$.33?

firm-w

 

a"

 

Trials

(5?

5%!“ 755

 

LIBRARY

 

University

Michigan State

 

 

This is to certify that the

dissertation entitled

ESSAYS ON THE ECONOMIC IMPACT OF DISEASE-
RESISTANT BEAN RESEARCH IN HONDURAS

presented by

David Len Mather

has been accepted towards fulfillment
of the requirements for

Ph.D. degreein Agricultural 'Economics

K RV ”By/MT
\Jufl’t‘ q,03 /~A “W850? 41.4. _

Datefl ; 200: Dissertation Advisor

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE
’ Ann:

 

W 7.qu

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuepes-p. 1 5

 

ESSAYS ON THE ECONOMIC IMPACT OF DISEASE-RESISTANT
BEAN RESEARCH IN HONDURAS

By

David Len Mather

A DISSERTATION
Subnfifled

in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Agricultural Economics

2003

ABSTRACT

ESSAYS ON THE ECONOMIC IMPACT OF DISEASE-RESISTANT
BEAN RESEARCH IN HONDURAS

By

David Len Mather

The three essays in this dissertation address different aspects of the impact of
disease-resistant bean research in Honduras and are each basedupon a random sample
survey of Honduran bean farmers (N=210) in 2001. The first essay presents evidence of
recent adoption rates of disease-resistant bean varieties (RVs), the farm-level benefits of
RV adoption, and the ex post rate of return to bean research in Honduras from 1982-2010.
The results demonstrate that the investment in breeding disease-resistant beans in
Honduras has been profitable; under the base scenario assumptions, the ex post economic
rate of return to disease-resistant bean research in Honduras during the period 1982-2010
is 41.2 %. In addition, adoption of RVs is widespread and scale-neutral; 42% of sample
bean farmers used an RV in the postrera season of 2000.

The second essay addresses the methodological challenges involved in the
evaluation of a technology which maintains rather than augments yield, and demonstrates
an econometric method (borrowed from the wage differential literature) to construct
appropriate counterfactuals in the estimation of the impact of maintenance research
technologies. The method is applied to test for selection bias and estimate differentials
between RV and traditional variety (TV) bean yields in survey data from Honduras by

constructing the counterfactual to RV yields as what yields RV users would have

obtained had they continued to grow TVs (i. e., the TV yield of RV users). The corrected
yield model predicts that RV growers enjoy net income gains of 13 to 19% in the postrera
season, compared to what they would have earned growing TVs; while the uncorrected
OLS model predicts that RV growers see either no income gain or even losses by growing
RVs (relative to TVs). This application demonstrates that the implications of this method
are significant for the assessment of maintenance research impacts.

The third essay uses the farm survey data in probit analysis to investigate the
significance and magnitude of “varietal/ breeding” factors vs. “access” factors in the
adoption of early and recent generation RVs in Honduras. The results indicate that
adoption of early RVs, Dorado and Don Silvio, are more constrained by poor market
characteristics (breeding factors) than lack of access. The poor market characteristics of
these early generation RVs, reflected in average price discounts of 15% relative to TVs,
has constrained adoption primarily to disease-intense areas, and has reduced the net
income gains for adopters. By contrast, the results of analysis of a recent generation RV,
Tio Canela (released in 1997), indicate that the market acceptance (breeding) aspects of
the variety appear to be much improved over those of Dorado, as Tio Canela has been
adopted in areas of both historically high and low disease pressure. However, the
principal constraint to further Tio Canela adoption is information and seed access (a
policy aspect). With increased access, current Dorado users (in disease-intense areas)
would be expected to switch to Tio Canela and enjoy larger net incomes due to Tio
Canela’s smaller price discounts. The adoption of Tio Canela outside of traditionally
high-disease pressure areas implies that the variety offers positive net benefits for growers

in low-disease pressure areas as well.

