible. For the activities of MAS+ for beans producers we 
conduct an evaluation with a control group but no randomization design by comparing the 
outcomes among participants to the outcomes among comparable non-participants. Namely, a
difference in difference analysis combined with propensity score methods to match participants
to non-participants. 
6
Both methodologies are explained in more detail in what follows.
Randomized Control Trial Methodology: Coffee Producers Impact Evaluation
Clustered randomized trials are a type of Randomized Controlled Trials (RCTs) in which groups 
of subjects, rather than individual subjects, are randomized in order to evaluate the impact of a 
particular intervention, policy or project. Succinctly, RCTs have become recognized as the best 
way to run ex ante impact evaluations as randomly selecting beneficiaries should imply that any 
improvement in outcomes is due to the program or project benefits, and not due to something 
else changing over time, since nothing else would affect the beneficiary group differently from the 
control group. Relative to individual level RCTs, CRTs reduce the risk of biases in the evaluation 
due to spillover and crossover effects between individuals and are a natural fit when the 
intervention is applied at the group level. When projects involve information sharing, community 
mobilization, access to trainings and extension through farmer groups, financial services provided 
by, e.g. village savings, they take advantage of the capacity communities (or farmer group) have 
to coordinate actions, disseminate and share information, and distribute resources. Therefore, a 
CRT is a natural fit for evaluating components of the MAS+. 
Several criticisms of RCTs have been advanced in both the economics and the more general 
program evaluation literature (Harrison, 2011; Ravallion, 2015; Deaton and Cartwright, 2017). 
One argument is that RCTs may not be ethical, as some potential beneficiaries are excluded from 
an intervention to measure its impacts. RCTs may generate unbiased estimates of average 
treatment effects for outcomes of interest, but they often lack information about how those 
outcomes came about. Finally, specific to agricultural projects, there are potential spillovers to 
neighboring areas. 
We plan to minimize these criticisms in the design of this RCT. First, we acknowledge that the 
MAS+ design is complex and it would not be possible to randomize access to the entire 
intervention; we therefore plan the impact evaluation around specific activities within the whole 
project. To ensure we understand how the whole project worked, the ongoing monitoring and the 
mid-term process evaluation will provide information about the effectiveness of other activities.
Second, we randomize at the farmer group or community level so jealousies do not arise within 
communities based on who gets access to specific components, as would occur with individual 
randomization. Furthermore, since randomization generates unbiased estimates of the benefits 
of programs, in combination with cost data the use of randomization can help us understand what 
projects create better value-for-money.
We note that external validity is always a threat to the proper interpretation of results from impact 
evaluations, whether randomized or not. We will test for impact heterogeneity, particularly with 
regards to gender and region. The latter might help us understand how interventions would work 
in other contexts. 
Finally, we note that there are specific difficulties related to agricultural interventions. There is a 
great deal of chance for spillovers to control groups. We plan to deal with these issues in two 
7
ways; first, we will design the survey form to learn about potential spillovers to the control group. 
Second, we will ask for cell phone numbers of household members at baseline; in a previous 
IFPRI work tracking migrants, we found that phone numbers were the most effective way to track 
migrants. 
Propensity Score Methodology: Beans Producers Impact Evaluation
Under the assumption that we can identify “treated” communities or areas reasonably well, we 
will apply propensity score methods to estimate impacts of MAS+ activities in bean growing areas. 
Through comparisons with experimental estimators, Heckman, Ichimura and Todd (1997, 1998) 
show that propensity score matching can provide reliable, low-bias estimates of program impact, 
under the following assumptions: (i) the same data source is used for participants and non￾participants, (ii) the data include meaningful explanatory variables capable of identifying program 
participation and outcomes, and (iii) participants and non-participants have access to the same 
markets. This logic can be extended to any estimates based on either covariate matching or 
reweighting estimators.
To measure impacts of MAS+ in the beans value chain we explicitly make the following 
assumptions. First, we assume that the households in the control group are similar to participating 
households. Second, the treatment and control groups have access to roughly the same markets 
(as above), and we assume that the explanatory variables that are included can identify 
hypothetical program participation among the control group. Note that we use retrospective 
variables for estimating propensity scores, so we need to assume that any recall biases that exist 
are similar between the treatment and control groups for the period under study. Third, we assume 
that households in the control group contain enough households that would have been group 
members to make meaningful comparisons between the treatment and control groups. Fourth, 
we assume there are no spillovers from the treatment to control groups, which could be an 
innocuous assumption, but may not be if growing techniques spread rapidly.
Unlike an RCT, in which the treatment and control groups are randomly selected, in using 
propensity score methods, the goal is to find control observations that are as similar as possible 
to project participants but did not have access to the program. The matching of control 
observations to treated observations must take place with observable variables, and then we must 
assume that unobservable characteristics are also roughly equivalent between the treatment and 
control groups. This assumption is quite strong; in this context, an important challenge will be 
finding control group observations that would have participated in trainings had they been offered, 
since training groups will not be set up in villages in the control group. 
Second, matching methods (including PSM) can be criticized as being easy to manipulate, based 
on the covariates that go into the matching process. To ensure that matching is not manipulated, 
we will use an algorithm to choose variables to include in propensity scores (Hirano and Imbens, 
2002; Imbens, Newey, and Ridder, 2005). For the algorithm to function properly, it needs a 
rigorously collected list of baseline characteristics. 
8
To further to improve the design we ensure that the control group is comparable to the project 
beneficiaries using a purposely selected baseline to identify non-participants and construct a 
control group that is similar to the participants at the start of the project. The method to select the 
baseline for the beans producers involves matching of just two sampling units (clusters) to 
construct matched pairs. With this approach, the number of treatment and comparison units is 
the same and the selection of individual items is correlated, since either both members of a 
matched pair are included in the sample, or neither is. Not only is the selection event correlated 
(among sampling units), but the two items in each cluster pair are highly correlated with respect 
to the matching variables. 
In addition, to account for external selection bias given that the projects purposively target 
particular household types, in the evaluation we compare the changes – instead of the level – of 
a given indicator between the group of participants and the control group. Assuming the change 
in the indicator in the control group is a good representation of what the change in the indicator 
would have been among the beneficiaries, this “difference-in-differences” estimate can provide a
valid way to neutralize the external selection bias and hence provide an unbiased assessment of 
the project’s impact. 
4 Data Sources 
Small farmers often keep no records and much information collected about agricultural production 
activities depends on farmer recall. However, the easiest and often least expensive way of 
obtaining agricultural production data is to simply ask farmers and other producers directly. The 
evaluation and this report main source of information comes from personal interviews with 
farmers. Below we describe the instrument and platform used in data collection, and statistical 
survey design.
4.1 Survey Design 
IFPRI designed the survey instrument, in consultation with both TechnoServe and the local survey 
firm that was contracted to conduct the survey by TechnoServe, ESA Consultores. The software 
application to administer the questionnaire through computer-assisted personal interviews (CAPI)
was written by ESA Consultores. The survey instrument was translated into a CAPI application in 
the software package Survey Solutions. This software was chosen by its flexibility to handle 
complex modules, which facilitated the order and flow of the interview. The software includes all 
necessary crosschecks and directions of how to answer questions for the enumerators, i.e. help 
screens and error messages are coded so that information can be corrected on the field. This 
electronic platform supports complex skip patterns and has simple questionnaire navigation 
capabilities. The platform also automates error-prone aspects of data collection, such as the filling 
in of ID codes or selecting at which level would each section be asked. Finally, this platform allows 
the capture and integration of non-traditional data, such as audio-visual media and geographical 
coordinates. The finalized surveys are validated in extenuating detail by the time the supervisors 
see them, and then they go to the server after a final check. IFPRI made a thorough review of the 
survey application before fieldwork to ensure that the CAPI application minimize errors in the data 
9
collection stage, maximize the internal consistency of the questionnaire and provided 
opportunities to identify and correct errors while in the field. 
The content of the questionnaire revolves around 3 themes:
 Agriculture Production and Commercialization. Measuring the production,
productivity, and commercialization of coffee and dry-beans
 Extension and Technology Practices. Measuring the prevalence of agricultural 
practices of coffee and dry-beans, participation in extension programs, and access to 
other sources of agricultural information.
 Producer Association and Credit Access. Measuring and its participation in community 
groups, communal activities, producer associations, participation in the credit market, and 
household savings. 
The questionnaire design benefited from IFPRI’s experience in Honduras, from the inputs from 
the monitoring and evaluation team of TechnoServe, and the inputs of ESA Consultores. The 
questionnaire adopted was the product of this exchange 
The survey instrument includes the following survey modules:
 Module 1: Household composition, education and demography
 Module 2: Dwelling characteristics 
 Module 3: Land access: Coffee
 Module 4: Coffee production and commercialization 
o Inputs and labor
o Agricultural technologies and practices
o Pests
o Extension
 Module 5: Land access: Beans
 Module 6: Beans production and commercialization 
o Inputs and labor
o Agricultural technologies and practices
o Pests
o Extension
 Module 7: Agriculture production and commercialization of other crops
 Module 8: Investments, assets, social organizations and collective action
 Module 9: Access to credit markets
 Module 10: Opinions, perceptions and access to information
 Module 11: Off-farm labor
 Module 12: Transfers and remittances
Below we briefly summarize the objectives of each module.
Cover
10
The objective of this section is to identify each interview with a unique sample frame identification
number, the geographic location, and directions to arrive to the household. GPS coordinates are 
collected for each household. The GPS coordinates allow us to locate the household and obtain
other geographic information. 
Consent
An informed consent script was read to the survey respondent explaining the main objective of 
the survey, the voluntary nature of it, that the information would remain strictly confidential, and 
that the information would be used only for research purposes. Farmers were asked if they agreed 
to participate and provided a verbal consent, which was recorded by the software. 
Module 1: Household composition, education and demography
Collects information on the household structure, determine household members, and the 
education level of these members. We also obtained this demographic information and education 
levels for non-household members in charge of crop management.
Module 2: Dwelling characteristics
Collects information about the living situation of the household. We obtain information on the value 
of the house where they live and household related expenses; as well as characteristics of the 
dwelling. These include: materials use on the floors, walls and ceiling, access to electricity, potable 
water and energy use.
Module 3: Land access for coffee
Collects information on the land ownership for coffee producers. It provides general 
characteristics of parcels, like accessibility to roads and markets, irrigation status, size, etc.
Module 4: Coffee production and commercialization
Collects information about coffee production, commercialization, practice and techniques, pests 
and training and extension. It is exclusively asked of coffee producers.
Sub-Section 4.1: Coffee production 
Collects information on coffee production at the variety level. For each variety planted, it 
gathers information about: total area, area in plantía, area in commercial production and 
area in recepa, plant density, age of plants, and total production. It also gathers information 
on losses, both field losses and post-harvest losses and productivity expectations under 
good, regular and bad conditions. 
Sub-Section 4.2: Coffee sales and inputs
Collects information about the total number of sales in the last 12 months. Once the 
number of cuts a year is obtained, detailed information about the buyers, quantity sold per 
type of processed coffee, value of the sale, timing, contract, quality and price is obtained 
11
at the cut level. We also gather information about expectations of sales under good, 
regular and bad market conditions, labor use and input use. 
Sub-Section 4.3: Coffee practices and techniques
Collects information about important practices targeted by the MAS+ program and 
suggested by the TechnoServe technical team. These practices are divided by the activity 
level and include: plot management, harvesting practices, post-harvest practices and 
coffee quality. 
Sub-Section 4.4: Pests in coffee
Collects information about the main pests that affected the coffee crop during the last 12 
months. For each pest, collects information about estimated losses, control measures, 
cost of control measures and if the farmers received any help to prevent such pests. 
Sub-Section 4.5: Participation in coffee training and extension
Collects information about the participation of household members in extension programs 
or training, on or off the farm. It inquiries about the entity that provided extension, costs, 
specific practices learned, frequency and use of such information learned.
Module 5: Land access for beans 
Collects information on the land ownership for bean producers. It provides general characteristics 
of parcels, like accessibility to roads and markets, irrigation status, size, etc.
Module 6: Beans production and commercialization
Collects information about beans production, commercialization, practice and techniques, pests 
and training and extension. It is exclusively asked of beans producers.
Sub-Section 6.1: Beans production 
Collects information about beans production at the variety level for each season (primera 
2017 and postrera 2017-2018). For each season, it obtains information for varieties
planted, area planted and harvested, total production, sales, price and production 
expectations under good and bad conditions. 
Sub-Section 6.2: Beans sales and inputs
Collects information about sales in the last 12 months the buyer, and expectation of prices 
under good and bad market conditions. It also collects information labor use and input use 
and costs associated.
Sub-Section 6.3: Beans practices and techniques
Collects information about important practices targeted by the MAS+ program and 
suggested by the TechoServe technical team. These practices are divided by the activity 
level and include: planting, harvest and post-harvest practices. 
12
Sub-Section 6.4: Pests in beans
Collects information about the main pests that affected the bean crop during the last 2 
agricultural seasons. Then for each pest, it asks specific questions, such as estimated 
losses per season, control measures, cost of control measures and if the farmers received 
any help to prevent such pests. 
Sub-Section 6.5: Participation in beans training and extension
Collects information about the entity that provided extension, the training or extension 
cost, specific practices learned in such training activities, frequency and use of the
information learned, post-harvest, quality and certification, marketing, finance and 
contracts and market access).
Module 7: Production and commercialization of other crops
Collects information about all other crops planted. For each crop the module collects the number 
of harvests in the last 12 months, the value of production, whether it was consumed by the 
household, and the sales value. 
Module 8: Investments, assets, social organizations and collective action
Collects information about the main investments and equipment of the household. In addition, this 
section asks about each type of organization that exists in the community, the level of involvement 
in such organizations, and the main benefits they obtain from such organizations. 
Module 9: Access to credit markets
Collects information about the sources of credits used by households, the degree of participation 
in the credit market, and household savings. Access to credit is of paramount importance for 
agricultural production and the household economic welfare. We do not limit question to formal 
credit; we include informal sources, input, services/machinery, and other assets that the members 
received against a promise to pay in the future- be it in product, cash, or labor.
Module 10: Opinions, aspirations and information and communication technology
Elicits the opinion of the interviewee with respect to agricultural technologies and events, and the 
diffusion of new techniques in the household’s social network, farmers’ perceptions of aspirations 
for the future, self-esteem and locus of control. We end the section asking about sources of 
agricultural information and use of communication technologies.
Module 11: Off-farm labor
Collects information about the productive activities of each individual in the household over 12 
years old. It serves to know the different sources of income of the household besides agricultural 
income, both salaried and self-employment activities.
Module 12: Transfers and remittances
13
This section asks about the main sources of transfers and remittances, frequency of transfers, 
amounts transferred and use of such transfers. 
Back Cover
Collects information about the result of the survey, the personnel that participated, and any further 
comments from the enumerators and the supervisors.
The survey instrument is provided as an attachment/appendix to this document.
4.2 Data Collection: Training, Pilot and Fieldwork
The Baseline Survey was conducted between July and August of 2018. IFPRI researchers, jointly 
with ESA Consultores, conducted a 14-day training between June 4th to June 17th: 7 days with 
field managers and groups’ supervisors and the remaining 7 days with the whole team of 
enumerators to review agricultural concepts facilitated by TechnoServe. Enumerators and 
supervisors alike were evaluated to be able to be part of the final team of interviewers. The training 
also included 2 pilot activities with both coffee and beans farmers; these pilot farmers were in 
areas outside of the impact evaluation survey sample and were not anticipated to be part of the 
control or treatment groups. Pilot activities were supervised by ESA, IFPRI and TechnoServe 
teams. These pilot activities were useful to refine the survey instrument, collect farmer’s 
impressions, refine technical questions, adjust the interview length, and further evaluate 
enumerators interviewing skills.
Training was based on the following topics:
 Introduction to the survey: survey objectives, sample, survey modules, survey 
implementation, confidentiality, interviewer’s role, assignment to supervisors
 Conducting the interview: giving general guidance, approaching the household, building 
rapport, converting refusals, obtaining informed consent, ensuring privacy, asking 
questions, probing, following interview instructions on the questionnaire and tablet, noting 
differences between the printed questionnaire and tablet screens, and flagging issues to 
be discussed with the Field Supervisor
 Questionnaire content: household roster; informed consent; dwelling characteristics; 
improved agriculture technologies and productivity, etc.
 Fieldwork procedures: following field team members’ roles and responsibilities, 
managing the household interview, reporting to the field supervisor, following up missed 
interviews, ensuring high-data quality, and monitoring and review of interviewers’ 
performance
 Entering and managing data on the tablet: understanding the tablet and screen 
components, starting a questionnaire on the tablet, navigating the questionnaire, 
advancing through survey modules and groups, entering responses, dealing with refusals, 
troubleshooting, transmitting data
 Completing survey modules: knowledge of general instructions, administering each 
survey module, asking questions, and entering responses question by question
14
 Improved agriculture technologies: understanding improved agriculture technologies, 
measuring crop productivity 
At the conclusion of the interviewers’ training, the entire field team conducted a pilot test in rural 
communities and included individuals who are similar to the planned survey respondents. The 
pilot areas were not part of the sample. At the end of the pilot, everyone participating in the pilot 
test met to discuss issues and challenges and to identify solutions. Issues with the survey 
instrument and program were communicated to IFPRI and corrections were made, documented, 
and tested before the fieldwork started.
Revisions to the instrument and data entry program and other preparations for fieldwork took 
several days, so there was a brief period between the pilot and the initiation of fieldwork. 
Interviewers worked in teams of 5: 4 interviewers and 1 supervisor, for a total of 8 teams (mixed 
female and males). The teams covered equally both the coffee and beans samples. Two 
additional field supervisors were in charge of visiting different areas, homogenizing the protocol 
across teams and providing general coordination. The ESA team was also responsible for 
reviewing the quality of the data. ESA delivered weekly reports to the IFPRI team and 
TechnoServe. Throughout the data collection, IFPRI provided timely responses to issues found in 
the field to ensure homogeneous protocols across survey teams.
4.3 Statistical Power and Sample Selection
The evaluation will primarily use two surveys: a baseline survey before the beginning of activities 
and endline survey after the conclusion of the MAS+. The baseline survey will serve two purposes. 
First, it provides a representative baseline estimates for measurement of performance indicators 
that will be measured among beneficiaries throughout the project. Second, it provides a baseline 
for impact estimates, which will compare treatment groups to a control group of non-participants. 
As such the sample design follows an analytical survey design. The purpose of an analytical 
survey is to develop detailed models that describe the relationship of dependent variables to a 
variety of explanatory (independent) variables (including treatment variables). The principal 
objective of the surveys is to provide high precision for estimates of the difference-in-difference 
estimates of program impact.
We designed the sample to target an increase in coffee yields of 24% (40% among participants, 
at a 60% participation rate) and bean yields of 21% at the 5 percent significance level and 80 
percent power, through the baseline and endline surveys. 
Assuming sampling clusters with an average size of 25 producers; a 20 percent of the variance 
in the endline outcomes using covariates collected at baseline, which includes the baseline 
outcome in an ANCOVA model (e.g. McKenzie, 2012); and intracluster coefficient of 0.05; we 
estimated the minimum detectable effects for coffee and beans yields. In Table 1 we find that for 
bean yields, the minimum detectable effect is 21% yield increase with a sample of 500 farmers 
per treatment condition; for coffee, we find that with below 12 clusters per treatment the difference 
could be detected with 80% power. However, for small numbers of clusters (e.g. less than 30) 
15
standard t-tests are likely to over-reject the null hypothesis for a given level of statistical 
significance, and require adjustment (e.g. Cameron and Miller, 2015). As a result, we included
20 clusters per treatment arm in the coffee sample, so the minimum detectable effect improves 
and drops to 19 percent of the mean. The resulting sample is 40 total clusters per treatment arm; 
the minimum detectable effect on the logarithm of the value of agricultural output is then about 19 
percentage points.