Cepyright by
DAVID LEN MATHER
2003

ACKNOWLEDGMENTS

I am grateful to my dissertation advisor Dr. Jim Oehmke for providing me with
patient and constructive feedback at all stages of this research, and for helping me
conceptualize the three essays. I am also grateful to my major professor Dr. Rick
Bernsten for guidance during my program, for his confidence in me, and for the
opportunity to work with the Bean/Cowpea CRSP on this and other research projects. I
also thank Dr. Eric Crawford and Dr. Irv Widders for serving on my committee and for
providing valuable feedback on the essays. I thank Dr. Jeff Wooldridge and Dr. Roy
Black, neither of whom served on my committee but provided invaluable econometrics
assistance. I am grateful to the Bean/Cowpea CRSP and the Department of Agricultural
Economics for funding my assistantships and fieldwork, and to the College of Agriculture
and Natural Resources for a Dissertation Completion Fellowship. In the Department of
Agricultural Economics, 1 thank Dr. Eric Crawford for so ably running the graduate
program, Sherry Rich for solving any and all administrative problems I may have
encountered in the program, and Judi Dow for making the Ag Econ Reference Room such
an efficient and pleasant place to work.

In Honduras, there were several institutions and many individuals who supported
the field component of this research. The survey on which these three essays are based
was funded by the Bean/COWpea CRSP and PROF RIJOL. We also received significant
logistical support from the Programa de Investigaciones en Frijol at Zamorano, directed

by Dr. Juan Carlos Rosas. Dr. Rosas provided invaluable assistance throughout the entire

process of this research by sharing his extensive knowledge of bean production in
Honduras. For their help with the survey design, I thank: Dr. Bernsten, Dr. Rosas,
Aracely Castro (Zamorano), Abelardo Viana Ruano and Julio Martinez (PROFRIJOL),
and Danilo Escoto (DICTA). I am extremely grateful to the team of enumerators, who
worked tirelessly to find and to carefully interview each of the survey farmers: Danilo
Escoto, Roger Ramos, Reynaldo Argenal, Cristobal Artica Alvarado, Manuel de la
Amaya, Rafael Ramon Elvis, and Calixto. For assistance with data entry, I am grateful to
Maria Bravo. For assistance with the collection of secondary data, I thank Aracely
Castro, Roger Ramos, and Wolfgang Pejuan. Of course, this research is fundamentally
based on the cooperation and generosity of the bean farmers in El Paraiso, Francisco
Morazan, and Olancho who agreed to share their time and their knowledge of and
experience with bean production.

I will cherish the times of joy and frustration that I shared with so many fellow Ag
Econ grad students as we worked our way through courses, exams, and research. I
especially thank Matt Schaefer, Gregg Hadley, George Young, Jim Stems, Chris “Buck”
Penders, Roger Bairstow, David Yanggen, Alyse Schrecongost, Maria Laura Donnet, and
Kraig Jones for their friendship. I also especially thank Kofi Nouve and Asfaw Negassa
for their support during the final stages of my dissertation. I am especially grateful to
Brady J. Deaton for his advice, support and friendship throughout my program. I also
thank his wife, Justine Richardson, for her friendship, peerless hospitality and amazing

cooking. Brady and Justine (and Liam) have been my family in Michigan.

vi

I thank Rev. Allen Kannapell (Episcopal Ministry to MSU) for spiritual guidance
and friendship, and I cherish the time I spent playing worship music with Allen,
Nokuthula Cele, Tapiwa Gandiya, and Nate Sten.

I thank my fellow futbolistas in No Dignity, Flying Eagles, and the pick-up soccer
group for sharing the beautiful game.

I also want to thank a few of my college professors: Dr. Robin Gottfried, Dr.
Charles Brockett, and Dr. Elwood Dunn of The University of the South, Sewanee, TN.
Their love of teaching and their knowledge of developing economies and politics inspired
me to study international development.

Words cannot express my gratitude and my love for my mother, father, and
brother. I am truly blessed by God to have and be a part of such a family. Finally, I thank
Jesus Christ for patience, strength, courage, wisdom, and above all for His redeeming and

sustaining love.

vii

TABLE OF CONTENTS

LIST OF TABLES ..................................................... x
LIST OF FIGURES ..................................................... xi
INTRODUCTION ....................................................... 1
ESSAY # 1: THE ECONOMIC IMPACT OF DISEASE-RESISTANT BEAN
RESEARCH IN HONDURAS ....................................... 5
1. Introduction .................................................... 6
2. Disease-resistant bean varieties .................................... 7
3. Survey methodology ............................................ 10
3.1 Previous Research ....................................... 10
3.2 Current Research ........................................ ll
4. Adoption of disease-resistant varieties .............................. 12
5. Estimation of the farm-level impact of disease-resistant varieties ......... 16
5.1 Limitations of experimental trial and survey data ............... 16
5.2 F arm-level price discounts ................................. 17
5.3 Experimental and survey yields ............................. 18
5.4 Disease frequency ....................................... 20
5.5 Expected utility framework ................................ 20
5.6 Incremental farm-level costs ............................... 22
6. Economic surplus analysis ....................................... 23
6.1 Aggregate benefits ....................................... 23
6.2 Research costs .......................................... 24
6.3 Rate of return ........................................... 26
6.4 Sensitivity analysis ....................................... 27
6.5 Distribution of benefits ................................... 29
7. Conclusion ................................................... 30
References ...................................................... 30
Appendix ....................................................... 33
ESSAY # 2: IMPACT ASSESSMENT OF TECHNOLOGIES THAT MITIGATE
ADVERSE CIRCUMSTANCES: DISEASE-RESISTANT BEANS IN
HONDURAS .................................................... 36
1. Introduction ................................................... 37
2. Model ....................................................... 39
3. The sample selection problem .................................... 40

viii

4. Estimation procedure ........................................... 43
5. Background and data ............................................ 46
6. Estimation results of varietal choice function ......................... 50
7. Estimation results of bean yield function ............................ 54
8. Counterfactual predictions ....................................... 58
9. Implications for impact analysis ................................... 64
10. Conclusions .................................................. 65
References ...................................................... 66
Appendix ....................................................... 68

ESSAY # 3: FACTORS INFLUENCING THE ADOPTION OF DISEASE-RESISTANT

BEAN VARIETIES IN HONDURAS ................................. 69
1. Introduction ................................................... 70
2. Background ................................................... 72
2.1 Organizations ........................................... 72

2.2 BGYMV-resistant varieties ................................ 72

2.3 Seed distribution channels ................................. 73

2.4 Post—Mitch seed distribution ............................... 74

2.5 Survey farmers .......................................... 74

3. Adoption model ................................................ 77
4. Estimation of adoption equations ................................... 82
4.1 Dorado ................................................ 82

4.2 Tio Canela ............................................. 88

4.3 Breeding factors ......................................... 91

4.4 Access factors .......................................... 91

5. Conclusions ................................................... 92
References ...................................................... 93
CONCLUSIONS ....................................................... 95
1. Economic impact of disease-resistant bean research in Honduras .......... 95
2. Methodological contributions ..................................... 95
3. Factors influencing the adoption of RVs in Honduras ................... 97

ix

LIST OF TABLES

ESSAY ONE
Table 1. Improved Bean Varieties Released in Honduras Since 1987 ............... 9
Table 2. Characteristics of Sample Farmers, Mideast and Northeast Honduras, 2000 . . 12
Table 3. Farmer Varietal Use, Mideast and Northeast Honduras, 2000 ............. 14
Table 4. Farmer Bean Seed Source: Improved Varieties, Honduras, 2000 .......... 15
Table 5. Farmers’ Bean Prices, Mideast and Northeast Regions, Honduras, 2000 . . . . 18
Table 6. Sensitivity of returns to bean breeding research to changes in key parameters 28
Table A-1. Bean Yields by Variety, Primera 1998-2000, Honduras ............... 33
Table A-2. Bean Yields by Variety, Postrera 1998-2000, Honduras ............... 33
Table A-3. Costs of Bean Varietal Development, Honduras, 1984-1997 ............ 34
ESSAY TWO
Table 1. Hypothetical yield profiles by variety and disease pressure ............... 40
Table 2. Adoption of RVs and Bean Yields by Season, 1999-2000, Honduras ....... 49
Table 3. Characteristics of sample farmers, Honduras, 2000 ..................... 50
Table 4. Probit Estimates of Resistant Variety Adoption Equation, Primera and Postrera,
1999-2000 ...................................................... 54
Table 5. Adjusted and Unadjusted OLS Estimates of Yield Equation, Primera and
Postrera, 1999-2000 .............................................. 58
Table 6. Predictions of RV and TV bean yields from Corrected and Uncorrected Yield
Models, 1999-2000 ............................................... 60
Table 7. Rates of Return to Disease-Resistant Bean Research in Honduras, Postrera
Seasons 1982-2010 ............................................... 65
Table 6b. Predictions of RV and TV bean yields from Unconditional and Conditional
Corrected Yield Models, 1999-00 .................................... 68
ESSAY THREE
Table 1a. Characteristics of Sample Bean Farmers, Postrera 2000, Honduras ........ 76
Table 1b. Characteristics of Sample Bean Farmers, Postrera 2000, Honduras ........ 77
Table 2. Probit Estimates of Dorado Adoption, Postrera, 2000 ................... 83
Table 3. Probit Estimates of Tio Canela Adoption, Postrera, 2000 ................ 84