TABLE 1 PROJECTED BASELINE SAMPLE CHARACTERISTICS AND MINIMUM DETECTABLE EFFECT
Variable Mean
Standard 
Deviation
(after 20% 
variance 
reduction)
Clusters per 
Treatment 
Condition
Size of 
Clusters
Total 
Sample 
Size
Minimum 
Detectable 
Effect 
Achievable
% of 
Mean
Bean Yields 18.1 12.2 N/A N/A 1000 3.9 21.5
Coffee Yields 63.9 29.1 40 25 1000 12.1 19.0
Notes: Yields in quintales per hectare (qq/ha).
At the time of sample selection, and due to constraints in obtaining a sample frame or listing 
of all farmers in the treatment and control clusters from where to select the treatment and 
control sample producers, the effective sample varies slightly from the number of clusters and 
the size of each cluster that was initially proposed. Table 2 and 
Table 3 show the sample distribution in the initial sample design for coffee and beans, respectively, 
at the start of the fieldwork. In the survey design, we set out to interview 500 producers in 20 
clusters in each treatment condition and for each commodity. Treatment and control producers
were selected from the list of potential participants provided by TechnoServe when available. This 
was the case for all treatment and control farmers in the case of coffee. For the coffee sample we 
listed 550 control coffee producers in 40 clusters and 572 treatment coffee producers in 38 
clusters (with additional pre-selected farmers as replacements to reduce attrition).
In the case of beans, we could not obtain the list of farmers in advance of the fieldwork. Control 
farmers in the beans sample that were not listed at the beginning of fieldwork were selected 
randomly by IFPRI from a listing of beans producers obtained before visiting the cluster. For the 
beans sample we listed 71 control beans producers in 34 clusters and 503 treatment beans 
producers in 34 clusters prior to data collection (with additional pre-selected producers as 
replacements to reduce attrition). During data collection, more complete lists of farmers were 
developed in control areas; data were then collected among the more complete list within the 34 
clusters.
16
TABLE 2 COFFEE SAMPLE DESIGN
Control Treatment
Department Producers Replace￾ments Total Clusters 
Control Producers Replace￾ments Total Clusters 
Treatment
COMAYAGUA 168 20 188 12 164 26 190 11
EL PARAÍSO 187 23 210 14 122 15 137 8
FRANCISCO 
MORAZÁN 30 2 32 2
LA PAZ 10 10 1 15 3 18 1
OLANCHO 129 9 138 9 166 20 186 11
YORO 56 4 60 4 75 6 81 5
Total 550 56 606 40 572 72 644 38
TABLE 3 BEANS SAMPLE DESIGN
Control Treatment
Department Producers Replacements Total Clusters 
Control Producers Replacements Total Clusters 
Treatment
EL PARAÍSO 29 29 5 69 12 81 5
FRANCISCO 
MORAZÁN 15 15 2 30 6 36 2
OLANCHO 16 16 16 240 38 278 16
YORO 11 11 11 164 26 190 11
Total 71 71 34 503 82 585 34
Survey Response Rates
Table 4 shows the distribution of the completed surveys and the response rates by department 
for coffee in panel A and for beans in panel B. The overall attrition was low, with above 95 percent 
completed in both the coffee and beans sample. The table shows the difficulty we had identifying 
control producers in the beans samples in some departments and that we recuperated some of 
the sample by selecting additional clusters in other departments. For example, in Yoro the coffee 
farmers listed were re-assigned to Olancho, thus the response rate is above one.
17
TABLE 4 COMPLETED SAMPLE AND RESPONSE RATES
Control Treatment
Completed Expected Response 
Rate
Completed Expected Response 
Rate
Panel A: Coffee Sample
COMAYAGUA 151 168 0.90 153 164 0.93
EL PARAÍSO 163 187 0.87 110 122 0.90
FRANCISCO MORAZÁN 22 30 0.73
LA PAZ 9 10 0.90 13 15 0.87
OLANCHO 150 129 1.16 190 166 1.14
YORO 0 56 0.00 0 75 0.00
Total 473 500 0.95 488 500 0.98
Panel B: Beans Sample
EL PARAÍSO 66 57 69 0.83
FRANCISCO 
MORAZÁN
22 33 30 1.10
OLANCHO 228 240 240 1.00
YORO 159 163 164 0.99
Total 475 500 0.95 493 503 0.98
Note: The total “expected” is not equal to the sum by department as we listed (“expected”) more potential households
to be able to arrive at the 500 interviewed farmers per treatment condition.
Effective Minimum Detectable Effects
Given the changes in the sample and the difference in the survey instrument used in the surveys, 
it is important to verify the minimum detectable effects (MDE) within the realized sample.
The power analysis using estimated means and variances from the survey is presented in Table 
5. The MDE is 25% of the mean for the beans yields and 30% of the mean for the coffee yields. 
Compared to the initial sample design above (21% for beans and 19% for coffee) these are larger 
and highlight the need to maintain low attrition in the endline and ensure that the production and 
area measurements are of good quality.
In summary, the results show that the sample design is well powered to detect economically 
significant differences in yields across the treatment and control groups in the sample.
TABLE 5 MINIMUM DETECTABLE EFFECTS OF SAMPLE
Variable Mean
Standard 
Deviation
Clusters 
per 
Treatment 
Condition
Avg Size 
of 
Clusters
Total 
Sample 
Size
MDE
Achievable
MDE after
(after 20% 
variance 
reduction)
% of Mean
Bean Yields 
(Primera) 12.73 8.6 37 T/39 C 12.7 968 4.06 3.2 25%
Bean Yields 
(Postrera) 9.31 6.4 37 T/39 C 12.7 968 3.02 2.42 26%
Coffee Yields 45.78 36.22 38 T/39 C 12.6 961 16.97 13.55 30%
Notes: Coffee yields in qq/ha wet parchment; for the MDE calculation the standard deviation is the important quantity. 
18
5 General Characteristics of the Intervention Area
Honduras is the second poorest country in the Western Hemisphere. In 2017, it reported a poverty 
rate of 64.3 percent below the national poverty line, according to national official records. Two out 
of three people are poor in the country poverty is more severe in rural areas than in urban areas.
In 2013, 46.6% of the population or 3.8 million people lived in rural areas2
. With a poverty rate of 
69.3% in rural areas, this is 2.6 million people living in poverty in rural areas. Comayagua, El 
Paraíso, Olancho and Yoro, the departments where most of the MAS+ beneficiaries are 
concentrated, are home to approximately 1.5 million of such families living in poverty. Figure 1
shows the geographic distribution of the sample.
FIGURE 1: COFFEE AND BEANS SAMPLE GEOGRAPHIC DISTRIBUTION
 
2 XVII Censo de Poblacion 2013, INE(http://www.ine.gob.hn/)
19
In 2017, 40.7% of the country’s population lived in extreme poverty. In rural areas, with a poor 
population being 58.8% many farmers to afford the basic food basket3
. Poor households in rural 
areas are further below the poverty line than in urban areas. 
According to the last National Census, in 2013, 54.3% of the population had unsatisfied basic 
needs. This share has barely decreased since the previous census in 2001, where the share of 
the population with unsatisfied basic needs reached 55.4% of the population. Since 2001, there 
has been a decrease in the share of the population that has no access to most basic needs or 
services: access to water, sanitation, access to basic education and overcrowding. But there are 
two particular necessities that have seen an increase in the share of the population that has no 
access to these services: the share of the population without subsistence capacity (which 
increased from 17 to 30% from 2001 to 2013) and the share of the population with dwellings in 
bad conditions (which increased from 0.5 to 9% from 2001 to 2013). 
A third of the country’s population is employed in the agricultural sector. Agricultural households 
in the area of intervention cultivate three main traditional crops: maize, beans and coffee on small 
plots, often on hillsides. The country is also vulnerable to national disasters, such as hurricanes 
and droughts, and the area of intervention is particularly characterized by little rainfall and variable 
climatic conditions. The poor economic circumstances of the area condition its people to struggle 
under such variable environments and do not allow them to mitigate shocks with risk management 
mechanisms, such as credit or insurance. The lack of resources constraints people in the area to 
invest in improved technologies, such as irrigation systems, that will also allow people’s resilience 
to environmental shocks. The use of traditional cultural practices produces poor yields, which 
leads to poor income and nutrition, and depletes soil of nutrients, which also leads to 
environmental degradation. On top of this, farmers’ access to markets is hindered by poor roads 
and long distances. The coffee sector is also vulnerable to external shocks, such as the declining 
of international coffee prices. 
 
3
 INE, LVIII Encuesta Permanente De Hogares de Propósitos Múltiples - EPHPM, June 2017, 
www.ine.gob.hn
20
6 Analysis of Coffee Producers in MAS+ Evaluation
In this section, we describe the sample of coffee growers. After describing the households 
demographically, we examine agricultural statistics among households related to coffee. We 
include balancing tests throughout the themes in these sections to show the validity of the RCT 
design for the coffee samples and that the evaluation design balances important outcomes across 
treatment and control groups.
6.1 Demographic Characteristics
First, we examine demographic characteristics among the sample in Table 6. The average 
household has 4.77 members, distributed as follows. On average, the households include 0.69 
members aged 5 years old or under, split nearly evenly by gender. Members include 1.1 boys and 
girls combined, on average, aged between 6 and 14 years old (Table 6). Most other members 
are working age adults (15 to 64 years old), as there are only 0.2 elderly members on average. 
There are fewer working aged women on average than working aged men in coffee sample 
households, but otherwise households are relatively gender balanced.
TABLE 6: DEMOGRAPHIC CHARACTERISTICS FOR COFFEE SAMPLE
Mean SD Min Median Max Observations
Number of people living in the house 4.77 2.12 1 5 14 961
Boys under 6 0.35 0.64 0 0 5 961
Girls under 6 0.34 0.62 0 0 6 961
Boys between 6 and 14 years old 0.52 0.92 0 0 11 961
Girls between 6 and 14 years old 0.58 1.43 0 0 14 961
Men between 15 and 64 years old 1.18 0.71 0 1 3 961
Women between 15 and 64 years old 0.88 0.42 0 1 3 961
Men older than 65 0.11 0.32 0 0 2 960
Women older than 65 0.09 0.29 0 0 2 961
People who manage agricultural activities within the coffee sample range from 11 to 87 years old 
(Table 7). On average, the primary farmer is 41 years old. About 42% of people working in 
agriculture in the coffee sample are women. The average number of years of education is 5 years;
most farmers have completed primary school. The literacy rate is 85%. Looking at the same 
characteristics by gender (Table 8), we observe that women are just a little older on average than 
men (42.5 years old as opposed to 40 years old for men). Average years of education and literacy 
ratios for both genders are about the same.
21
TABLE 7: DEMOGRAPHIC CHARACTERISTICS
Age Female Years of Education Literacy
Mean 41.13 0.42 4.67 0.85
SD 16.11 0.49 3.30 0.36
Min 11 0 0 0
Median 40 0 6 1
Max 87 1 18 1
Observations 2198 2197 2197 2198
Note: These demographic characteristics have been calculated for all people in charge of agricultural decisions in the 961 coffee 
households.
TABLE 8: HUMAN CAPITAL CHARACTERISTICS, BY GENDER
Male Female
Age Years of Education Literacy Age Years of Education Literacy
Mean 40.16 4.68 0.85 42.48 4.65 0.85
SD 17.14 3.28 0.36 14.45 3.33 0.36
Min 11 0 0 14 0 0
Median 39 6 1 40 6 1
Max 87 18 1 85 17 1
Observations 1280 1280 1280 917 916 917
Note: These demographic characteristics have been calculated for all people in charge of agricultural decisions in the 961 coffee 
households.
Besides asking about income earned from coffee sales (analyzed below), we asked farmers if 
they had other sources of income. We use this information to calculate total household monthly 
income and estimate what percentage coffee sales represent out of total household monthly 
income. We find that income earned from coffee sales represents on average 69% of household 
income (Table 9). It is worth mentioning that these shares are only estimated for households 
reporting positive earnings from either agricultural or non-agricultural activities (907 farmers).
TABLE 9: AGRICULTURAL AND NON-AGRICULTURAL INCOME
Monthly Non-Agricultural 
Income in USD
Monthly Agricultural 
Income in USD
Share that Coffee Income 
represents
from Total Income
Mean 448.74 267.21 0.69
SD 1261.02 570.36 0.37
Min 4.10 0.03 0.00
Median 147.60 124.76 0.98
Max 14760.00 10284.17 1.00
Observations 478 886 907
Note: Only 478 farmers from the coffee sample report earning income from other non-agricultural activities; 886 farmers report 
earning income from selling coffee to the market. The share that coffee income represents from total income is calculated only for 
farmers earning positive amounts from either of both sources (907 farmers)
22
6.2 Agricultural Outcomes
We next examine agriculture practiced by sample households, focusing on coffee production. 
First, we examine the amount of land being cultivated in coffee, before discussing production and 
yields. We then discuss coffee sales in the past 12 months. Sample for coffee consists of 961 
farmers; across the different estimates the sample may vary because of non-response, non￾applicability or the answer were considered extreme outliers.
Size of Cultivated Land
Measures of area are fundamental components of agricultural statistics, as they are required for 
calculating many indicators of productivity. Ideally, measures of both production and area should 
be highly accurate. However, errors in the denominator (area) magnify any errors in the numerator 
(production); thus, accurate measures of area are arguably more critical to minimizing potential 
errors in calculating agricultural yield. As many farmers in developing countries have no real 
means of accurately determining how much land they use to produce crops or other agricultural 
products, accurate measures of area can be difficult to obtain.
The relevant measure for area is the area planted (cultivated), rather than the area harvested, or 
owned. This is an important distinction since not all parts of a field or farm that are planted will 
necessarily produce any yield or be harvested. Area cultivated is determined by farmer estimates 
(self-reported) of how much area they cultivate in the previous 12 months.
Table 10 shows the estimates of land use for coffee production. The average household has 2.3 
hectares (ha) planted with coffee; there is substantial variation in the total area planted, with the 
standard deviation slightly larger than the mean.4 The median household in the sample has 1.4 
ha planted with coffee, so the average is skewed by some larger farms. For area planted with 
coffee either in plantía, recepa, or in commercial production, we present conditional averages; 
only non-zero values. Note that not all households have coffee currently in commercial production; 
54 surveyed farmers fit this category. We estimated that the area planted with coffee in 
commercial production is 1.9 ha, which is lower than the total area planted. About half of 
households have coffee in plantía, and average area planted in plantía is 1.9 ha. Medians are 
somewhat lower, suggesting that there are some households with larger holdings.
 
4 Most of the responses in the survey were given in manzanas, a standard unit of land area in Honduras; a 
Honduran manzana is equivalent to approximately 0.7 hectares. 
23
TABLE 10: SIZE OF PLANTED AND CULTIVATED AREA FOR COFFEE SAMPLE
Total area planted 
with coffee (in HA)
Area planted with 
coffee in plantía (in 
HA)
Area planted with 
coffee in 
commercial 
production (in HA)
Area planted with 
coffee in recepa (in 
HA)
Mean 2.3 1.1 1.9 0.9
SD 2.9 2.1 2.2 0.9
Min 0.0 0.0 0.0 0.1
Median 1.4 0.7 1.4 0.7
Max 55.7 27.8 31.3 6.3
Observations 925 506 871 150
Notes: The Total sample for coffee consists of 961 farmers. Differences in sample sizes for total area are due to 3 farmers that 
do not report area planted and observations that were determined to give extreme outliers in the yield estimation. All other 
differences are due to farmers that have all their plantation either in plantía, in commercial production, or for recepa, but not all of 
them together.
The modal and median farmer has one plot of coffee; as some farmers have more than one plot, 
the average is 1.7 plots across the sample (Table 11).5 A few farmers have substantially more 
coffee plots. Coffee farmers in the sample do not all have title to their plots; rather, farmers of 
less than half of the plots in the sample have title, which might affect their access to finance.
TABLE 11: NUMBER OF TOTAL PLOTS AND OWNED PLOTS FOR COFFEE SAMPLE
Number of plots for coffee Number of plots that are owned with 
property title
Mean 1.71 0.78
SD 1.12 1.12
Min 1.0 0.0
Median 1.0 0.0
Max 10.0 9.0
Observations 961 961
Note: Coffee sample consists of 961 farmers. All farmers have at least one plot where they plant coffee.
Coffee Production 
Asking farmers to estimate their total production is perhaps one of the most convenient and least 
expensive ways to gather data on agricultural production. It is often employed through surveys, 
relying on the ability of farmers to remember how much they might have harvested of a crop or 
from a plot. The accuracy of production estimates from farmer recall varies tremendously; 
evidence of farmer error has been based primarily on differences between farmer estimates of 
production and those calculated with individual measurement approaches. Additionally, if there is 
 
5 The average landholdings are 2.3 hectares and there are 1.7 plots held per farmer, the average plot size 
is 1.35 hectares.
24
not bias in farmers’ recall, the mean from a large enough sample is an unbiased estimate of the 
true mean. In the absence of bias or if bias is constant over time, recall estimates can provide 
accurate estimates of change over time. 
Farmers may express production in local units, which must then be converted to standard units 
such as pounds, kilograms, etc. Errors easily accumulate through multiple conversions and 
rounding. The information was collected using the units used by the farmer and selected with from 
a pre-loaded list of units that can be accurately converted to standard units to minimize errors and 
ensure that the units could be accurately converted in the analysis. In the analysis that follows we 
convert all other weight measurements to quintales (qq) (1 quintal = 100 lbs= 45.359 kg).
Coffee can provide multiple products and depending on the stage of the processing have different 
weight and densities. For example, coffee may be sold as either berries, wet or dried, etc. To 
measure coffee production, we standardize all measures to pergamino húmedo or wet parchment 
using conversion factors proposed by IHCAFE (Instituto Hondureño del Café) and Technoserve 
that adjust for the humidity and processing in the weight. The following conversion factors are 
used:
 100 lbs café uva or coffee in berries = 43.7 lbs wet parchment
 100 lbs café uva or coffee in berries = 21.53 lbs dry parchment
 100 lbs pergamino seco = 81.92 lbs green parchment
All production values for each processing level were converted to quintales (qq) and then 
processing levels were all converted from quintales to quintales in pergamino húmedo or wet 
parchment to obtain a standardized measure of production for coffee. For example, if the 
respondent said total production was 50 cargas in uva or berries, we convert this to 100 qq in 
berries, since each carga has 2 qq; then 100 qq in berries is equivalent to 43.7 quintales in wet 
parchment (qq WP) using the conversion factor above.
In practice, most farmers selected the standard unit of measurement quintal and wet parchment 
as the units of production and presentation, as shown in Table 12. This minimizes the effect of 
possible errors in the conversions.
TABLE 12 UNITS AND PROCESSING LEVELS SELECTED BY RESPONDENTS
Units Processing Level
Freq. Percent Freq. Percent
Quintal 1,347 93.0 Uva/Berry 138 11.1
Carga (2 quintales) 24 1.7 Wet parchment 862 69.2
Libras 6 0.4 Dry parchment 67 5.4
Arrobas 1 0.1 Oro/Green 179 14.4
Cans/Latas (sp. no. Per quintal) 54 3.7
Other (sp. no. Per quintal) 16 1.1
Total 1,448 100 Total 1,246 100
Notes: Observations at the variety level; each household can have multiple varieties.