LIST OF FIGURES

Figure 1. Resistant Variety Adoption, Mideast and Northeast Honduras ............ 35

xi

INTRODUCTION

International agricultural research systems are currently faced with the dilemma
that while funding levels for agriculture and agricultural R&D have declined in recent
years, demands on agricultural research systems to develop agricultural technologies
which are more sustainable, equitable, and better-targeted to marginal areas have
increased. Available evidence suggests that after several decades of strong support,
international funding for agriculture and agricultural R&D began to decline in both
absolute and relative terms around the mid-19805 as support for economic infrastructure
as well as health, education, and other social services began to grow (Pardey and
Bientema, 2001).

There is an abundance of literature concerning the positive and negative impacts
on poverty, equity, and the environment in developing countries of Green Revolution
modern rice and wheat varieties (MVs) - the technologies that embody the early decades
of agricultural R&D. A recent review of 292 impact studies demonstrates that rates of
return to agricultural research remain quite high and have not fallen over time (Alston et
01., 2000). Yet, favorable rates of return alone have not alleviated continued (and
increasing) under-investment in agricultural R&D.

International agricultural research systems have switched from the “high payoff
input” approach (Hayami and Ruttan, 1985; Schultz, 1964), based on W5 and the use of
external inputs, to a broader concern for sustainable agriculture and technologies for

marginal areas, including maintenance research and reduced reliance on external inputs

(Byerlee, 1996). In fact, since the early Green Revolution MVs, a significant share of
crop improvement research (both biotech and traditional) in grains, vegetables, fruits,
etc., has shifted in focus from that of increasing yield to improving resistance to disease,
drought, and pests both to maintain yield and to enable reductions in pesticide
application. Yet, there is a much smaller body of literaturelconcerning the development
of such technologies.

These developments present a challenge for impact assessment, given the
increasing number of impact indicators demanded, and the fact that — as some argue — the
“low-hanging” fruit of agricultural research have already been picked. That is, the high-
payoff input model technologies have already been developed and extended, and the
returns to research are not likely to be as high for technologies geared towards directly
reducing poverty, inequality, and improving environmental outcomes. In addition,
generating evidence to document links between these technologies and desired outcomes
more challenging than implementing traditional impact studies.

Bean research in Central America provides an interesting example of agricultural
R&D within the context of the developments discussed above. Funding for bean research
in Central America has followed the same pattern as that of agriculture in general: the
Centro Intemacional de Agricultura Tropica (CIAT), bilateral donor-funded research
networks such as USAID Bean! Cowpea Collaborative Research Support Program
(CRSP) and the Swiss-funded Programa Cooperativo Regional de Frijol para
Centroamerica, Mexico, y El Caribe (PROFRIJOL), and NARS in Central America have

all been experienced substantial budget cuts. Yet, in Central America, beans are

predominantly produced by small, resource-poor farmers, and represent an important
source of cheap protein for rural and urban poor. Both the CRSP and the CIAT Bean
Improvement Program have set breeding for disease and drought resistance as a top
priority - both to better address the constraints of resource-poor farmers and to reduce
their dependence on pesticides and fungicides.

In the 19705, Bean Golden Yellow Mosaic Virus (BGYMV) began to spread
through Central America, threatening the production of beans, an important food crop in
the region (Morales and Anderson, 2001). The virus arrived relatively late to Honduras,
but in 1989, severe virus incidence resulted in yield losses ranging from 10 to 100 percent
(Rodriguez et al., 1994). Since the mid-19803, bean research in Honduras has focused on
developing improved bean varieties resistant to key diseases, principally BGYMV, which
has led to the release of five resistant varieties since 1990.

The three papers in this dissertation address different aspects of the impact of
disease-resistant bean research in Honduras and are each based upon a random sample
survey of Honduran bean farmers (N=210) implemented by the CRSP and PROFRIJOL
in February, 2001. The first paper presents evidence of recent adoption rates of disease-
resistant bean varieties (RVs), the farm-level benefits of RV adoption, and the ex post rate
of return to disease-resistant bean research in Honduras. The second paper addresses the
methodological challenges involved in the evaluation of a technology which maintains
rather than augments yield, and demonstrates an econometric method (borrowed from the
wage differential literature) to construct appropriate counterfactuals in the estimation of

the impact of technologies which mitigate adverse outcomes. The third paper investigates

the characteristics of farmers who have adopted disease-resistant bean varieties, and the
relative role of access (extension) and agronomic/economic (breeding) factors in their

varietal decision.

References

Alston, J ., Norton, 6., Pardy, P., 1995. Science Under Scarcity: Principles and Practices
for Agricultural Research and Priority Setting. Cornell University Press, Ithaca,
NY.

Alston, J.M., Marra, M., Pardey, P., Wyatt, T.J. 2000. Research returns redux: a meta-
analysis of the returns to agricultural R&D. The Australian Journal of
Agricultural and Resource Economics. 44: 1 85-215.

Byerlee, D. 1996. Modern Varieties, Productivity, and Sustainability: Recent Experience
and Emerging Challenges. World Development. 24:697-718.

Gary, A. 1997. USAID and Agricultural Research; Review of USAID Support for
Agricultural Research, World Bank, ESDAR.

Morales, F.J., Anderson, P.K., 2001. The Emergence and Dissemination of Whitefly-
Transmitted Geminiviruses in Latin America. Arch. Virol 146:415-441.

Pardey, P., Beintema, NM. 2001. Slow magic: Agricultural R&D a century after
Mendel. IFPRI Food Policy Report No. 31. Washington, DC.

Rodriguez, F., Diaz, 0., Escoto, D., 1994. Situacion actual del mosaico dorado del frijol
en la America Central: Honduras. In: Morales, F. J. (Ed.), El Mosaico Dorado del
Frijol: Avances de Investigacion.

Schultz, T.W. 1964. Transforming Traditional Agriculture. University of Chicago
Press, Chicago, IL.

THE ECONOMIC IMPACT OF DISEASE-RESISTANT BEANS IN HONDURAS

ESSAY ONE

1. Introduction '

In the 19705, Bean Golden Yellow Mosaic Virus (BGYMV) began to spread
through Central America, threatening the production of beans, an important food crop in
the region (Morales and Anderson, 2001). The virus arrived relatively late to Honduras,
but in 1989, severe virus incidence resulted in yield losses ranging from 10 to 100 percent
(Rodriguez et al., 1994): Since the mid-19805, bean research in Honduras has focused on
the development of improved bean varieties resistant to key diseases, principally
BGYMV, which has led to the release of five resistant varieties since 1990. The principal
objective of this paper is to estimate the ex post rate of return to disease-resistant bean
research in Honduras.

Breeding for disease resistance in beans is an example of productivity
maintenance research. Unlike productivity enhancement research, which develops
technology to increase yield given a specified level of inputs, productivity maintenance
research counteracts yield losses that result from changes in the biological or physical
environment (Smale et al. , 1998). While productivity enhancement is measured in terms
of positive yield gains associated with adoption of new technology, productivity
maintenance must be estimated in terms of the yield losses that would have occurred in
the absence of the new technology — the yield loss averted (ibid, 1998; Morris et al.,

1994).

 

' This paper is based on: Mather, D., Bemsten, R., Rosas, J .C., Viana, A.R.,
Escoto, D. 2003, forthcoming. The Economic Impact of Disease-Resistant Bean
Research in Honduras. Agricultural Economics. It includes sections from an earlier
version of the paper which was presented at the Conference on Impacts of Agricultural
Research and Development, San Jose, Costa Rica, February 4-8, 2002.

6

Various authors have used experimental trial data to estimate yield loss averted
through use of a disease-resistant cultivar (Smale eta1.,1998; Morris et al., 1994),
combined with adoption rates from surveys to estimate aggregate benefits. In contrast,
Johnson and Klass (1999) use experimental trial data to estimate yield loss averted,
develop a climate-based GIS model to predict disease incidence, and then use an expected
utility framework to predict adoption rates.

A methodological contribution of this paper is the use of a combination of
experimental trial and farm-level survey data within an expected utility framework to
estimate the farm-level benefits of RV adoption. We then use recent survey evidence on
RV adoption rates as well as the costs of varietal development to estimate the ex post rate
of return to disease-resistant bean research in Honduras.