25
Total production is defined as the amount that is produced, regardless of how it was ultimately 
used. It also includes any postharvest loss (i.e. postharvest losses are not subtracted from total 
production). All production quantities are considered only for farmers having positive plantía and 
commercial production or field or post-harvest losses. To identify outliers, we processed the 
production and area information to compute yields and then identified yields with implausible 
values given the observed distribution. Once we identified these observations, we also excluded 
those observations from the average area and average production estimates (fewer than 10 
observations on the top of the distribution of yields were excluded).
Table 13 shows coffee production in quintales of wet parchment coffee. We estimate total 
production, production in the area designated as plantía, production in the area designated for 
commercial production, and then losses in the field and post-harvest. The spread of total 
production is consistent with the skewed distribution we observe in area planted. Total production 
at the median is 34.3 qq, and commercial production is 31 qq at the median; almost half the 
estimated mean, showing that there is a long tail over high production values after excluding 
extreme outliers. Production itself is quite variable, with standard deviations above the average. 
Finally, it is notable that only a small share of households report post-harvest losses; even then, 
average post-harvest losses are small or around 10 percent. Note that as the production 
estimates are all conditional (they must be greater than zero), so the subcategories do not add 
up to the total mean in the first column.
TABLE 13: COFFEE PRODUCTION 
Coffee 
production in 
qq WP, total
Coffee 
production in 
qq WP, plantía
Coffee 
production in 
qq WP, 
commercial 
production
Coffee 
production in 
qq WP, field 
losses
Coffee 
production in 
qq WP, post 
harvest losses
Mean 70.4 27.4 65.5 7.2 7.6
SD 125.0 49.5 117.8 13.4 16.8
Min 0.3 0.1 0.1 0.1 0.1 
Median 34.3 10.0 31.0 2.6 2.6 
Max 1,529.3 300.0 1,376.3 152.9 179.2
Observations 925 177 873 239 179
Notes: See text for conversion procedure. All figures are in qq and húmedo/wet processing level. Sample for coffee consists of 
961 farmers. All production quantities are considered only for farmers having positive plantía and commercial production or field 
or post-harvest losses. Outliers above the 99th percentile in yields were excluded from the production and area averages, and 
categorized as extreme outliers.
Coffee Yields
Agricultural commodities measure of the total output of production of an agricultural commodity 
relative to the amount of land used to produce it. Yield is calculated as the ratio of the total 
production in quintales of wet parchment coffee and the total area planted in hectares (qq WP/ha) 
and is a measure of productivity from that farm or producer. 
26
Table 14 shows some sample statistics of yields; the average productivity is 30.9 qq/ha WP, with 
a median of 21 qq/ha WP. Evidenced by the difference between the mean and the median, it is 
important to analyze the shape of the yield distributions. Figure 2 shows the distribution of the 
yields where we can appreciate that most farmers are below 50 qq/ha WP6
. Note that once we 
divide production by cultivated land, the standard deviation drops below the mean indicating the 
yield distribution is less spread out than the production distribution. The figure shows both the 
household level distribution and the distribution of the yields at the variety level (for households 
with more than one variety). We find lower yields both in commercial area and in plantía; yields in 
plantía are more variable in general. These yields are consistent with averages measured by 
IHCAFE, which estimates 85 qq/ha WP in Comayagua and 61.5 qq/ha WP in El Paraiso, for 
example.
TABLE 14: COFFEE YIELDS 
Total yields (qq/ha WP)
Yields in commercial 
area (qq/ha WP)
Yields in plantía (qq/ha 
WP)
Mean 30.9 36.8 23.0 
SD 30.1 36.7 26.6 
Min 0.3 0.4 0.2 
Median 21.0 23.9 12.9 
Max 157.2 201.6 123.1 
Observations 925 861 177 
Notes: See text for conversion procedure. All figures are in qq and húmedo/wet processing level. Sample for coffee consists of 
961 farmers. All production quantities are considered only for farmers having either positive plantía and commercial production or 
field/post-harvest losses. Outliers above the 99th percentile in yields were excluded from the production and area averages, and 
categorized as extreme outliers.
 
6 The last bars in the figure represent the Winsorized values. Winsorizing consists in decreasing the effect 
of outliers by smoothing the tails of the distribution by placing values above a value x at x or on the values 
below x sequentially to maintain the shape of the distribution.
27
FIGURE 2 DISTRIBUTION OF COFFEE PRODUCTIVITY
To check whether the yields are reasonably sensible, we inspect the production values against 
area in 
Figure 3. If all yields were the same, we would see the production increase linearly with area with 
a strong correlation. However, we expect some dispersion from a line, based on differences in 
productivity of farmers and land, shocks, or other factors. We find a marked dispersion in total 
productivity, perhaps more than one would expect; however, commercial production appears to 
have a more linear pattern. In general, the yields look reasonable for self-reports.
We observe some bunching around what heuristic estimates of plot sizes (e.g. around 1.4, 2 or 
2.8 hectares, equivalent to reports of 2, 3 or 4 manzanas in the survey). Farmers are estimating 
the amount of land they have rather than knowing precise amounts, which will affect the 
denominator in the yields calculation. Abate et al. (2018), highlights that inaccurate estimates of 
land area can affect the estimated effects of extension programs on yields, so we will consider 
whether an alternative method of measuring land area is needed for the endline. Otherwise, the 
pattern is consistent with expectations.
28
FIGURE 3: COFFEE AREA AND COFFEE PRODUCTION, TOTAL AND FOR COMMERCIAL PRODUCTION
29
Coffee and Markets: Buyers, Sales and Prices
We next look at some markets variables to better understand the agents that interact in the coffee 
markets represented in the sample. Table 15 shows the distribution mean of quantity sold to 
different type of coffee buyers. The principal message of this table is that most farmers sell to 
intermediaries and not exporters and in wet parchment and not green/oro or uva/berry for the 
most part. This indicator will be important in the future to gauge how successful was MAS+ linking 
farmers to exporters. Quantities sold are presented in metric tons (MT or 1,000 kgs.) and the 
prices are in USD per MT in each processing level. The aggregate or total price is calculated as 
the value of sales over the quantity sold, after converting to wet parchment units. On average 
farmers sold 4 MT WP to intermediaries and 7.8 MT WP to exporters, but there are few farmers 
in this group, indicating that big farmers are the few selling to exporters7
.
TABLE 15 COFFEE BUYERS AND QUANTITY SOLD
Who did you sell your coffee to?
Quantity of 
coffee sold 
in Berries/
Uva, MT WP
Quantity of 
coffee sold in 
Wet/Húmedo,
MT WP
Quantity of 
coffee sold 
in Dry/Seco,
MT WP
Quantity of 
coffee sold 
in
Green/Oro,
MT WP
Quantity of 
coffee sold 
in Total MT
WP
Intermediary or 
Coyote
Mean 2.6 3.1 3.7 10.0 4.0
SD 8.5 9.2 8.3 27.9 13.3
Observations 96 581 40 104 809
Exporter
Mean 0.7 3.4 8.3 5.3 7.8
SD . 3.0 14.5 6.3 10.4
Observations 1 9 7 7 16
Other Buyers
Mean 3.4 2.5 3.0 6.7 4.3
SD 1.5 2.9 3.4 13.0 8.2
Observations 4 48 8 24 73
We expect that after standardizing prices, coffee that has undergone more processing will sell for 
higher amounts.
8
In Table 16 we estimate the price that coffee farmers are receiving for each level 
of processing, and confirm that the price increases by processing level, reflecting the value added. 
 
7 We note that the average quantity sold, approximately 4 MT, is above the 3.2 MT in the production data 
after converting the figures to WP units. This is due to the farmers that sell their coffee green and the 
conversion factors used. The figures could suggest that farmers are trading coffee but we doubt this is the 
case. In the endline, we will include questions to explore if farmers are also acting as intermediaries of 
others or aggregating their coffee. Further note that the average for all other types of processing 
(none/berry, wet, and dry) are consistent with the production values (around 3 MT).
8 The prices (calculated) presented are obtained by dividing the total sales by the quantity sold; as opposed 
to farmers reporting the price per ton or per qq they received. The distribution of reported vs. calculated 
prices are similar and we presented the calculated figures to directly link the sales values and quantities 
presented before.
30
Coffee sold through intermediaries in uva/berry is 332.60 USD per ton; coffee in wet parchment
is sold at 1,322 USD per ton. In contrast, if sold in green/oro, the price per ton is 1,743.80 USD 
per ton. We note that the number of observations in some categories is small, since not all farmers 
sell to different buyers; thus, the price estimate for those categories is less precise, evidenced by 
the large standard deviation (SD) compared to the estimated average price. Converting all sale 
quantities to wet parchment unit to obtain an average price per wet parchment that included the 
conversion factors and price differentials across processing levels, the average price for coffee 
sold to intermediaries is 1,143.60 USD per ton9
.
These estimated prices are interesting, in that farmers are receiving very different prices even 
within a specific processing level. We present Figure 4 with the calculated prices in USD per MT
from the transactions reported by farmers in the 12 months before the survey. The figure shows 
that there is wide variation in prices depending on the state in which coffee was sold, and 
moreover that the distribution of the aggregate price in the sample is bi-modal; the first mode 
representing the prices of coffee sold in uva/berry and the second mode representing coffee sold 
in húmedo/wet. This indicates that there are two different prices and that the price that a farmer 
obtains depends on some other attributes of their production. Note that this also the case for those 
that sold as wet parchment, thus these differences not driven by the conversions due to humidity 
and density. 
TABLE 16 COFFEE BUYERS AND SALE PRICES
Who did you sell your coffee 
to?
Price of coffee 
sold in Berry/
Uva, per MT
Price of coffee 
sold in 
Wet/Húmedo,
per MT
Price of coffee 
sold in
Dry/Seco, per 
MT
Price of coffee 
sold in 
Green/Oro, per 
MT
Price of 
coffee sold 
in Total per 
MT WP
Intermediary 
or Coyote
Mean 332.6 1,322.9 1,437.9 1,743.8 1,143.6
SD 234.2 429.5 808.6 1,853.5 499.3
Observations 95 564 40 98 787
Exporter
Mean . 1,381.8 1,718.9 2,014.1 1,031.6
SD . 386.3 515.5 1,116.1 398.2
Observations - 9 6 7 16
Other Buyers
Mean 212.0 1,404.2 1,981.2 3,461.2 1,196.4
SD 27.7 470.3 698.3 4,742.1 536.1
Observations 4 48 8 23 72
Note: Prices are winsorized at the 2nd and 98
th percentile to decrease the influence of extreme outliers. Prices are calculated as 
the ratio of total sales and total quantity sold in each processing level.
 
9 This price is a weighted average of the prices received in each processing level, where the weights are a 
function of the proportion sold in each processing level and of the conversion factor to WP from each 
specific processing level. In the figures, this price is identified as “Total-Aggregate WP.”
31
FIGURE 4 DISTRIBUTION OF PRICES OF COFFEE: CALCULATED BY PROCESSING LEVEL
Most coffee farmers reported sales in the last 12 months before the survey. In Table 17 we 
estimate the average value of total sales is 3,220 USD per year. Most sales are as wet parchment, 
and the median farmer sold 1,500 USD in the previous 12 months.
Comparing production to sales, median production is 1.6 MT per year and the median quantity 
sold is 1.4 MT, for a 12.5 percent loss at the median (Table 18). Figure 5 shows the distribution of 
sales, with the right side of the tail winsorized to better see the spread. 
TABLE 17: COFFEE SALES: VALUE
Sale value 
coffee Berry/Uva 
(USD)
Sale value 
coffee WP 
(USD)
Sale value 
coffee Dry/Seco 
(USD)
Sale value 
coffee
Green/Oro 
(USD)
Total Sale value 
(USD)
Mean 1,520.7 2,819.5 3,009.2 5,839.8 3,220.5 
SD 4,071.5 3,659.5 6,519.4 14,671.4 6,886.5 
Min 24.6 0.9 46.7 0.4 0.4 
Median 476.6 1,640.0 1,102.7 1,482.7 1,499.8 
Max 36,080.0 28,103.0 42,213.6 123,410.0 123,410.0 
Observations 99 621 54 128 873 
Notes: Sample for coffee consists in total of 961 farmers. In this table, only farmers selling positive quantities are included. 39 farmers 
made no coffee sales, and 74 farmers reported no monetary revenue from coffee sales.
32
TABLE 18 COFFEE SALES: QUANTITY BY PROCESSING LEVEL
Quantity 
of coffee 
sold in
Berry/
Uva, MT
WP
Quantity of 
coffee sold in 
Wet/Húmedo,
MT WP
Quantity of 
coffee sold 
in Dry/Seco,
MT WP
Quantity of 
coffee sold 
in 
Green/Oro,
MT WP
Quantity of 
coffee sold 
in Total MT
WP
Mean 2.6 3.0 4.2 9.2 4.1
SD 8.3 8.8 8.8 25.1 13.0
Min 0.0 0.0 0.1 0.0 0.0
Median 0.8 1.4 1.4 2.7 1.4
Max 78.3 147.3 50.6 216.8 216.8
Observations 101 638 55 135 898
Notes: Sample for coffee consists in total of 961 farmers. In this table, only farmers selling positive quantities are included. 39 
farmers made no coffee sales, and 74 farmers reported no monetary revenue from coffee sales.
FIGURE 5 DISTRIBUTION OF COFFEE SALES
Evaluation Validity Tests at Baseline: Agricultural Productivity for Coffee
In what follows we provide balance tests for the treatment and control groups to validate the 
impact evaluation design. By balance tests, we mean that we are testing for statistically significant 
differences in the average values for outcomes we plan to study in later survey rounds. In the 
33
coffee sample, we expect that the unconditional variables should be balanced; e.g. the null 
hypothesis that the mean for the treatment and control groups are equal should be accepted. 
From a statistical perspective, we should expect a few statistically significant differences. Recall 
that if we test 20 independent null hypotheses at the 5 percent level, in expectation we should 
reject one hypothesis out of chance. So, if a handful of rejections occur, it does not necessarily 
mean that there is a lack of balance; rather, if we find a pattern of rejections, we need to develop 
a strategy to overcome that pattern. 
First, we examine the variables measuring the coffee area (Table 19). On the left side, we show 
the number of observations and clusters for the control group followed by the mean and the 
standard error of the mean. On the right side, we show the same information for the treatment 
group, and finally we provide a p value for the t-test that the two means are equivalent. Note that 
where we presented conditional averages in the previous section, here we provide unconditional 
averages, since in a randomized control trial we can best identify the intent-to-treat effect. 
Whereas we find that the point estimates for each type of area under coffee cultivation are slightly 
larger in the treatment group, the averages are sufficiently similar that none of the differences are 
statistically different. To examine if the balance is similar across the distribution and not just the 
means, we present Figure 6 with the distribution of both area planted and commercial area planted 
for treatment and control groups. The distributions are very similar, with treatment and control 
lines intertwine throughout the range of the distribution.
TABLE 19: BALANCE TEST FOR COFFEE AREA SIZE
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Total coffee area (in ha) 473 2.12 488 2.46 0.23
[39] [0.17] [38] [0.23]
Coffee area in plantía (in ha) 472 0.50 487 0.67 0.13
[39] [0.05] [38] [0.10]
Coffee area in commercial production (in ha) 473 1.61 488 1.83 0.34
[39] [0.15] [38] [0.18]
Coffee area in recepa (in ha) 473 0.12 487 0.16 0.40
[39] [0.02] [38] [0.04]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
34
FIGURE 6 AREA PLANTED DISTRIBUTION ACROSS TREATMENT ASSIGNMENT
Second, we examine average production levels and losses across the treatment and control 
groups (Table 20). We find averages across the two groups that are broadly similar; the smallest 
p value is in fact 0.24, far from significance at the 5 or 10 percent level. Potentially interesting is 
that with the unconditional figures, we can now clearly observe that post-harvest losses are below 
2 percent on average at the farm level.
TABLE 20: BALANCE TEST FOR COFFEE PRODUCTION
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p￾value
Total Production in qq WP 473 64.40 488 71.39 0.60
[39] [8.51] [38] [10.22]
Total Production in qq WP, Total (exclude 
loss) 473 61.26 488 68.11 0.60
[39] [8.24] [38] [10.13]
Total Production in qq WP, Plantia 473 5.63 488 4.57 0.51
[39] [1.27] [38] [1.01]
Total Production in qq WP, Commercial 473 55.63 488 63.54 0.53
[39] [7.81] [38] [9.73]
Total Production in qq WP, Field loss 473 2.03 488 1.56 0.36
[39] [0.47] [38] [0.27]
Total Production in qq WP, Post-harvest loss 473 1.11 488 1.72 0.25
[39] [0.22] [38] [0.48]
35
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
Since neither area cultivated nor coffee produced was statistically different between the treatment 
and control groups, it follows that the yields should not be statistically different from one another 
either (Table 21). Indeed, we test for differences in yields for total area planted, the area in 
commercial production, and the area in plantía, and we find no significant differences. Point 
estimates for the control group are slightly higher than the treatment groups, but as we have 
discussed this difference is not statistically meaningful. Figure 7 shows that the yield estimates 
are balanced across the distribution.
TABLE 21: BALANCE TEST FOR COFFEE YIELDS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Yields qq/ha WP, Total area 453 30.73 475 31.16 0.93
[39] [3.31] [38] [3.38]
Yields qq/ha WP, Commercial 422 36.41 442 37.10 0.91
[39] [3.94] [38] [4.29]
Yields qq/ha WP, Plantía 96 21.15 82 24.92 0.47
[34] [3.44] [34] [3.83]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
36
FIGURE 7 YIELD DISTRIBUTION ACROSS TREATMENT ASSIGNMENT
To complete this line of thought, we examine the amount of coffee that was sold, as well as the 
value of those sales (Table 22). The quantity and value of coffee sold in berry form is marginally 
significant (at the 10 percent level), indicating that farmers in the treatment group are receiving 
higher prices. As mentioned, we expect some differences and this difference is not statistically 
different across groups at conventional significance levels. Figure 8 shows that the quantity sold 
is balanced across the distribution. Finally, Table 23 shows that the proportion of farmers that sell 
to different buyers is balanced, including exporters.
37
TABLE 22: BALANCE TEST FOR COFFEE SOLD AND VALUE OF SALES
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Quantity sold in berry in MT (1000 kgs)- WP units 455 0.11 473 0.46 0.09*
[39] [0.04] [38] [0.21]
Quantity sold in wet in MT (1000 kgs)- WP units 455 4.15 471 2.42 0.48
[39] [2.40] [38] [0.44]
Quantity sold in dry in MT (1000 kgs)- WP units 455 0.25 473 0.25 0.99
[39] [0.12] [38] [0.13]
Quantity sold in green/oro in MT (1000 kgs)- WP units 455 1.79 473 0.97 0.26
[39] [0.64] [38] [0.34]
Quantity sold total in MT(1000 kgs)- WP units 455 6.29 471 4.11 0.39
[39] [2.44] [38] [0.65]
Sale value berry (USD) 455 71.73 473 253.77 0.10*
[39] [21.78] [38] [107.41]
Sale value wet parchment (USD) 455 1746.33 471 2064.19 0.35
[39] [186.53] [38] [282.70]
Sale value dry parchment (USD) 455 151.17 473 201.43 0.71
[39] [57.75] [38] [123.13]
Sale value green/oro (USD) 455 708.23 473 920.03 0.60
[39] [203.84] [38] [350.52]
Total Sale value (USD) 455 2677.46 471 3445.27 0.18
[39] [284.82] [38] [489.86]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
38
FIGURE 8 QUANTITY SOLD ACROSS TREATMENT ASSIGNMENT
TABLE 23: BALANCE TEST FOR QUANTITY SOLD BY BUYERS OF COFFEE
(1) (2) t-test
Control Treatment p-value
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE (1)-(2)
Sold to intermediary or coyote 455 0.92 473 0.88 0.21
[39] [0.02] [38] [0.03]
Sold to exporter 455 0.01 473 0.03 0.25
[39] [0.01] [38] [0.01]
Sold to other buyers 455 0.06 473 0.09 0.32
[39] [0.02] [38] [0.02]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
39
Inputs, Technologies and Management Practices
We seek to track progress in the introduction of improved technologies and management 
practices. We consider two concepts—application and adoption. Application is the use of 
technology or management practice by a farmer or other producer over at least one production 
period. Adoption is the use of technology or management practice by a farmer in a sustainable 
way over an extended period of time. 
The fact that farmers or other beneficiaries have applied a technology or management practice 
for a year or two does not mean that they have sustainably adopted it – or will continue to do so 
after a project ends. In addition, improved technology and management practices are often 
promoted as packages comprising several independent technologies or practices. For example, 
“conservation agriculture” is often promoted as an improved technology/management practice 
package and may include any combination of several independent elements (e.g., zero-tillage, 
use of cover crops, integrating livestock, direct seeding), each of which can lead to improved 
production outcomes but are more effective when applied together. In the survey we track 
independent elements of a technology or practices to allow identification of barriers to application 
of some technologies relative to others. Finally, note that the application or adoption of many of 
these techniques will be monitored among beneficiaries. The goal here is to measure the 
application of those techniques relative to the control group.
Also, important to growing coffee are the types of inputs that are used by farmers. Note that inputs 
may be important to yield, but not necessarily to profits; farmers should ideally be optimizing their 
profitability rather than yields. We first report the proportion of farmers that applied different types 
of inputs (Table 24). We find that the largest proportion of farmers used chemical fertilizers; a 
smaller proportion also used either foliar fertilizer or compost. A substantial share of farmers also 
used herbicides (36 percent); fewer famers used fungicides or other inputs. Note that farmers 
might only use these chemicals if they were affected by some fungus or pest
TABLE 24: SHARE OF FARMERS THAT APPLIED COFFEE INPUTS
Compost
/ Organic 
fertilizer
Chemical 
fertilizer
Foliar 
fertilizer Fungicide Herbicide Parasi
toids
Other 
inputs
Share of farmers that 
applied 0.15 0.63 0.27 0.19 0.36 0.01 0.02
SD 0.36 0.48 0.44 0.40 0.48 0.10 0.14
Observations 961 961 961 961 961 961 961
Notes: Sample for coffee consists of 961 farmers. Farmers that did not respond to the inputs module because they reported 
not having used any input have also been included in this table as farmers that did not apply any inputs.
The total cost of inputs used by farmers appears to be substantial (Table 25). Conditioning on any 
positive spending, we find the total cost of inputs other than labor is 373 USD lempiras on average; 
labor costs for households that use hired labor is over 3,000 USD. These figures suggest that 
coffee production, in the way that coffee is produced, is either not very profitable or farmers are 
overestimating their labor costs. 
40
TABLE 25: COST OF INPUTS AND LABOR COSTS
Total cost 
of inputs
Annual cost of male 
labor
Annual cost of female 
labor
Total annual labor 
cost
Annual cost in USD 373 2487 1168 3163
SD 755 3966 2302 5111
Observations 775 782 515 805
Notes: Only farmers spending positive monetary value on inputs are included in this table.
Table 26 shows practices across the production cycle. Thirty percent of farmers perform tissue 
management and 21 percent dry their coffee. The tissue management practice that farmers 
mention is sanitary pruning, with 70 percent implementing it (Table 27). In other words, among all 
farmers doing any tissue management, 70 percent do sanitary pruning; so only 21 percent of the 
full sample is doing sanitary pruning. Most farmers report harvesting when the coffee berries are 
in a uniform maturity level and sell their coffee cleaned but not dried or in wet parchment. Also 
70% of farmers that have their coffee in shade have it permanent shade (Table 28).
TABLE 26: COFFEE PRACTICES
During last 12 months, Mean SD Observations
did you clean, control weeds? 0.98 0.13 961
did prune plantations, tissue management? 0.30 0.46 959
did you do pruning of excess shoots? 0.30 0.46 958
did you do soil fertilization? 0.64 0.48 959
did you do foliar fertilization? 0.24 0.42 957
did you do shade regulation? 0.48 0.50 957
did you do pest management? 0.21 0.41 960
did you do disease control? 0.17 0.38 960
did you cut coffee? 0.94 0.24 960
did you de-pulp coffee? 0.82 0.38 961
did you wash and ferment coffee? 0.81 0.39 960
did you dry coffee? 0.21 0.41 958
did you store coffee? 0.07 0.25 953
Notes: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to 
missing values on specific questions.
41
TABLE 27: COFFEE PRACTICES: PLANT AND TISSUE MANAGEMENT
Mean SD Observations
Did you perform tissue management during the last 12 months? 0.3 0.2 961
For tissue management:
did you do sanitary pruning? 0.7 0.5 255
did you do recepa (0.40m from ground)? 0.2 0.4 255
did you do descope/pruning (1.70m)? 0.2 0.4 255
did you do medium pruning (1.50m )? 0.1 0.2 255
did you do other type of pruning? 0.0 0.2 255
did you do pruning of excess shoots? 0.4 0.5 255
Note: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to missing values to 
specific questions.
TABLE 28: COFFEE PRACTICES: QUALITY
Mean SD Observations
Did you do any shade management practice? 0.5 0.5 960
How many bloomings did you have in 2017? 2.1 1.5 956
Do you monitor bloomings to program harvesting? 1.8 0.4 961
Do you harvest when coffee cherries have same maturity level? 0.9 0.3 959
When you harvest, do you separate the green beans? 0.7 0.4 959
Do you sample your harvest to scrutinize coffee grains? 0.0 0.2 959
% of coffee beans ripe 76.3 20.8 42
% of coffee beans green 13.7 13.8 42
% of coffee beans over ripe 8.8 19.5 42
% of coffee beans dry 6.9 17.0 42
Do you have shade of type: permanent? 0.7 0.5 436
Do you have shade of type: temporal? 0.0 0.2 436
Do you have shade of type: both? 0.3 0.4 436
Do you have shade at a level of: 10-30%? 0.3 0.5 436
Do you have shade at a level of: 31-50%? 0.5 0.5 436
Do you have shade at a level of: more than 50%? 0.2 0.4 436
Is your coffee grain small size? 0.1 0.3 957
Is your coffee grain medium size? 0.5 0.5 957
Note: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to missing values to 
specific questions.
42
TABLE 29: COFFEE PRACTICES: SOIL CARE
Mean SD Observations
Did you perform a soil analysis in any of your plots? 0.11 0.31 954
Did you add lime to soil? 0.16 0.36 961
Did you perform a water analysis? 0.01 0.10 958
How much space in cm do you leave between coffee plants? 111.47 78.72 955
How much space in cm do you leave between grooves? 174.08 135.54 955
Note: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to missing values 
to specific questions.
Only 11% of farmers perform a soil analysis and only 1% of farmers performed a water analysis, 
showing these practices are still low in the area (Table 29). Seventy percent of coffee farmers 
own a de-pulping machine, but only 35% report doing pulp managing; the majority of farmers 
compost it. To sell their “beneficiado húmedo” or wet processing, half the farmers use water from 
a source such as river, creek, canal, lake; showing that practices for water collection are not used 
extensively. Eighty seven percent of farmers sell their coffee humid and the same share of farmers 
de-pulp the coffee before selling (Table 30). 
Only about 30 farmers in the coffee sample report having received training in coffee quality, and 
half of them have tested for quality at baseline level. Most farmers (86 percent) report that their 
coffee grain is white, followed by 30 percent that also report it being brown and 25 percent dark 
brown. Given the small number of farmers that performed quality tests at baseline (15), sample 
sizes on quality are still very small to draw conclusions on both the quality results they got from 
tests and the quality classification. Although anecdotally, 9 out of those 15 farmers don´t know or 
were not informed about the quality classification.
43
TABLE 30: COFFEE PRACTICES: POST-HARVEST
Mean SD Observations
Do you have a cherry coffee receipt hopper? 0.49 0.50 961
Do you have your own de-pulping machine? 0.69 0.46 835
Did you manage coffee pulp during 2017? 0.35 0.48 835
What do you do with the coffee pulp? Compost 0.54 0.50 294
What does he do with the coffee pulp? Leave if on the farm 0.33 0.47 294
What do you do with the coffee pulp: Accumulate it 0.17 0.38 294
Source of water for wet processing: Rainwater / water harvest 0.06 0.25 835
Source of water for wet processing: River, creek, canal, lake, dam 0.50 0.50 835
Source of water for wet processing: Private well 0.08 0.27 835
Source of water for wet processing: Public well 0.01 0.08 835
Source of water for wet processing: Spring 0.20 0.40 835
Did you perform any water management during 2017? 0.20 0.40 835
What do you do with the honey waters? Compost 0.08 0.27 168
What do you do with the honey waters? Pour them on the farm 0.18 0.39 168
What do you do with the honey waters? Pour them into fountains / streams 0.10 0.30 168
What do you do with the honey waters? Pour them in oxidation lagoon 0.35 0.48 168
What do you do with the honey waters? Pour them on the street, ditch 0.21 0.41 168
Do you sell humid? 0.87 0.34 835
Do you dry the coffee bean? 0.92 0.28 108
Do you store coffee? 0.36 0.48 99
Do you sell your coffee immediately? 1.31 0.47 99
Do you transport your harvested coffee in your own vehicle? 2.13 0.76 31
Type of tolva: In water 0.33 0.47 475
Type of tolva: Dry 0.67 0.47 475
Do you de-pulp coffee? 0.87 0.34 960
Do you have shade at a level of: more than 50%? 0.36 0.48 835
Is your coffee grain small size? 0.14 0.35 835
Is your coffee grain medium size? 0.25 0.43 835
Note: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to missing values to specific 
questions.
44
TABLE 31: COFFEE PRACTICES: MORE ON QUALITY
Mean SD Observations
Have you received training on coffee quality tests? 0.30 0.46 99
Did any of the lots of coffee you produce has been tested for quality? 0.50 0.51 30
What was the color of the coffee grain: white 0.86 0.34 835
What was the color of the coffee grain: light yellow 0.09 0.28 835
What was the color of the coffee grain: green yellow 0.02 0.13 835
What was the color of the ferment: brown 0.30 0.46 835
What was the color of the ferment: dark brown 0.25 0.43 835
What was the color of the ferment: black 0.05 0.22 835
What result did you get? from 80 to 83 points 0.47 0.52 15
What result did you get? more than 83 points 0.40 0.51 15
What was the classification it got: standard 0.07 0.26 15
What was the classification it got: High altitude coffee HG 0.27 0.46 15
What was the classification it got: special coffee/gourmet 0.07 0.26 15
What was the classification it got: you were not informed 0.60 0.51 15
Note: These questions were asked of the entire sample of 961 coffee farmers; differences in sample sizes due to missing values 
to specific questions.
Coffee Extension
Coffee farmers report in general very little exposure to extension (Table 28). The survey form 
asked specifically about exposure to 10 different topics related to extension, and only 182 out of 
961 farmers, or about 19 percent of farmers, report receiving extension on any of the 10 topics. 
Among farmers receiving some coffee extension, the two most common topics are coffee pruning 
and crop maintenance; very few farmers have received any extension on marketing, finance and 
contracts, or market access. Among farmers reporting receiving extension, they tend to rate the 
extension as very good. Since a main goal of MAS+ is to increase productivity through extension, 
the finding of little exposure is quite positive if it will lead to stronger impacts.
45
TABLE 32: COFFEE EXTENSION
Mean SD Observati
ons
Number of topics of received extension 0.35 0.93 961
Share of farmers that received extension about:
Coffee Pruning 0.09 0.29 961
Shade regulation 0.06 0.24 961
Maintenance 0.10 0.30 961
Coffee cut 0.03 0.17 961
Post-harvest 0.01 0.12 961
Quality and certification 0.02 0.14 961
Marketing 0.02 0.14 961
Finance and contracts 0.01 0.07 961
Market access 0.00 0.06 961
Duration of coffee extension in days 55.1 254.23 182
Quality of extension, 1 to 4 3.57 0.56 175
Notes: All coffee farmers have been considered for this table. All farmers who claimed not having received extension in any topic, 
and so they were not asked these questions directly, are coded as “no” for each topic. Sample for duration of training and quality 
of extension include only farmers that received extension.
Coffee Pests
We next examine damages that farmers report being caused by pests, as one specific type of 
negative shock affecting coffee farmers (Table 33). Farmers claim substantial losses to pests 
when they claim losses; on average 24 percent of the crop was loss when they had a loss. Among 
the more common insects doing damage to crops were coffee rust and broca. Among farmers 
reporting that they incurred in some cost to manage a pest, the average expenditure was 5,478 
lempiras; note that not all farmers attempted to solve pest problems with additional inputs.
46
TABLE 33: PESTS IN COFFEE
Mean SD Observations
Percent of damage done by pests 23.97 22.94 628
Share of farmers that had problems by:
Broca 0.40 0.49 961
Coffee rust 0.40 0.49 961
Minador 0.02 0.16 961
Ojo de gallo 0.15 0.36 961
Mal de hilachas 0.05 0.22 961
Antracnosis 0.03 0.17 961
Mancha de hierro 0.10 0.29 961
Other 0.03 0.17 961
Average cost to control pest 5,478.17 16,947.47 246
Notes: Information on having encountered each specific pest includes all farmers. Sample sizes for percent of crop damage caused 
by pests and control costs includes only farmers that had positive values for these two variables.
Investments and Equipment
Next, we examine what investments and equipment the farmers in the sample have. The most 
reported investment are plant nurseries, with 18 percent of coffee farmers having them. The 
proportion of farmers that have post-harvest equipment for coffee is low, with only 7 percent 
reporting having a pulping machine and 2 percent having a designated place to dry their coffee 
(Table 34).
TABLE 34: INVESTMENTS IN THE AGRICULTURAL PLOTS 
Investment in the last 12 months: Mean SD Observations
Live barriers 0.04 0.19 952
Dead barriers, fences 0.07 0.26 956
Water harvests 0.00 0.06 954
Alley cultivation 0.02 0.13 956
Plant nursery 0.18 0.39 957
Greenhouses 0.00 0.05 956
Irrigation canal 0.00 0.00 953
Irrigation system 0.01 0.10 953
Solar dryer 0.02 0.14 953
Pulping machine 0.07 0.26 955
Other investments 0.01 0.07 944
In the survey we asked 19 different types of durable goods that could either be associated with 
higher incomes or production (Table 35). We find that the most important are primarily consumer 
durables, such as televisions, radios, motorcycles, and cell phones (though the latter could also 
47
be considered useful for production). Equipment associated with agriculture, such as sprayers, 
are not own by many farmers in the sample.
TABLE 35: INVESTMENT IN EQUIPMENT
Type of machinery and equipment: Mean SD Observations
Sprayer with motor 0.04 0.18 961
Motor mounter 0.03 0.16 961
Tractor 0.00 0.00 961
Car 0.02 0.15 961
Truck or pick-up 0.17 0.38 961
Motorcycle 0.33 0.47 961
Bicycle 0.08 0.27 961
Radio 0.64 0.48 961
Television 0.47 0.50 961
Refrigerator 0.30 0.46 961
Computer 0.06 0.23 961
Cell phones 0.85 0.35 961
Gas stove 0.09 0.28 961
Livestock (oxen, cattle) 0.17 0.37 961
Small livestock (goats, pigs, sheep) 0.21 0.41 961
Hens, Ducks, Turkeys, Pigeons 0.81 0.39 961
Horses, mares, mules, males, donkeys 0.35 0.48 961
Group Membership and Participation
Group membership can play a role in fostering improvements in agricultural production, though it 
is not sufficient (e.g. Waddington et al., 2014). Learning about the extent of these organizations 
is important, especially given the focus of MAS+ in fostering group participation. We therefore ask 
farmers if they participate in any of such organizations at baseline (Table 36). To be counted as 
participants, they can be either active or inactive organization members, or leading such 
organizations. We find that the most common type of group are water user associations (with 20% 
participation), followed by civic or charitable groups (with similar participation). Productive groups 
are rarer; only about a quarter of households report the existence of credit or microfinance groups, 
and only 17 percent report the existence of producer groups. Other organizations have close to 
zero participation. Therefore, there is a lot of scope for organizing coffee farmers within the 
sample.
48
TABLE 36: GROUPS PRESENT IN THE COMMUNITY
Mean SD Observations
Share of 
farmers that 
are members
Exists in the community: Productive agricultural 
groups or org. of producers 0.17 0.38 946 0.13
Exists in the community: Productive agricultural 
groups for commercialization 0.02 0.13 946 0.01
Exists in the community: Water users groups 0.72 0.45 959 0.20
Exists in the community: Civic or charitable groups 0.77 0.42 961 0.19
Exists in the community: Women groups 0.06 0.24 949 0.01
Exists in the community: Credit or micro-finance 
groups 
0.25 0.44 958 0.14
Exists in the community: Insurance or mutual aid 
groups 
0.00 0.00 951 0.00
Exists in the community: Commerce and business 
associations 0.00 0.00 957 0.00
Exists in the community: NGO / External Project 0.01 0.11 956 0.01
Exists in the community: Other 0.01 0.12 938 0.01
Evaluation Validity Tests at Baseline: Inputs, Extension and Pests for Coffee
Our next tests of the validity of the evaluation design focuses on inputs, extension participation 
and the prevalence of pests in the coffee. We study whether the share of households using 
specific types of inputs in coffee differs between the treatment and control group (Table 37). In all 
seven cases, the average proportion of households using specific inputs is not statistically 
different between the treatment and control groups. We additionally examine whether material or 
labor input costs differ by treatment group (Table 38). As with the discrete indicators, we find no 
evidence of statistically significant differences. To this point, as expected through the 
randomization, we have found no differences between the treatment and control groups for any 
indicators related to coffee inputs, outputs, or sales by sample households.
49
TABLE 37: BALANCE TEST FOR SHARE OF FARMERS THAT APPLIED INPUTS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Applied Compost/ Organic fertilizer 473 0.14 488 0.16 0.43
[39] [0.02] [38] [0.02]
Applied Chemical fertilizer 473 0.63 488 0.64 0.89
[39] [0.04] [38] [0.05]
Applied Foliar fertilizer 473 0.25 488 0.28 0.51
[39] [0.03] [38] [0.04]
Applied Fungicides 473 0.17 488 0.22 0.29
[39] [0.03] [38] [0.03]
Herbicides 473 0.38 488 0.34 0.41
[39] [0.03] [38] [0.03]
Parasitoids 473 0.01 488 0.01 0.31
[39] [0.01] [38] [0.00]
Other inputs 473 0.02 488 0.02 0.80
[39] [0.01] [38] [0.01]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
TABLE 38: BALANCE TEST FOR COST OF INPUTS AND LABOR COSTS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Input costs (USD) 387 332.31 399 401.51 0.40
[39] [55.18] [38] [61.44]
Annual labor cost (USD)- men 392 2416.33 421 2369.42 0.91
[39] [285.38] [38] [292.57]
Annual labor cost (USD)- women 392 626.65 421 845.10 0.29
[39] [107.66] [38] [177.64]
Annual labor cost (USD)- total 392 3042.97 421 3214.52 0.76
[39] [349.28] [38] [427.12]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
The next set of variables that relate to one of the research questions involve access to extension 
for growing coffee. We test first for whether the average number of topics covered through 
extension differs by treatment and control groups; it does not (Table 39, row 1). We then test 
whether there are differences in receipt of extension by practice; again, we find no differences. 
We finally examine extension by days and rating; there are no statistical differences between 
treatment and control groups.
50
TABLE 39: BALANCE TEST FOR COFFEE EXTENSION
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Number of topics of received extension 473 0.30 488 0.39 0.34
[39] [0.06] [38] [0.07]
Received extension about Coffee pruning 473 0.08 488 0.10 0.48
[39] [0.02] [38] [0.02]
Received extension about Shade regulation 473 0.05 488 0.06 0.65
[39] [0.01] [38] [0.01]
Received extension about Maintenance 473 0.09 488 0.11 0.32
[39] [0.02] [38] [0.02]
Received extension about Coffee cut 473 0.03 488 0.03 0.84
[39] [0.01] [38] [0.01]
Received extension about Post-harvest 473 0.01 488 0.02 0.09*
[39] [0.00] [38] [0.01]
Received extension about Quality and certification 473 0.02 488 0.02 0.62
[39] [0.01] [38] [0.01]
Received extension about Marketing 473 0.02 488 0.02 0.95
[39] [0.01] [38] [0.01]
Received extension about Finance and contracts 473 0.00 488 0.01 0.26
[39] [0.00] [38] [0.00]
Received extension about Market access 473 0.00 488 0.01 0.32
[39] [0.00] [38] [0.00]
Duration of coffee extension in days 81 25.39 101 78.89 0.25
[28] [7.38] [29] [45.46]
Quality of extension from 1 to 4 77 3.53 98 3.60 0.39
[28] [0.07] [29] [0.04]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
Our final balance tests for coffee relate to experience with pests (Table 40). We test whether 
households have differences in the percent of their crop damaged; we also test whether the cost 
of pest control differs between the treatment and control groups. We find that there are no 
significant differences in the percent of damages or the cost of pest control. However, for one 
specific pest (the minador), we find a significant difference. We believe this difference is not 
pertinent for two reasons. First, it is the only difference we find, out of dozens of tests. We 
expected, in fact, to find more differences. Second, the share of households affected by the 
minador are extremely small, at about 3 percent of all households. Since the share is so small, it 
would not have meaningful effects on production over the entire sample.
51
TABLE 40: BALANCE TEST FOR PESTS IN COFFEE
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Percent of damage 308 24.40 320 23.56 0.78
[39] [2.27] [38] [1.98]
Had problems with broca 473 0.39 488 0.40 0.91
[39] [0.05] [38] [0.03]
Had problems with roya 473 0.41 488 0.39 0.67
[39] [0.05] [38] [0.04]
Had problems with minador 473 0.04 488 0.01 0.01**
[39] [0.01] [38] [0.00]
Had problems with ojo de gallo 473 0.16 488 0.15 0.80
[39] [0.02] [38] [0.03]
Had problems with mal de hilachas 473 0.06 488 0.04 0.33
[39] [0.02] [38] [0.01]
Had problems with antracnosis 473 0.03 488 0.03 0.59
[39] [0.01] [38] [0.01]
Had problems with mancha de hierro 473 0.10 488 0.09 0.74
[39] [0.02] [38] [0.01]
Had problems with other 473 0.03 488 0.03 0.67
[39] [0.01] [38] [0.01]
Cost of pest control in Lps 136 3986.26 110 7322.70 0.11
[38] [1312.67] [30] [1582.25]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
In summary, we find almost no statistically meaningful differences between the treatment and 
control groups for outcomes among the coffee sample, demonstrating that the randomization 
worked quite well. Even without controlling for other variables in analysis, the implication is that 
the difference in means either midway through the project, or at endline between treatment and 
control groups is an unbiased estimate of the program impacts. Controlling for other factors in 
analysis may remain useful, however, to explain some of the variance in the outcome, to improve 
statistical inference on the difference in means.
52
7 Analysis of Beans Producers in MAS+ Evaluation
As with the coffee subsample in the previous section, in this section we describe the sample of 
beans growers. After describing the households demographically, we examine agricultural 
statistics among households from the perspective primarily of growing beans, including both 
outputs and inputs. We include balancing tests throughout the themes in these sections to show
the validity of the propensity score matching evaluation design to verify that the selected sample 
is balanced for important outcomes across treatment and control groups.
7.1 Demographic Characteristics
We first examine demographic characteristics among the beans sample; households are slightly 
larger than among the coffee sample, but the different is not statistically significant (Table 41). 
The households are also largely nuclear, meaning that there are few elderly residents; there are 
0.2 total residents older than 65, and there are on average around 2 children who are under 15 
living in the households. As with the coffee sample, young and elderly household members are 
relatively gender balanced. However, we find that adults of working age are not gender balanced; 
we find more males than females in households, at 1.16 males on average versus 0.87 females. 
TABLE 41: DEMOGRAPHIC CHARACTERISTICS, BEANS SAMPLE
Mean SD Min Median Max Observations
Number of people living in the house 4.84 2.11 1 5 15 968
Boys under 6 0.43 0.70 0 0 5 968
Girls under 6 0.35 0.70 0 0 9 968
Men between 6 and 14 years old 0.54 1.01 0 0 13 968
Women between 6 and 14 years old 0.55 1.25 0 0 15 965
Men between 15 and 64 years old 1.16 0.66 0 1 4 968
Women between 15 and 64 years old 0.87 0.42 0 1 3 968
Men older than 65 0.11 0.31 0 0 1 968
Women older than 65 0.09 0.29 0 0 2 968
People managing the farm in the beans sample range between 11 to 85 years old (Table 42). The 
average age is 41 years old. About 42% of beans farmers are women. The average number of 
years of education is 4.3 years and the average literacy ratio is 82%, just a little below the coffee 
sample. Looking at the same characteristics by gender (Table 43), we see that women are about 
two years older on average than men (42 years old vs 40 years old for men), similar to the coffee 
sample. Average years of education and literacy ratios for both genders are about the same.
53
TABLE 42: DEMOGRAPHIC CHARACTERISTICS
Age Female Years of Education Literacy
Mean 41.00 0.42 4.32 0.82
SD 16.28 0.49 2.98 0.38
Min 11 0 0 0
Median 40 0 5 1
Max 85 1 16 1
Observations 2188 2188 2188 2188
Note: Demographic characteristics are calculated among all persons in charge of agricultural decisions in the 968 beans 
households.
TABLE 43: DEMOGRAPHIC CHARACTERISTICS BY GENDER
Male Female
Age Years of Education Literacy Age Years of Education Literacy
Mean 40.26 4.30 0.81 42.03 4.33 0.83
SD 17.26 3.02 0.39 14.76 2.92 0.38
Min 11 0 0 11 0 0
Median 38 5 1 42 5 1
Max 85 16 1 84 15 1
Observations 1273 1273 1273 915 915 915
Note: Demographic characteristics are calculated among all persons in charge of agricultural decisions in the 968 beans 
households.
We also calculate non-agricultural income for the beans sample. But as opposed to the coffee 
sample, there is a smaller share of beans farmers that sell their product to the market, so we have 
information for 448 farmers that report earning positive income from non-agricultural activities,
and 438 farmers that report earning income from selling beans. Considering the sample of beans
farmers that earn positive amounts of either agricultural or non-agricultural activities, we find that 
bean sales represent 60% of total monthly income (Table 44).
TABLE 44: AGRICULTURAL AND NON-AGRICULTURAL INCOME
Monthly Non-Agricultural 
Income in USD
Monthly Agricultural 
Income in USD
Share that Coffee Income 
represents
from Total Income
Mean 302.51 126.27 0.60
SD 574.16 504.94 0.44
Min 0.21 0.06 0.00
Median 147.60 49.20 1.00
Max 6150.00 8200.00 1.00
Observations 448 438 513
Note: Only 448 farmers from the beans sample report earning income from other non-agricultural activities; 438 farmers report 
earning income from selling beans to the market. The share that beans income represents from total income is calculated only for 
farmers earning positive amounts from either of both sources (513 farmers).
54
7.2 Agricultural Outcomes
We next explore bean production, first from the perspective of cultivated area, and then examine 
production and yields. Before turning to agricultural inputs, the section also discusses beans 
sales. Note that we expect different outcomes with sales, as beans are not necessarily a cash 
crop.
Size of Cultivated Land
There are two main seasons for growing field crops in Honduras, and most of Central America—
the primera (April-July) and postrera (August-October) seasons. Decisions about which season 
to plant in depend upon several factors, such as temperature, rainfall, and soil drainage 
capabilities, among others. Since beans can be grown in either season, we split the sample to 
look at area planted and harvested by season (Table 45). We find that conditional on planting, 
households tend to plant about 1 ha in beans, regardless of the season, though more households 
plant in postrera than in primera. The distributions are not as skewed as with coffee; the largest 
areas planted and harvested are 10.4 ha and 12.5 ha in the primera and postrera, respectively.
TABLE 45: AREA SIZE OF PLANTED AND HARVESTED AREA WITH BEANS
Total area planted 
with beans in 
Primera (ha)
Total area harvested 
with beans in 
Primera (ha)
Total area 
planted with 
beans in 
Postrera (ha)
Total area 
harvested with 
beans in 
Postrera (ha)
Mean 1.00 0.93 1.04 0.97
SD 1.01 0.93 1.18 1.09
Min 0.04 0.01 0.01 0.01
Median 0.70 0.70 0.70 0.70
Max 10.44 10.44 12.53 12.53
Observations 657 656 766 747
Notes: Sample for beans consists of 968 farmers. Difference in sample sizes is due to farmers that only plant in primera or 
postrera. 683 farmers report planted areas in Primera, and 795 farmers in postrera. 
Outliers were dealt with in the following ways: If commercial area reported was greater than total area, it was replaced with total 
area. We replaced total area by commercial area. 14 cases in primera, 23 in postrera. 28 extra observations were dropped from 
total area and 16 from harvested area because they are yield outliers or have missing values in the primera; similarly, we 
removed 49 for total area and 23 for harvested area in the postrera.
55
Beans Production 
Next, we examine the total amount of beans produced by season (Table 46). We find that among 
producing households, beans production is a bit higher in the primera than the postrera season, 
at about 863 kg on average versus 646 kg in the latter season. As with area cultivated, we find 
that the distribution is somewhat skewed; the medians are smaller than the means, at 544 and 
453 kg for the primera and postrera, respectively.
TABLE 46: BEANS PRODUCTION
Beans production in 
Primera (in kg)
Beans production in 
Postrera (in kg)
Mean 862.56 645.80
SD 943.82 761.87
Min 0.191 0.340
Median 544.31 453.59
Max 8890.36 6667.77
Observations 657 739
Notes: Only farmers with positive production in either primera or postrera have been considered for this table; 14 observations 
were dropped from production in primera and 25 from production in postrera because they contained yield outliers or missing 
observations.
Given that beans are not necessarily a cash crop and can constitute an important share of home 
consumption, and as such, have an important role in food security, we look at how farmers report 
using their total production. Table 47 and Table 48 illustrate how the beans production was 
distributed for primera and postrera, respectively. The possible uses included in the survey were: 
sale, consume in the home, make sub-products, animal feed, and losses. We find that most 
farmers use their beans production for home consumption and sales.
TABLE 47: BEANS PRODUCTION FINAL USE, PRIMERA
Sales
kg- primera
Home 
consumption
kg - primera
Exchange
kg -
primera
Sub 
products
kg - primera
Animal feed
kg - primera
Gifts
kg - primera
Lost
kg -primera 
Mean 106.14 140.03 162.19 90.72 110.56 85.02 259.03
SD 303.32 191.81 402.31 45.36 90.26 212.07 215.30
Min 0.91 0.23 18.14 45.36 22.68 1.36 0.91
Median 45.36 90.72 90.72 90.72 68.04 45.36 181.44
Max 4535.90 3039.05 1814.36 136.08 272.15 2177.23 907.18
Observations 537 633 19 3 8 195 55
Notes: Sample for beans consists of 968 farmers. Difference in sample sizes is due to farmers that only plant in primera or postrera. 657
farmers had planted area in Primera, and 739 farmers in postrera. 
56
TABLE 48: BEANS PRODUCTION FINAL USE, POSTRERA
Sales
kg- postrera
Home 
consumption
kg - postrera
Exchange
kg - postrera
Sub 
products
kg - postrera
Animal feed
kg - postrera
Gifts
kg - postrera
Lost
kg - postrera 
Mean 87.74 144.44 115.30 60.48 101.76 57.18 310.86
SD 192.07 218.65 104.32 26.19 103.17 96.99 288.05
Min 0.34 6.80 14.51 45.36 22.68 2.27 18.14
Median 45.36 90.72 90.72 45.36 90.72 45.36 226.80
Max 2902.98 3655.94 362.87 90.72 453.59 1088.62 1814.36
Observations 539 708 16 3 15 144 60
Notes: Sample for beans consists of 968 farmers. Difference in sample sizes is due to farmers that only plant in primera or postrera. 657 farmers had 
planted area in Primera, and 739 farmers in postrera. 
Beans Food Security
To investigate whether a household’s consumption meets minimum requirements, we use the 
Standard of Per Capita Consumption of Basic Grains for countries affected by Mitch hurricane, 
which for Honduras is estimated in 25.7 kg per capita / year10. We find that 48% of households 
have a yearly per capita consumption that is greater than the recommended threshold (Table 49). 
The average consumption in the area is estimated to be greater than the standard, at 35.6 kg per 
capita a year. 
TABLE 49: BEANS FOOD SECURITY
Beans Consumption (KG/PC/year)
% of HH above the Standard for 
Per Capita Consumption in 
Honduras (25.7)
Mean 35.56 0.48
SD 44.57 0.50
Min 0.57 0.00
P50 22.68 0.00
Max 731.19 1.00
Observations 749 749
Notes: These statistics are produced only for the sample of farmers that reported having home consumption.
If farmers only produced in primera and not in postrera, only their consumption in primera was considered; 
and vice-versa
Beans Yields
Combining the production and land plated information, we then measure yields for beans, by 
season (Table 50). Yields are higher in the primera on average than the postrera among sample 
households at 888 and 670 kg/ha, respectively. Median yields are slightly lower, at 782 and 586 
kg/ha. In general, these averages are in line with expectations.
 
10 http://www.fao.org/docrep/003/y2784s/Y2784S07.htm
57
Figure 9 shows the distribution of the yields for primera and postrera. Most farmers are below 
3,000 kg/ha. Note that the yields distribution is more spread out in primera, and primera has higher 
productivity estimates; evidenced by the thicker tail on the right (green). We find lower yields than 
we would expect using a semi-technical growing system, which would be between 1,000 kg/ha
and 1,824 kg/ha (around 1038 and 1818 kg/ha depending on the variety). 
TABLE 50: BEANS YIELDS
Yield in Primera (kg/ha) Yield in Postrera (kg/ha)
Mean 888.47 670.44
SD 596.68 440.58
Min 0.27 0.98
Median 781.94 586.45
Max 3127.76 2345.82
Observations 657 738
Notes: Sample for this table consists only of farmers having positive area planted and positive production in either primera or 
postrera. 311 farmers did not have any area in commercial production or had production in primera, and 230 did not have 
commercial area planted or had production in postrera.
FIGURE 9 DISTRIBUTION OF BEANS PRODUCTIVITY BY SEASON
58
As with coffee yields, we next graph production against area reported planted in both seasons 
(Figure 10). A difference between the graphs of coffee and bean yields is that we observe clear 
bunching around what are clear estimates of plot sizes (e.g. around 1.4, 2 or 2.8 ha -equivalent 
to the reported: 2, 3 or 4 manzanas in the postrera). Clearly, farmers are estimating the amount 
of land they have rather than knowing precise amounts, which will affect the denominator in the 
yields calculation. As shown by Abate et al. (2018), inaccurate estimates of land area can affect 
impact estimates from programs on yields, so we will consider whether an alternative method of 
measuring land area is needed for the endline. Otherwise, the pattern is consistent with 
expectations, as production in both seasons grows with land area in a linear relationship.
59
FIGURE 10: BEANS AREA AND BEANS PRODUCTION IN PRIMERA AND POSTRERA
60
Beans Sales
To complete this section, we examine the quantity and value of bean sales (Table 51). As 
discussed at the outset, not all households sell beans, and we would not expect households to 
sell all their production in most cases. Indeed, more farmers producing beans in the primera
appear to also sell beans after the season (81 percent of producers); those who sell do tend to 
sell what appears to be a large share of their production, as the average sales are 0.8 MT (1 
metric tons/MT= 1,000 Kgs). In the postrera season, 73 percent of households recorded sales; 
among those selling, the average amount sold was 0.6 MT. Households report an average sales 
value of around 631 USD in the primera season, which implies a price of 1,051 USD per ton; the 
average price reported was about 2,150 USD per MT in the postrera, higher than in primera.
Figure 11 show the distribution of these sales in each agricultural season. In primera, the 
distribution is more concentrated to the right; this thicker tale shows that primera production is 
more likely to be sold. In the case of postrera we find that the mass of sales is more concentrated 
below 1 MT.
TABLE 51: BEANS SALES, QUANTITY AND VALUE OF SALES
Beans quantity sold 
in Primera (in MT)
Value of Sales in 
Primera (in USD)
Beans quantity sold 
in Postrera (in MT)
Value of Sales in 
Postrera (in USD)
Mean 0.81 631 0.61 1311
SD 0.92 724 0.76 6637
Min 0.00 0.41 0.00 1.48
Median 0.54 369 0.36 277
Max 8.66 5437 6.53 98400
Observations 535 487 541 491
Notes: Sample for beans consists of 968 farmers. Difference in sample sizes for columns due to farmers that did not sell positive quantities 
or have positive revenues.
61
FIGURE 11 DISTRIBUTION OF BEANS SALES BY SEASON
Evaluation Validity Tests at Baseline: Agricultural Productivity - Beans
In this section, we provide balance tests for the treatment and control groups for variables that 
reflect agricultural productivity: area planted, production, yields, and sales. The balance tests, 
look to detect statistically significant differences in the average values for outcomes we plan to 
study in later survey rounds across treatment and control groups. In the beans sample, we might 
expect more differences initially, as the beans sample was not randomly selected. For the 
propensity score matching methodology this validation test is very important not just a 
confirmation and in the coffee evaluation that uses an RCT. The purpose of these test is to, first, 
see if the matching exercise in the sample selection resulted in a sample that is comparable to 
across the treatment assignment. Second, for the variables that are not balanced we can obtain 
some context to inform the matching exercise using the baseline sample, to create weights that 
adjust for this imbalance at baseline and adjust the impact estimates at endline using these 
weights. In what follows we focus on the first of these aspects.
From a statistical perspective we should expect a few statistically significant differences if the 
treatment and control groups are randomly chosen from the same underlying population. This is 
not the case for the beans evaluation design; which uses a propensity score matching. We visited 
treatment areas because we had them on a selected list to be treated, so they are bound to be 
different from the control areas we visited because we knew they were not going to be touched 
by MAS+. To minimize this selection bias, we implemented a pairing or matching of areas which 
should decrease the differences across the treatment and control groups.
Recall that if we test 20 independent null hypotheses at the 5 percent level, in expectation we 
should reject one hypothesis out of chance. Thus, if a handful of rejections occur, it does not 
62
necessarily mean that there is a lack of balance and thus no valid comparisons can be made; 
rather, if we find a pattern of rejections, we need to develop a strategy to overcome that pattern. 
The balance tests for beans are slightly more complicated than the tests for coffee. Farmers can 
plant beans in either the primera or the postrera season, or both. So substantial numbers of 
farmers do not grow beans in one of the two seasons. We adjust our methodology to account for 
this complication. We present tests for whether the conditional means are different from one 
another (e.g. production in postrera, conditional on growing beans in postrera), but we initially test 
whether there is a difference in the proportion of households conducting each activity (e.g. 
planting beans, producing beans) between the treatment and control groups. We can then isolate 
whether any differences would be due to differences in participation in the activity, or due to the 
level of participation.
We first test for differences in area planted and harvested, by season (Table 52). The p-values 
associated with null hypotheses that each variable in the table have the same mean between the 
treatment and control groups are statistically similar at conventional levels (5 percent). As a result, 
there are no statistically significant differences between the treatment and control groups either 
for planting in the primera or postrera seasons, or for the amount produced conditional on planting. 
Figure 12 shows the distribution of area planted for the treatment and control groups. The figure 
shows that the distributions are very similar and near each other across the distribution.
TABLE 52: BALANCE TEST FOR BEANS AREA SIZE PLANTED AND HARVESTED
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Dummy for whether farmer planted beans in Primera 475 0.68 493 0.68 0.92
[39] [0.04] [37] [0.04]
Beans Area Planted in Primera (in ha) 321 0.97 336 1.03 0.54
[39] [0.08] [36] [0.07]
Beans Area Harvested in Primera (in ha) 321 1.87 335 1.55 0.30
[39] [0.29] [36] [0.12]
Dummy for whether farmer planted beans in Postrera 475 0.84 493 0.77 0.19
[39] [0.03] [37] [0.05]
Beans Area Planted in Postrera (in ha) 401 1.07 380 1.00 0.64
[39] [0.10] [37] [0.10]
Beans Area Harvested in Postrera (in ha) 375 0.98 375 0.95 0.79
[39] [0.10] [37] [0.09]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
63
FIGURE 12 DISTRIBUTION OF AREA PLANTED BY TREATMENT ASSIGNMENT
We next test whether household production differs between the treatment and control groups 
(Table 53). Because households may have differentially lost all their production, we also include 
dummy variables for whether or not they produced in each season; as in Table 52, we find no 
systematic differences in production patterns. We find no significant differences. We further find 
that we cannot reject the null hypothesis of equal production levels. So, we find no differences in 
either cultivated area or production for both seasons.
TABLE 53: BALANCE TEST FOR BEANS PRODUCTION
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Farmer had beans production in Primera 475 0.68 493 0.68 0.92
[39] [0.04] [37] [0.04]
Beans Production in Primera (in kg) 321 801.18 336 921.20 0.33
[39] [93.16] [36] [80.92]
Farmer had beans production in Postrera 475 0.78 493 0.77 0.85
[39] [0.03] [37] [0.05]
Beans Production in Postrera (in kg) 371 607.87 380 672.01 0.54
[39] [69.17] [37] [80.21]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
64
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
Consequently, there should not be statistical differences in mean yields, conditional on growing 
beans, between the treatment and control groups. Sure enough, we cannot reject the null 
hypothesis of equivalent mean yields, for either season (Table 54). So, studying production from 
the perspective of cultivated area, total beans production, and yields, we find no statistically 
significant differences. Figure 13 shows the distribution of yields by treatment assignment. The 
distributions are similar; however, the primera yields seem to be higher between 1,000 – 2,000 
kg/ha, with treatment farmers having higher yields. These distributional differences are not 
reflected in the balancing test. However, these are the kind of differences in the distribution that 
should be accentuated in the endline impact estimates if the MAS+ has a large impact on yields 
on different parts of the distribution.
TABLE 54: BALANCE TEST FOR BEANS YIELDS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Yields kg/ha Primera 321 839.86 336 961.58 0.15
[39] [54.59] [36] [64.69]
Yields kg/ha Postrera 366 643.34 372 696.46 0.27
[39] [31.79] [37] [36.47]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
65
FIGURE 13 DISTRIBUTION OF YIELDS BY TREATMENT ASSIGNMENT
Our final comparison related to production is among bean sales. We initially test whether different 
proportions of beans producers sell any beans, by primera and postrera seasons (Table 55, rows 
1 and 4). We find no statistical differences in proportions. We then test whether either the sales 
volume or value differs over the treatment and control groups; we again find no significant 
differences. So, we can extend the earlier statement about production to also include sales.
Figure 14 shows the distribution quantities sold. Reflecting the distribution of yields, the quantity 
sold in primera for the treatment is slightly higher around 2 MT.
66
TABLE 55: BALANCE TEST FOR BEANS SALES, QUANTITY AND VALUE OF SALES
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Dummy for whether farmer sold beans in Primera 475 0.52 493 0.59 0.35
[39] [0.05] [37] [0.05]
Total sales beans in MT - Primera 249 0.77 291 0.83 0.57
[37] [0.09] [36] [0.07]
Value of sales in Lempiras - Primera 224 562.18 268 680.85 0.23
[37] [55.38] [36] [80.47]
Dummy for whether farmer sold beans in Postrera 475 0.53 493 0.61 0.14
[39] [0.04] [37] [0.04]
Total sales beans in MT - Postrera 250 0.59 302 0.62 0.80
[38] [0.07] [37] [0.08]
Value of sales in USD - Postrera 227 1215.81 274 1347.14 0.86
[38] [398.06] [37] [611.69]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
FIGURE 14 DISTRIBUTION OF BEANS QUANTITY SOLD BY TREATMENT ASSIGNMENT
67
We next turn to input use (Table 68). Recall from the descriptive tables that some input use was 
heavy, such as specific fertilizers, whereas other inputs were seldom used, such as inoculant. 
Whether or not input use was heavy or light, we find no significant differences between the 
treatment and control groups for the use of any specific inputs. We further study both total 
material input costs and labor costs by gender (Table 69) in conditional figures; we find no 
significant differences for these variables, either. Even though the sample clusters were not 
chosen randomly, we find no significant differences between the treatment and control groups for 
outcomes related to either production, area planted or productivity.
Inputs
Bean farmers appear to be more likely to report using inputs than coffee farmers in the sample 
(Table 56). Whereas few farmers are using inoculants to promote crop rooting, many farmers use 
one or more types of fertilizer, particularly chemical or foliar fertilizers. Moreover, a relatively large 
proportion of farmers use fungicides (47 percent), herbicides (68 percent), or insecticides (43 
percent) in producing beans.
TABLE 56: SHARE OF FARMERS THAT USED BEANS INPUTS
Inoculant Compos/ 
fertilizer
Chemical 
fertilizer
Foliar 
fertilizer Fungicide Herbicide Insecticide Other 
Inputs
Mean 0.06 0.09 0.41 0.56 0.47 0.68 0.43 0.01
SD 0.23 0.28 0.49 0.50 0.50 0.47 0.50 0.09
Observations 968 968 968 968 968 968 968 968
Notes: Sample for beans consists of 968 farmers. Farmers that did not respond to the inputs module because they reported not having 
used any inputs, are included as “no” for all inputs in figures above.
However, the amount of inputs used, when they are used, appears to be much smaller among 
bean farmers than among coffee farmers (Table 57). The total cost of inputs among those with 
positive values 56 USD. When farmers hire labor, they hire substantially more male labor than 
female labor to work on their beans crops. The total cost of labor, among households hiring any 
labor, is around 688 USD, though note that the figures are highly variable with large SD, implying 
that a few farmers in the sample are hiring substantial amounts of labor whereas most may be 
hiring very little labor. 
TABLE 57: TOTAL COSTS OF INPUTS AND LABOR COSTS
Total cost of 
inputs
Annual cost in male 
labor
Annual cost in female 
labor
Total annual labor 
cost
Annual cost in 
USD 55.94 665.33 22.33 687.66
SD 93.37 1628.17 194.75 1715.92
Observations 848 739 739 739
Note: Only farmers spending positive monetary value on inputs are included.
68
Practices
Next, we examine practices related to bean production (Table 58). The average farmer in the 
sample reports living far from a main road, at 31 minutes away (row 1). The market at which they 
can sell beans is not surprisingly much farther; the average farmer reports needing to travel more 
than 80 minutes, though the median in this case is 30 minutes. Most farmers report that they do 
not irrigate their bean plots. 
TABLE 58: PRACTICES BEANS: ACCESS TO MARKETS, VARIETIES AND SELLING POINTS
Mean SD Observatio
ns
Time from dwelling to main road in minutes 31.79 41.30 966
Time in minutes to place where they sell beans 82.33 323.10 731
Do you have irrigation in any of your plots? 1.95 0.21 965
Did you plant seed:
Criolla 0.70 0.46 800
Amadeus 0.19 0.39 726
Carrizalito 0.01 0.08 717
Deorho 0.04 0.19 719
Azabache 0.00 0.00 715
Paraisito mejorado 0.02 0.13 717
Honduras nutritivo 0.00 0.00 714
INTA Cárdenas 0.00 0.04 717
Other improved 0.03 0.17 723
Other 0.16 0.37 726
Notes: These questions were asked of the sample of 968 beans farmers; differences in sample sizes due to missing 
values to specific questions.
The most common seed used that is reported is the criolla or native; the Amadeus seed is also 
used by most farmers. Some of the seeds that we asked about were not planted at all, and about 
16 percent of farmers noted using a seed that we did not list. 
Table 59 shows to whom households sell their produce and the quantities. Most beans farmers 
sell to intermediaries, estimated at 83 percent; 12 percent of farmers also sell their product at a 
local main square.
TABLE 59 SALES OF BEAN PRODUCTION BY BUYERS
Point to buy or sell: Mean SD Observations
Local main square or farm 0.12 0.33 692
Intermediary or coyote 0.83 0.37 755
Exporter 0.00 0.00 686
IHMA 0.01 0.11 688
Point to buy or sell: Industrial 0.00 0.04 687
Organization 0.01 0.10 687
Other 0.05 0.22 684
Notes: These questions were asked of the sample of 968 beans farmers; differences in sample sizes due to missing 
values to specific questions.
69
We next asked about growing practices, which gives a baseline for primary research question 3 
(Table 60). We find that almost all farmers state that they cleared the bean plot in the past year 
(97 percent). There is clearly some confusion about zero or minimum tillage, as 56 percent of 
farmers state they used one of those two techniques, but they should really be mutually exclusive 
with reporting having prepared soil (61 percent). Whereas few farmers use mechanized planting, 
which would optimize the planting rate, most farmers fertilize their plots (53 percent), control 
weeds (77 percent), or try to limit pests (80 percent). Fewer farmers attempted to do disease 
control (43 percent); however, they may not have experienced a recent disease outbreak.
TABLE 60: PRACTICES IN BEAN PARCELS
During last year in your plot, Mean SD Observations
Did you clean? 0.97 0.17 967
Did you prepare the soil? 0.61 0.49 964
Did you plant with zero tillage? 0.36 0.48 963
Did you plant with minimum tillage? 0.20 0.40 966
Did you do mechanized planting? 0.07 0.26 964
Did you do fertilization? 0.53 0.50 963
Did you do weed control? 0.77 0.42 960
Did you do pest management? 0.80 0.40 965
Did you do disease control? 0.43 0.50 961
Did you harvest? 0.94 0.25 967
Did you pre-dried? 0.56 0.50 963
Did you do aporreado? 0.94 0.23 965
Notes: These questions were asked of the sample of 968 beans farmers; differences in sample sizes due to missing 
values to specific questions.
Finally, we ask about a few remaining practices that could also improve productivity, largely 
related to purchased input requirements (Table 61). We ask first about whether seeds were mixed 
or coated with inoculant, which helps rooting and productivity. Only 5 percent of farmers use 
inoculants. Second, we ask about the number of seeds used per hole; the average is 3, but there 
is some variation to that number. Twenty-five percent of households use either a balanced 
formula fertilizer, but 59 percent do apply a foliar fertilizer (which is a liquid fertilizer applied to 
leaves of a plant). In general, there is room for improvement to yields; however, recall that those 
improvements are costly, and might only improve yields and not profits.
70
TABLE 61: BEANS FARMING PRACTICES
Mean SD Observatio
ns
Did you mix the seeds with inoculant? 0.05 0.22 913
How many seeds did you use per time? 3.00 0.72 912
Did you apply fertilizer, balanced formula? 0.24 0.43 913
Did you apply foliar fertilizer? 0.59 0.49 913
Did you pre-dry? 0.52 0.50 913
Notes: These questions were asked of the sample of 968 beans farmers; differences in sample sizes due to 
missing values to specific questions.
Beans Extension
Few farmers have received extension on beans, also important for realizing the goal of the third 
research question (Table 62). We asked about whether farmers had received extension about 
nine different topics related to beans. Households reported receiving extension messages on 
0.41 topics on average; most farmers (83 percent) have received no extension on beans. Among 
farmers reporting that they received extension, the most common topics that come up are about 
planting and soil preparation, then maintenance and harvest techniques. Interestingly, there are 
very few farmers who have received advice on marketing or market access, so a strong entry 
point with these farmers could be ensuring that they have better market access. Finally, among 
farmers who have received extension, they tend to rate it highly.
TABLE 62: EXTENSION RECEIVED FOR BEANS SAMPLE
Mean SD Observation
s
Number of topics of received extension 0.41 1.15 968
Share of farmers that received extension about:
Soil Preparation 0.09 0.29 968
Planting 0.12 0.33 968
Maintenance 0.06 0.24 968
Harvest 0.06 0.24 968
Post-harvest 0.02 0.14 968
Quality and certification 0.02 0.13 968
Marketing 0.02 0.14 968
Finance and contracts 0.01 0.08 968
Market access 0.01 0.10 968
Duration of beans extension in days 40.63 97.84 167
Quality of extension, 1 to 4 3.57 0.56 160
Notes: All beans farmers have been considered for this table. All farmers who claimed not having received extension in any topic,
and so they were not asked these questions directly, are coded as “no” for each topic. Sample for duration of training and quality 
of extension include only farmers that received extension.
71
Beans Pests
We next examine damages that farmers report being caused by pests, as one specific type of the 
negative shocks affecting beans farmers (Table 63). About 31 percent of all farmers state that 
some damage was done during the primera season, whereas 43 percent of all farmers claim pest￾caused damages during the postrera season. Farmers claim substantial losses due to pests when 
they claim losses; an average of 7.6 quintales in the primera and 8.8 in the postrera, assuming 
that any losses occurred. Among the more common insects causing damage to crops were 
weevils, white files, and gallina ciega (a type of beetle). Other pests were not as common though 
a type of fungus (mancha angular) was mentioned by 11 percent of all farmers as well. Total costs 
reported by farmers of dealing with pests were 25.8 USD on average, conditional on any 
spending; note that not all farmers attempted to solve pest problems with additional inputs.
TABLE 63: PESTS THAT AFFECTED BEANS CROPS
Mean SD Observations
Percent of damage done by pests in primera 32.41 25.40 300
Percent of damage done by pests in postrera 36.08 26.69 418
Losses in QQ due to pest damages in primera 7.57 9.47 316
Losses in QQ due to pest damages in postrera 8.79 11.29 418
Share of farmers that had problems by:
Picudo de la vaina 0.19 0.39 968
Mosca blanca/White Fly 0.20 0.40 968
Gallina ciega/Fall armyworm 0.18 0.38 968
Lorito verde 0.04 0.19 968
Lepidópteros 0.03 0.18 968
Mustia hilachosa 0.09 0.29 968
Roya/Rust 0.08 0.27 968
Mancha angular 0.11 0.31 968
Mosaico viral 0.08 0.27 968
Other 0.11 0.31 968
Average cost to control pest 25.8 34.26 266
Notes: Information on having encountered each specific pest includes all farmers. Sample sizes for percent of crop damage 
caused by pests and pest control costs includes only farmers that had positive values for these two variables.
Investments and Assets of Beans Farmers
To measure investments and agricultural equipment used in bean production, we count a number 
of different types of investments that farmers could have made over the past 12 months (Table 
64). We specifically ask about 11 types of investments. Similarly, we ask how many consumer 
or producer assets out of 19 farmers might own, some of which could be important in producing 
beans. We find that few households made investments in the past year; the average number of 
investments is 0.23. This number might not be all that surprising, if investments last for several 
years, then they do not need to be made each year. Households own, on average, 4.8 different 
72
consumer or producer durables (assets) out of the 19 categories enumerated. We explain both 
investments and equipment further below.
TABLE 64: INVESTMENTS AND ASSET COUNTS
Mean SD Min Median Max Observations
Total no. of investments in the last 12 months (out of 11 options) 0.23 0.56 0.00 0.00 6.00 968
Total no. of equipment owned for beans (out of 19 options) 4.82 2.23 0.00 5.00 16.00 968
A standard way to create an index from a number of potentially related variables is principal 
component analysis. We conduct principal component analysis around investments, and report 
eigenvalues for the first principal component in Table 65. The eigenvalues effectively tell us which 
investments were the more important in generating a variable that incorporates variation in the 
investments across households, with larger values being more important. We find that the “alley 
cultivation” investment is the most important in this regard, with barrier creation being the second 
and third most important. Other investments, such as plant nurseries, greenhouses, or types of 
investments that are tied to irrigation, are less common.
TABLE 65: EIGENVALUES FOR FIRST COMPONENT OF PCA FOR INVESTMENTS
Investment in the last 12 months: EV
Live barriers 0.41
Dead barriers and fences 0.35
Water harvest system 0.30
Alley cultivation 0.51
Plant nursery 0.38
Greenhouses 0.13
Irrigation canal 0.19
Irrigation system 0.37
Solar dryer 0.16
Pulping machine -0.05
Other investments 0.00
Similarly, households could own 19 different types of durable goods that could either be 
associated with higher incomes or production (Table 66). We find that the most important in 
developing the first principal component are primarily consumer durables, such as televisions, 
radios, and cars. Equipment associated with agriculture, such as sprayers, are somewhat less 
important in the first principal component.
73
TABLE 66: EIGENVALUES FOR FIRST COMPONENT OF PCA FOR EQUIPMENT
Type of machinery and equipment: EV
Sprayer with motor 0.22
Motor mounter 0.25
Tractor 0.22
Car 0.39
Truck or pick-up 0.18
Motorcycle 0.14
Bicycle 0.01
Radio 0.41
Television 0.43
Refrigerator 0.34
Computer 0.23
Cell phones 0.29
Gas stove 0.09
Livestock (oxen, cattle) 0.08
Small livestock (goats, pigs, sheep) -0.12
Hens, Ducks, Turkeys, Pigeons -0.03
Horses, mares, mules, males, donkeys 0.05
Other -0.04
Group Membership and Participation
Group membership can play a role in fostering improvements in agricultural production, though it 
is absolutely not sufficient (e.g. Waddington et al., 2014). Given that MAS+ has a focus on 
fostering group participation, particularly within the beans value chain, we investigate the existing 
degree of group participation in the beans sample (Table 67). Participation is defined as either 
active or inactive group membership, or group leadership. We find that the most common type of 
group are water user associations (about 25% of farmers), followed by civic or charitable groups
(18%). Productive groups are rarer; only about a quarter of households report the existence of 
credit or microfinance groups, and only 13 percent report the existence of producer groups. 
TABLE 67: GROUPS IN THE COMMUNITY, BEANS SAMPLE
Mean SD Observations
Share of 
farmers that 
are members
Exists in the community: Productive agricultural 
groups or org. of producers 
0.13 0.34 958 0.08
Exists in the community: Productive agricultural 
groups of commercialization 0.02 0.13 960 0.01
Exists in the community: Group water users 0.83 0.38 967 0.25
Exists in the community: Civic or charitable groups 0.67 0.47 965 0.18
Exists in the community: Women groups 0.11 0.31 961 0.04
Exists in the community: Credit or micro-finance 
groups 
0.25 0.43 966 0.11
Exists in the community: Insurance or mutual aid 
groups 
0.00 0.06 965 0.00
Exists in the community: Commerce and business 
associations 0.00 0.03 966 0.00
Exists in the community: NGO / External Project 0.01 0.11 964 0.01
Exists in the community: Other 0.02 0.13 950 0.01
74
Evaluation Validity Tests at Baseline: Inputs, Extension and Pests for Beans
Our next tests of the validity of the evaluation design focuses on inputs, extension participation 
and the prevalence of pests in beans farmers. First, we turn to input use (Table 68). Recall from 
the descriptive tables that some input use was heavy, such as specific fertilizers, whereas other 
inputs were seldom used, such as inoculant. Whether or not input use was heavy or light, we find 
no significant differences between the treatment and control groups for the use of any specific 
inputs. We further study both total material input costs and labor costs by gender (Table 69) in 
conditional figures; we find no significant differences for these variables, either. Even though the 
sample clusters were not chosen randomly, we find no significant differences between the 
treatment and control groups for outcomes related to either production outputs or inputs.
75
TABLE 68: BALANCE TEST FOR SHARED OF BEANS FARMERS THAT APPLIED INPUTS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Applied Inoculant to promote rooting 475 0.04 493 0.07 0.40
[39] [0.01] [37] [0.03]
Applied Compost/Organic fertilizer 475 0.08 493 0.09 0.81
[39] [0.02] [37] [0.02]
Applied Chemical fertilizer 475 0.41 493 0.40 0.94
[39] [0.04] [37] [0.05]
Applied Foliar fertilizer 475 0.56 493 0.55 0.81
[39] [0.03] [37] [0.05]
Applied Fungicide 475 0.46 493 0.48 0.65
[39] [0.03] [37] [0.04]
Applied Herbicide 475 0.68 493 0.67 0.74
[39] [0.03] [37] [0.04]
Applied Insecticide 475 0.44 493 0.43 0.94
[39] [0.04] [37] [0.04]
Applied Other Inputs 475 0.01 493 0.01 0.59
[39] [0.00] [37] [0.01]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
TABLE 69: BALANCE TEST FOR INPUT COSTS AND LABOR COSTS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Input costs 414 1378.32 434 1350.94 0.92
[39] [196.16] [37] [182.90]
Annual labor cost - men 353 18096.83 386 15488.36 0.50
[37] [2579.15] [37] [2931.83]
Annual labor cost - women 353 885.48 386 284.84 0.17
[37] [421.15] [37] [112.91]
Annual labor cost - total 353 18982.31 386 15773.21 0.44
[37] [2902.29] [37] [2961.16]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
We find significant differences between the treatment and control groups related to extension 
participation in Table 70. Most households in the treatment group have received some form of
extension in the previous 12 months before the survey. We find that 145 of the 160 households 
reporting any extension are in the treatment group. So not surprisingly, they report having received 
76
some extension on several topics, whereas the control group rarely report having received any 
extension. Notably, the extension appears to have been received by at least one household in 
many of the treatment groups, as 34 of the 37 treatment clusters have at least one household 
answering “yes.” It is likely that some of the farmers in the treatment areas have participated in
extension programs with similar topics. However, this slight imbalance will not prevent the
evaluation from providing information on the effects of MAS+. To ensure that we can understand
the impacts of MAS+, it will be important to be very specific on the types of practices and themes 
that are being discussed in MAS+ in any midline and final questionnaires, so the lessons are 
accurately reflected in the evaluation. In addition, we need to be careful to document any similar 
programs that might occur in the same areas, and to be sure we measure the amount of 
participation in MAS+, so we can provide context and explore heterogeneity between those who 
stay in the program from beginning to end and those that leave at some point.
Luckily, we find no differences between the treatment and control groups in terms of production 
above, so we do not have to be concerned that the relatively small share of households that have 
already received some extension by the baseline have received enough (or properly timed) 
extension to affect their beans production. The main consequence of this finding is that we will 
have to control for extension received prior to baseline in an appropriate way when measuring 
impacts of the program, for example, by directly controlling for the months or years that the 
farmers have participated in extension programs. To do so, we plan to collect data—both 
administrative and recall through the survey—on extension participation, and if it varies 
substantially, we can use continuous treatment effect models to measure impacts of additional 
participation (e.g. Imbens and Hirano, 2004).
77
TABLE 70: BALANCE TEST FOR BEANS EXTENSION
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Received extension 475 0.05 493 0.76 0.00***
[39] [0.02] [37] [0.10]
Received extension about Soil preparation 475 0.01 493 0.17 0.00***
[39] [0.00] [37] [0.02]
Received extension about Planting 475 0.01 493 0.23 0.00***
[39] [0.01] [37] [0.03]
Received extension about Maintenance 475 0.01 493 0.12 0.00***
[39] [0.00] [37] [0.02]
Received extension about Harvest 475 0.00 493 0.12 0.00***
[39] [0.00] [37] [0.02]
Received extension about Post-harvest 475 0.00 493 0.04 0.00***
[39] [0.00] [37] [0.01]
Received extension about Quality and certification 475 0.00 493 0.03 0.00***
[39] [0.00] [37] [0.01]
Received extension about Marketing 475 0.01 493 0.03 0.01***
[39] [0.00] [37] [0.01]
Received extension about Finance and contracts 475 0.01 493 0.01 0.96
[39] [0.00] [37] [0.00]
Received extension about Market access 475 0.00 493 0.02 0.06*
[39] [0.00] [37] [0.01]
Duration of coffee extension in days 15 16.53 152 43.00 0.02**
[11] [6.19] [34] [9.22]
Quality of extension from 1 to 4 15 3.53 145 3.57 0.81
[11] [0.16] [34] [0.06]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
Finally, we examine balance related to exposure to pests (Table 71). We first examine whether 
damage differs between the treatment and control groups (rows 1 and 2), by season. We find 
that the treatment group appears to have suffered less damage in the postrera season than the 
control group, and in terms of production losses, they are lower in the treatment group in both 
seasons (rows 3 and 4), at the 10 percent level or better. However, we find no differences in 
exposure to any specific types of pests, nor in the cost of treating crops for pests.
78
TABLE 71: BALANCE TEST FOR PESTS IN BEANS
(1) (2) t-test
Control Treatment (1)-(2)
Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value
Percent of damage in primera 243 20.00 280 17.37 0.32
[36] [2.32] [37] [1.26]
Percent of damage in postrera 243 33.42 280 24.86 0.01***
[36] [2.61] [37] [1.64]
Loss quantity due to pest in primera 137 9.04 179 6.44 0.05**
[35] [1.11] [37] [0.70]
Loss quantity due to pest in postrera 195 10.18 223 7.56 0.07*
[34] [1.13] [35] [0.87]
Had problems with picudo de la vaina 475 0.17 493 0.21 0.33
[39] [0.03] [37] [0.03]
Had problems with mosca blanca 475 0.17 493 0.22 0.17
[39] [0.03] [37] [0.02]
Had problems with gallina ciega 475 0.17 493 0.18 0.88
[39] [0.03] [37] [0.03]
Had problems with lorito verde 475 0.04 493 0.04 0.67
[39] [0.01] [37] [0.01]
Had problems with lepidópteros 475 0.03 493 0.04 0.55
[39] [0.01] [37] [0.01]
Had problems with mustila hilachosa 475 0.10 493 0.09 0.79
[39] [0.03] [37] [0.02]
Had problems with roya 475 0.07 493 0.10 0.17
[39] [0.01] [37] [0.01]
Had problems with mancha angular 475 0.10 493 0.12 0.45
[39] [0.02] [37] [0.02]
Had problems with mosaico viral 475 0.08 493 0.08 0.85
[39] [0.02] [37] [0.02]
Had problems with otra 475 0.09 493 0.12 0.39
[39] [0.02] [37] [0.02]
Does not know 475 0.02 493 0.02 0.90
[39] [0.01] [37] [0.01]
Cost of inputs to control pest 280 305.84 320 255.70 0.58
[37] [67.59] [37] [61.91]
Notes: Values displayed for t-tests are p-values.
Standard errors are clustered at the primary sampling unit level.
***, **, and * indicate significance at the 1, 5, and 10 percent critical levels, respectively.
The implication is that we will also need to continue to monitor whether pest exposure differs 
between treatment and control areas. If treatment areas randomly had fewer pests in the previous 
12 months, it should not affect outcomes at the endline or after households are exposed to the 
intervention. However, if the control areas just have endemic problems with pests that are worse 
than those in treatment areas, we will have to find a way to control for those differences at endline 
as well.
79
In sum, we find two main differences between the treatment and control groups related to beans 
production. First, the beans treatment group has received substantially more extension in the past 
twelve months than the control group. We hypothesize that some of these farmers may have 
received extension with similar themes as MAS+ in the past 12 months; however, it is a relatively 
small number of households. We will control for this exposure in the endline by directly controlling 
for experience in extension programs for farmers in the households and including practices that 
are uniquely promoted by MAS+. Second, we find less exposure to pests, specifically losses, 
among the treatment group relative to the control group. For this second difference, we will keep 
in mind to understand at endline whether pests are more endemic to control areas, so that we 
can control for this difference if necessary as well. In this case, we would include a refined 
matching based on the prevalence of these pests in treatment and control areas.
8 Discussion and Conclusions
This report has discussed the plan for the impact evaluation associated with the MAS+ project 
and the baseline survey conducted for that evaluation. For evaluating interventions among 
households growing coffee and beans, respectively, we split the sample into the households 
growing coffee and those growing beans. The evaluation of the coffee interventions was 
randomized at the cluster level, meaning that specific geographic areas were randomly selected 
for participation, while others will be excluded for the purpose of developing a credible 
counterfactual. The evaluation of the beans interventions, on the other hand, has not been 
randomized. To attempt to reduce selection bias, the research team pre-matched the clusters 
chosen for treatments with potential control groups using the available monitoring data, and 
selected households within control groups that provided the best match possible. The pre￾matching procedure helps minimize any unobservable differences between the targeted group for 
participation and the control group.
After describing how the impact evaluation fits into the overall monitoring and evaluation plan for 
MAS+, the report describes the overall research questions and the survey modules that we use 
to answer those research questions. The survey was developed collaboratively between IFPRI, 
TechnoServe, and ESA Consultores, the firm that collected the data. Data collection took place 
using CAPI and specifically the Survey Solutions application. The main challenge encountered 
during data collection revolved around the control groups for the beans sample; some of the 
communities pre-selected were quite sparsely populated and required replacement. The samples 
that were reached are just short of the projected samples, at 961 and 968 households for coffee 
and beans, respectively; differences will not materially affect statistical power.
The remainder of the report studies both some household characteristics of the sample, and many 
potential outcomes that could be affected by the interventions. After trimming some outliers in 
both samples, production levels seem quite comparable to other studies in Honduras. We notably 
find that neither the coffee nor the beans sample has been exposed to much extension, and in 
particular neither group show much experience with help marketing their crops. Since a main 
projected outcome of the MAS+ is to help farmers improve their connections to agricultural value 
chains, there is clearly scope for interventions helping farmers through this channel.
80
We finally provide balance tests for outcomes among both the coffee and beans samples of 
farmers, which measure whether observable variables differ between the prospective participant 
and control groups. We find that the coffee sample is quite well-balanced. The beans sample is 
balanced among outcomes related to area under cultivation and production, as well as inputs, but 
we find differences in exposure to extension. During the implementation period of MAS+ it will be 
important to track the areas where the program is being offered and be careful to exclude the 
areas that we assigned to control for this intervention. As a result, we will need to keep this point 
in mind during the endline analysis; it will be important to control for this exposure in data analysis 
after the program is complete.
We summarize the baseline findings for important agricultural productivity outcomes—the 
production, yields, volume sold, and the total reported value of sales- in Table 72 for coffee and 
Table 73 for beans. Farmers in the sample have a wide spread in their agricultural productivity, 
which for MAS+ creates an opportunity to provide a diverse set of training that can respond to 
different characteristics of farmers. Namely, for farmers that are below the median of the yield 
distribution, training aimed at increasing productivity can bring about large impact across these 
outcomes. In the case of those farmers that are above the median in the productivity distribution, 
an intervention aimed at increasing their access better markets, to export markets, credit, 
certifications, etc., have the opportunity to have large impact in the total income they derive from 
agriculture.
TABLE 72 SUMMARY BASELINE MEASURES: COFFEE KEY VARIABLES
Mean SD Median Observations
Coffee production in qq WP, total 70 125 34 925
Coffee production in qq WP, plantía 27 49 10 177
Coffee production in qq WP, commercial production 66 118 31 873
Coffee production in qq WP, field losses 7 13 3 239
Coffee production in qq WP, post-harvest losses 8 17 3 179
Total yields (qq/ha WP) 30.9 30.1 21.0 925
Yields in commercial area (qq/ha WP) 36.8 36.7 23.9 861
Yields in plantía (qq/ha WP) 23.0 26.6 12.9 177
Quantity of coffee sold in Wet/Húmedo, MT WP 2.60 8.33 0.78 101
Quantity of coffee sold in Berry/Uva, MT WP 3.02 8.83 1.36 638
Quantity of coffee sold in Dry/Seco, MT WP 4.19 8.80 1.40 55
Quantity of coffee sold in Green/Oro, MT WP 9.21 25.09 2.71 135
Quantity of coffee sold in Total MT WP 4.08 12.95 1.41 898
Sale value coffee berry/uva (USD) 1,521 4,071 477 99
Sale value coffee WP (USD) 2,820 3,660 1,640 621
Sale value coffee Dry/Seco seco (USD) 3,009 6,519 1,103 54
Sale value coffee oro (USD) 5,840 14,671 1,483 128
81
Total Sale value (USD) 3,220 6,887 1,500 873
Notes: Sample for beans consists of 961 farmers. Differences in sample sizes due to farmers that do not have positive quantities for the 
indicators.
TABLE 73 SUMMARY BASELINE MEASURES: BEANS KEY VARIABLES
Mean SD Media
n
Observations
Production, Primera (kgs) 862 943 544 657
Production, Postrera (kgs) 645 761 454 738
Yields, Primera (kg/ha) 888 596 782 657
Yields, Postrera (kg(ha) 670 440 586 738
Volume Sold, Primera (MT) 0.81 0.92 0.54 535
Value of Sales (USD), Primera 631 724 369 487
Volume Sold, Postrera (MT) 0.61 0.76 0.36 541
Value of Sales, Postrera USD) 1311 6637 277 491
Beans Production Home Consumption kg in Primera 140 192 91 633
Beans Production Home Consumption kg in Postrera 144 219 91 708
Notes: Sample for beans consists of 968 farmers. Differences in sample sizes due to farmers that do not have positive quantities for the 
indicators.
9 Recommendations for Monitoring
The results of the baseline study suggest two potential considerations for monitoring. First, note 
that in most cases the selected sample have demonstrated good balance between the treatment 
and control groups, both in the randomized coffee sample and in the non-randomized beans 
sample. It is important to make sure that control groups are left out of the intervention; if the 
intervention were to use those groups or areas, the study design would potentially be 
compromised. It is important to put into place measures that ensure that implementation teams 
do not move into control areas. 
Second, it is worth considering investing in measuring beans plots using GPS. The plot of 
production against area for beans clearly demonstrates bunching around ordinal land areas, 
which are almost surely farmer estimates and affect the measurement of yields. If a more 
accurate measure of land area can be available, it will likely tighten up the distribution of yields 
for beans. However, the challenge is that enumerators must be further trained, and the activity 
can be costly and time consuming in terms of going to the field and walking around it. If the 
monitoring data is clearly suggestive of impacts on yields close to the endline, we strongly suggest 
taking the opportunity to attempt to accurately measure beans plots.
82
References
Abate, G. T., Bernard, T., de Brauw, A., & Minot, N. (2018). The impact of the use of new 
technologies on farmers' wheat yield in Ethiopia: evidence from a randomized control 
trial. Agricultural economics (Amsterdam, Netherlands), 49(4), 409-421.
Cameron, C and Miller, 2015 “A Practitioner's Guide to Cluster-Robust Inference,” Journal of 
Human Resources, Spring 2015, Vol.50, No. 2, pp.317-373.
Deaton and Cartwright, 2017, “Understanding and misunderstanding randomized controlled 
trials,” Social Science & Medicine, Volume 210, August 2018, Pages 2-21
Harrison, Glenn W., 2011, “Randomisation and Its Discontents,” Journal of African Economies, 
20(4), 2011b, 626-652.
Heckman, J, Ichimura, H and Todd, P 1997, ”Matching as an Econometric Evaluation Estimator: 
Evidence from Evaluating a Job Training Program,” Review of Economic Studies, pp. 605-654.
Heckman, J, Ichimura, H, Smith, J and Todd, P 1998, “Characterizing Selection Bias Using 
Experimental Data, ” Econometrica, pp. 1017-1098.
Hirano, K and Imbens, GW 2002, “Estimation of Causal Effects using Propensity Score Weighting: 
An Application to Data on Right Heart Catheterization,” Health Services & Outcomes Research 
Methodology, vol 2, pp. 259-278.
Imbens, G.W., and K. Hirano, 2004. “The Propensity Score with Continuous Treatments,” working 
paper, Harvard University.
Imbens, G. W., Newey, W. K.,, and Ridder, G. , 2005 “Mean-Square-Error Calculations for 
Average Treatment Effects,” IEPR Working Paper No. 05.34. 
McKenzie, D., 2012. "Beyond baseline and follow-up: The case for more T in 
experiments," Journal of Development Economics, Elsevier, vol. 99(2), pages 210-221.
Ravallion, M., 2015. "On the Implications of Essential Heterogeneity for Estimating Causal 
Impacts Using Social Experiments," Journal of Econometric Methods, De Gruyter, vol. 4(1), pages 
1-7, January.
Waddington, H., Snilstveit,, B., Hombrados, J., Vojtkova, M., Phillips, D., Davies, P., White, H., 
2014, “Farmer Field Schools for Improving Farmer Practices and Farmer Outcomes: A Systematic 
Review,” Campbell Systematic Reviews 2014:6.
83
Appendix 1: Collective Action, Transfers, Credit, Shocks, 
Aspirations and ICTs
Coffee Sample
Productivity and price expectations of coffee farmers
Table 74 shows the productivity and price expectation of farmers. Their expectation of their 
productivity is above the realized productivity we found. For example, the median farmer expects 
to produce 20 quintales green per manzana under good conditions and 12 quintales green per 
manzana under bad conditions while the median productivity we estimate it at 11 quintales green
per manzana. The prices expectations are also above the realized prices with the median price 
expected under bad market conditions being 50 percent above the median price in the sample. 
Under these measures the farmers are having a bad year.
TABLE 74: COFFEE PRACTICES: ACCESS TO MARKET AND EXPECTATIONS
Mean SD Min Median Max Observations
Under good conditions, what would be the production 
in next season (QQ/MZ) 38.07 54.57 0.83 20.00 490.00 904
Under bad conditions, what would be the production 
in next season (QQ/MZ) 26.18 47.08 0.40 12.00 450.00 904
Under good market conditions, what would be the 
price next season? 2,434.27 1556.38 5.00 2250 25,000 587
Under bad market conditions, what would be the price 
next season? 1,592.47 717.37 2.20 1500 5,030 586
Notes: These questions were asked to sample of 961 coffee farmers. Difference in sample sizes is due to missing values to specific questions
Collective Action and Aspirations
We next ask about whether communities participate in a variety of different types of collective 
action (Table 75). We find very few examples of collective action; the most common one is either 
shared work or work exchanges and water conservation community work, but only 7 percent of 
households report them in their communities. Other types of community based collective action 
are rare, which suggest that community cohesion is not strong in rural Honduras.
84
TABLE 75: COLLECTIVE ACTION
Mean SD Observations
Community participated in: Construction for water conservation 0.07 0.25 960
Community participated in: Joint purchase of agricultural inputs 0.02 0.13 958
Community participated in: Shared work or hand back 0.06 0.23 960
Community participated in: Joint agricultural certification 0.00 0.03 960
Community participated in: Gathering / selling joint agricultural products 0.01 0.07 957
Community participated in: Other collective development activity 0.01 0.11 957
We next explore a set of questions asked about aspirations, control of one’s fate, and self-esteem 
(Figure 15). The bars left to right measure the proportion of the sample answering strongly agree 
to strongly disagree. The different questions show farmers tend to agree with the statements that 
mark independence and agency, having access to help in the community, and life improving in 
the future.
85
FIGURE 15: ASPIRATIONS, LOCUS OF CONTROL AND SELF-ESTEEM PERCEPTIONS
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
You can solve problems with your crops without help
You can decide what you are going to do in your life
If an adverse weather event occurs, however much you do, you…
You are very good at taking care of the crops you have
You can negotiate the sale price of your products and get better…
In the future, your life condition will improve
It is very important that parents keep their promises to their children
You have specific goals and plans for future growth of your crops
You are satisfied with the production level of your plot / business
Your neighbors don't like you
You do not have much to be proud of
Other people like to work with you
Most neighbors would be willing to help you if needed
You are afraid to try new things / something new
You have had conflicts with neighbors for when to sell or who to…
You have had conflicts with neighbors for those who participate in…
Strongly agree Agree Disagree Strongly disagree
86
Transfers and Remittances
Farmers can have different sources of income. Some of them are transfer from social programs 
or remittances from migrants abroad. We find that fully 20 percent of the sample received 
remittances from outside the country. The most common types of transfers received from public 
programs are the Bono Vida Mejor and Bono 10,000 programs; about 20 percent of households 
received one of these transfers. 
TABLE 76: REMITTANCES AND TRANSFERS RECEIVED BY COFFEE SAMPLE
Mean SD Observations
During the last 12 months, received: Remittances from relatives out of 
the country 0.20 0.40 958
During the last 12 months, received: Remittances from relatives within 
the country 0.05 0.22 959
During the last 12 months, received: Bono Vida Mejor/Bono 10,000 0.20 0.40 959
During the last 12 months, received: Stove, Ecofogón Vida Mejor 0.08 0.28 959
During the last 12 months, received: 3rd age bonus 0.00 0.06 957
During the last 12 months, received: Bolsa Solidaria 0.08 0.28 959
During the last 12 months, received: Fees, alimony 0.00 0.06 959
During the last 12 months, received: Other 0.00 0.07 932
Credit
We track household participation in group-based savings, microfinance, or lending programs. 
Group-based savings programs are formal or informal community programs that serve as a 
mechanism for people in poor communities with otherwise limited access to financial services to 
pool their savings. The specific composition and function of the savings groups group vary and 
can include rotating loan disbursement. The definition is inclusive of all of the different types of 
group-based savings programs. According to the World Bank, microfinance can be defined as 
approaches to provide financial services to households and microenterprises that are excluded 
from traditional commercial banking services. Typically, these are low-income, self-employed or 
informally employed individuals, with no formalized ownership titles on their assets and with 
limited formal identification papers. 
The benefits of financial inclusion include: lower transaction costs of day to day interactions (e.g. 
Mobile Money), ability to grow savings to smooth consumption and mitigate against shocks, and 
access to credit to invest in Micro, Small and Medium enterprises (MSME). Group-based savings 
programs are formal or informal community programs that serve as a mechanism for people in 
poor communities, with otherwise limited access to financial services, to pool their savings. The 
specific composition and function of the savings groups group vary and can include rotating 
disbursement as well as accumulating savings models. 
A household is considered to be participating in group-based savings, micro-finance or lending 
program if any member of the household saved money with or took a loan or borrowed cash or 
in-kind from a group-based savings, micro-finance or lending program in the past 12 months. Only 
27 percent (or 258 of 961 households) have received credit in the past 12 months (Table 77). 
Most of the credit they received comes from formal banks and credit and savings associations.
87
TABLE 77: CREDIT FOR COFFEE SAMPLE
Mean SD Observatio
ns
In the last 12 months, did you receive credit from a public or private 
institution? 0.27 0.44 961
Institution that gave you credit: BANHCOFFEE 0.02 0.15 258
Institution that gave you credit: Another bank 0.26 0.44 258
Institution that gave you credit: Savings and Credit Cooperative 0.24 0.43 258
Institution that gave you credit: Another regulated formal institution 0.14 0.35 258
Institution that gave you credit: Agricultural Cooperative or Producers 
Association 0.05 0.21 258
Institution that gave you credit: Communal bank or other 
unconventional 0.02 0.14 258
Institution that gave you credit: Caja Rural / Caja Municipal 0.17 0.38 258
Institution that gave you credit: Grupo solidario 0.01 0.09 258
Institution that gave you credit: NGO or Project 0.02 0.12 258
Institution that gave you credit: SAG program or government 0.00 0.00 258
Institution that gave you credit: Other non-regulated institution 0.02 0.14 258
Institution that gave you credit: Farming companies 0.00 0.06 258
Institution that gave you credit: Agricultural input store 0.00 0.06 258
Institution that gave you credit: Another store / merchant 0.01 0.11 258
Institution that gave you credit: Coyote / Local Copier 0.03 0.17 258
Institution that gave you credit: Coyote / foreign collector 0.00 0.00 258
Institution that gave you credit: Lender 0.01 0.11 258
Institution that gave you credit: Another farmer of the municipality 0.00 0.00 258
Institution that gave you credit: Family member or friend 0.02 0.14 258
Institution that gave you credit: Another informal lender 0.00 0.06 258
Shocks
Negative shocks can have deleterious effects on household production, and so it is important to 
measure shocks to understand whether they might have affected households. We ask households 
in the sample whether they have experienced negative shocks over the past 12 months (Table 
78). Many households noted experiencing shocks in the past 12 months; the most common 
reported negative shock was an increase in agricultural input prices (64 percent), followed by low 
agricultural product prices (46 percent). Notably, 20 percent also reported being affected by heavy 
rains, so both a lack of rain and too much rain affects different parts of the sample. These 
measures of shocks may be helpful in controlling for differences in productivity among farmers.
88
TABLE 78: SHOCKS EXPERIENCED IN THE LAST 12 MONTHS
Events that affected you in the last 12 months Mean SD Observations
Drought 0.23 0.42 947
Fires / burning 0.03 0.18 947
A lot of rain / floods 0.20 0.40 947
Migration of a household members 0.03 0.17 947
Pests / animal disease 0.14 0.35 947
Sick people working at home 0.16 0.37 947
Low agricultural product prices 0.46 0.50 947
Road strikes 0.07 0.25 947
Increase in agricultural input prices 0.64 0.48 947
Less work available than before 0.27 0.44 947
Lack of contracted labor 0.30 0.46 947
Robbery / assaults on house 0.02 0.13 947
Violent death of a household member 0.01 0.11 947
Information and Communication Technologies
The survey form asked about where households received information about agriculture, more 
generally (Table 79). The most common source of agricultural information is the radio, followed 
by the TV; however, only 19 percent of households mentioned the radio, and only 8 percent 
mentioned the TV. Other sources of agricultural information were mentioned very infrequently, 
meaning by only 1 or 2 percent of sample households at most. In general, the majority of sample 
households appear not to receive much widely disseminated agricultural information.
TABLE 79: SOURCES OF RECEIVED AGRICULTURAL INFORMATION
Mean SD Observations
Sources of agricultural information: Radio 0.19 0.39 957
Sources of agricultural information: TV 0.08 0.28 959
Sources of agricultural information: Internet 0.01 0.12 957
Sources of agricultural information: Messages 0.00 0.06 960
Sources of agricultural information: Newspapers 0.00 0.06 960
Sources of agricultural information: Merchant 0.02 0.14 956
Sources of agricultural information: Supplies 0.01 0.08 959
Sources of agricultural information: Association 0.02 0.14 959
Sources of agricultural information: Other 0.01 0.11 938
Finally, we asked about some specific uses of the internet and cell phones (Table 80). Although 
not specifically asked, through the averages presented 28 percent of households have 
smartphones, since they report both accessing the internet and using Whatsapp, a smartphone 
application. A few households report owning a computer or using a neighbor’s computer to access 
the internet. Most of the sample households do not have internet access. Far more households 
have cell phones; whereas only a few use cell phones to buy or sell agricultural products, the 
majority of the sample reports either sending or receiving messages by phone.
89
TABLE 80: INTERNET AND CELLULAR USE
Mean SD Observatio
ns
Internet access by: Own computer with internet 0.03 0.17 960
Internet access by: Computer of a negihbor or family 0.01 0.07 957
Internet access by: Mobile internet (cellphone) 0.28 0.45 960
Internet access by: Cybercoffee 0.00 0.03 958
Internet access by: Library, school, other institution 0.00 0.05 959
Internet access by: Other 0.00 0.00 953
Use cellular to: Make purchases of agricultural products or 
processed food 0.07 0.25 950
Use cellular to: Sales of agricultural products or processed food 0.08 0.28 950
Use cellular to: Send messages in general 0.54 0.50 960
Use cellular to: Receive messages in general 0.61 0.49 958
Use cellular to: use Whatsapp 0.28 0.45 955
90
Beans Sample
Collective Action and Aspirations
We ask about whether communities participate in a variety of different types of collective action 
(Table 81). We find very few examples of collective action; the most common one is either shared 
work or work exchanges, but only 11 percent of households report them in their communities, 
which does not mean they participate in labor exchanges. Other types of community based 
collective action appear quite rare, which is consistent with the idea that community cohesion is 
not strong in rural Honduras.
TABLE 81: COLLECTIVE ACTION
Mean SD Observations
Community participated in: Construction for water conservation 0.05 0.22 968
Community participated in: Joint purchase of agricultural inputs 0.01 0.12 967
Community participated in: Shared work or hand back 0.11 0.31 967
Community participated in: Joint agricultural certification 0.00 0.06 965
Community participated in: Gathering / selling joint agricultural 
products 0.01 0.12 967
Community participated in: Other collective development activity 0.02 0.14 965
As with the coffee sample, we next explore a set of questions asked about aspirations, control of 
one’s fate, and self-esteem (Figure 16). And as with the coffee sample, the bars left to right 
measure the proportion of the sample answering strongly agree to strongly disagree. 
91
FIGURE 16: ASPIRATIONS, LOCUS OF CONTROL AND SELF-ESTEEM PERCEPTIONS
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
You can solve problems with your crops without help
You can decide what you are going to do in your life
If an adverse weather event occurs, however much you do, you can not…
You are very good at taking care of the crops you have
You can negotiate the sale price of your products and get better price…
In the future, your life condition will improve
It is very important that parents keep their promises to their children
You have specific goals and plans for future growth of your crops
You are satisfied with the production level of your plot / business
Your neighbors don't like you
You do not have much to be proud of
Other people like to work with you
Most neighbors would be willing to help you if needed
You are afraid to try new things / something new
You have had conflicts with neighbors for when to sell or who to sell in…
You have had conflicts with neighbors for those who participate in…
Strongly agree Agree Disagree Strongly disagree
92
Remittances
High migration rates in Honduras make it important to ask about remittances, as they can affect 
production either positively or negatively. We also ask about other types of transfers that farm 
households might have received, largely related to social programs (Table 82). We find that fully 
21 percent of the sample received remittances from outside the country, which is similar to the 
coffee sample; however, more households in the bean sample receive remittances from within 
the country (9 percent) than within the coffee sample (5 percent). The most common types of 
transfers received from public programs are the Bono Vida Mejor and Bono 10,000 programs; 
about 19 percent of households received one of these transfers, whereas 13 percent received 
transfers from Bolsa Solidaria. 
TABLE 82: REMITTANCES AND TRANSFERS RECEIVED BY BEANS SAMPLE
Mean SD Observations
During the last 12 months, received: Remittances from relatives out 
of the country 0.21 0.41 966
During the last 12 months, received: Remittances from relatives 
within the country 0.09 0.29 967
During the last 12 months, received: Bono Vida Mejor/Bono 10,000 0.19 0.39 963
During the last 12 months, received: Stove, Ecofogón Vida Mejor 0.10 0.30 967
During the last 12 months, received: 3rd age bonus 0.00 0.06 966
During the last 12 months, received: Bolsa Solidaria 0.13 0.33 967
During the last 12 months, received: Fees, alimony 0.00 0.03 966
During the last 12 months, received: Other 0.00 0.05 949
Credit
Whereas finance is not seen as a constraint to bean production as it may be for coffee production, 
nonetheless we are interested in whether bean producing households have received any credit. 
We find that only 14 percent of households have received formal or informal credit in the past 12 
months (Table 84). Among those reporting they received credit, banks other than the 
BANHCOFFEE, perhaps not surprisingly, are the most common source at 35 percent. Other 
answers were the credit and savings association at 20 percent, and then a caja rural or caja 
municipal, which is an informal institution, at 16 percent. Nonetheless, the beans sample appears 
to have very little access to credit.
93
TABLE 83: CREDIT RECEIVED BY BEANS SAMPLE
Mean SD Observatio
ns
In the last 12 months, did you receive credit from a public or private 
institution? 0.14 0.35 968
Institution that gave you credit: BANHCOFFEE 0.01 0.08 140
Institution that gave you credit: Another bank 0.35 0.48 140
Institution that gave you credit: Savings and Credit Cooperative 0.20 0.40 140
Institution that gave you credit: Another regulated formal institution 0.13 0.34 140
Institution that gave you credit: Agricultural Cooperative or Producers 
Assoc. 0.02 0.15 140
Institution that gave you credit: Communal bank or other 
unconventional 0.04 0.20 140
Institution that gave you credit: Caja Rural / Caja Municipal 0.16 0.37 140
Institution that gave you credit: Grupo solidario 0.01 0.12 140
Institution that gave you credit: NGO or Project 0.01 0.08 140
Institution that gave you credit: SAG program or government 0.01 0.08 140
Institution that gave you credit: Other non-regulated institution 0.01 0.08 140
Institution that gave you credit: Farming companies 0.01 0.08 140
Institution that gave you credit: Agricultural input store 0.01 0.08 140
Institution that gave you credit: Another store / merchant 0.01 0.12 140
Institution that gave you credit: Coyote / Local Copier 0.00 0.00 140
Institution that gave you credit: Coyote / foreign collector 0.00 0.00 140
Institution that gave you credit: Lender 0.01 0.08 140
Institution that gave you credit: Another farmer of the municipality 0.00 0.00 140
Institution that gave you credit: Family member or friend 0.02 0.15 140
Institution that gave you credit: Another informal lender 0.00 0.00 140
Shocks
Again, negative shocks can have a very important impact on agricultural production, so we also 
ask households in the bean sample whether they have experienced negative shocks over the 
past 12 months (Table 84). Many households noted experiencing shocks in the past 12 months; 
the most common reported negative shock was an increase in agricultural input prices (51 
percent), followed by drought (47 percent). Notably, 41 percent also reported being affected by 
low agricultural output prices; it is not clear from this table whether farmers actually experienced 
decreases in output prices or increases in input prices, or if they are noting what they perceive as 
regular conditions. Also notably, 28 percent also reported being affected by heavy rains, so both 
a lack of rain and too much rain affects different parts of the sample. These measures of shocks 
may be helpful in controlling for differences in productivity among farmers. 
94
TABLE 84: SHOCKS EXPERIENCED DURING THE LAST 12 MONTHS, BEANS SAMPLE
Events that affected you in the last 12 months Mean SD Observation
s
Drought 0.47 0.50 959
Fires / burning 0.12 0.32 959
A lot of rain / floods 0.28 0.45 959
Migration of a household members 0.03 0.18 959
Pests / animal disease 0.14 0.35 959
Sick people working at home 0.12 0.33 959
Low agricultural product prices 0.41 0.49 959
Road strikes 0.08 0.27 959
Increase in agricultural input prices 0.51 0.50 959
Less work available than before 0.21 0.41 959
Lack of contracted labor 0.17 0.37 959
Robbery / assaults on house 0.02 0.15 959
Violent death of a household member 0.01 0.10 959
Information and Communication Technologies
As with the coffee sample, we ask about how farmers receive agricultural information and about 
their use of the internet and other information and communication technologies. First, we ask 
about whether farmers receive agricultural information from different sources, such as the radio, 
television, etc. We find very low response rates, suggesting that bean farmers are not receiving 
much information from widely disseminated sources (Table 85). The most common source of 
information mentioned is radio, followed by television, but only 10 and 6 percent of the sample 
mentioned those two sources, respectively.
TABLE 85: SOURCES OF AGRICULTURAL INFORMATION
Mean SD Observations
Sources of agricultural information: Radio 0.10 0.31 967
Sources of agricultural information: TV 0.06 0.23 967
Sources of agricultural information: Internet 0.00 0.06 967
Sources of agricultural information: Messages 0.00 0.06 967
Sources of agricultural information: Newspapers 0.01 0.08 964
Sources of agricultural information: Merchant 0.01 0.10 964
Sources of agricultural information: Supplies 0.02 0.14 965
Sources of agricultural information: Association 0.01 0.10 966
Sources of agricultural information: Other 0.05 0.21 956
Similar to the coffee sample, we find that around 26 percent of the beans sample appear to have 
a smartphone, as 26 percent use the internet on a mobile device and 27 percent report using 
Whatsapp (Table 86). Otherwise, few respondents appear to have access to the internet. 
95
Households report using their cell phones to send and receive messages most often; as with the 
coffee sample, few households appear to be either purchasing or selling agricultural products on 
their phones, at 11 and 16 percent, respectively.
TABLE 86: INTERNET AND CELLPHONE USE
Mean SD Observations
Internet access by: Own computer with internet 0.03 0.16 966
Internet access by: Computer of a negihbor or family 0.00 0.06 967
Internet access by: Mobile internet (cellphone) 0.26 0.44 968
Internet access by: Cybercoffee 0.01 0.09 967
Internet access by: Library, school, other institution 0.01 0.08 967
Internet access by: Other 0.00 0.00 965
Internet access by: Not applicable 0.01 0.10 911
Use cellular to: Make purchases of agricultural products or 
processed food 0.11 0.31 967
Use cellular to: Sales of agricultural products or processed food 0.16 0.37 965
Use cellular to: Send messages in general 0.48 0.50 966
Use cellular to: Receive messages in general 0.51 0.50 964
Use cellular to: use Whatsapp 0.27 0.45 960