2. Disease-resistant bean varieties

BGYMV is the principal bean disease in Honduras, and one of the main
production constraints in Honduran valleys (Martel, 1995). The virus is transmitted by
the whitefly species Bemisia tabaci, which is normally found below 1,000 m in all
growing regions of Honduras and more frequently in the drier postrera season
(September/October to December/January). Whitefly-transmitted geminiviruses cause
significant yield losses of important food and industrial crops in tropical and subtropical
agroecosystems around the world (Morales and Anderson, 2001). The exponential
increase in geminivirus-induced diseases in bean and other crops in the 19905 coincided
with the expansion in Honduran valleys of tomato and other horticultural crops

(reproductive hosts for the whitefly) valleys during the same period (ibid, 2001).

Genetic resistance, rather than pesticide use, is considered the most sustainable
means for small farmers to avoid yield losses from BGYMV, as RVs are scale-neutral
and reproducible on-farm (Martel et aI. , 2000). In contrast, pesticides can only partially
control the whitefly in the short term, are expensive for small farmers, and are often
associated with negative health effects on farm laborers. In addition, pesticides have
proven ineffective in the long term, as the whitefly often develop resistance to specific
pesticides (Morales and Anderson, 2001). Thus, compared with pesticides, RVs offer
small farmers a more financially and ecologically sustainable means to control the virus.

Honduras has two bean research programs, which are implemented by Zamorano
(Programa de Investigaciones en Frijol, Escuela Agricola Panamericana) and DICTA
(Direccion de Ciencia y Tecnologia Agricola), work in collaboration with three
organizations: the USAID-funded Bean/COWpea Collaborative Research Support Project
(CRSP); the Programa Cooperativo Regional de Frijol para Centroamerica, Mexico, y El
Caribe (PROFRIJOL), funded by COSUDE, the Swiss Agency for Development and
Cooperation; and CIAT (Centro Internacional de Agricultura Tropical). In the late 19805,
the Honduran Ministry of Natural Resources significantly decreased funding for
agricultural research and extension as a result of structural adjustment. In the wake of
these budget cuts, Zamorano became the bean breeding program in Hondurasz, while
DICTA retained its regulatory mandate, as well as some screening activities. The main
RVs developed by Zamorano in collaboration with DICTA are Dorado (released in 1990),

Don Silvio (1993), and Tio Canela (1997) (Table l).

 

2 Zamorano also coordinates bean trials for Central America.

8

Table 1. Improved Bean Varieties Released in Honduras Since 1987.

 

 

 

 

 

 

 

 

 

 

 

 

Variety Color Reaction to Diseases Year
BGYMV BCMV Rust WB CBB Released

Tio Canela small red R R S I I 1997
DICTA 122 small red R R S S I 1996
DICTA 113 small red R R S S I 1996
Don Silvio dark red R R I S I 1993
Dorado dark red T R I I I 1990
Oriente shiny red S R I I S 1990
Catrachita shiny red I R I S S 1987

 

 

 

 

 

BGYMV=Bean Golden Yellow Mosaic Virus, BCMV=Bean Common Mosaic Virus,
WB=Web Blight, CBB=Common Blight Bacteria.

R=Resistance, T=Tolerance, S= Susceptible, I= Intermediate

Source: Martel, 1995; Rosas and Varela, 1996.

 

 

A5 is the case throughout Latin America, Honduran consumers have strong
preferences for various characteristics of dry beans — color, size, shape, freshness,
cooking time, consistency when cooked, and the texture of the sauce — all of which can
influence the farm-level price of dry beans. Martel (1995) found that on average, farmers
received 16% less for Dorado (due to its dark red color), compared to traditional varieties.
Don Silvio (1993), the second RV released, is very similar to Dorado in color yet has a
shorter crop cycle than that of Dorado. However, Tio Canela (1997) was expected to
receive a higher farm-level price than Dorado, due to its lighter-red color (Rosas and

Varela, 1996).

 

3. Survey methodology
3.] Previous research

In 1993, the Bean/Cowpea CRSP funded a survey of a random sample of bean
farmers (N=239) in the two main bean-growing regions of Honduras (Mideast and
Northeast), as well as a survey of traders (N =5 7) in eight different major Honduran
markets (Martel, 1995). The surveys documented adoption rates of improved bean
varieties, the socioeconomic characteristic of adopters,the relative farm-level prices of
different varieties, and farmers’ preferences in varietal selection. This adoption survey
and a complementary subsector analysis, which provided increased evidence of the
importance of socioeconomic and market factors in farmer adoption of improved
varieties, helped Zamorano and DICTA set future breeding priorities.

In 1996, PROFRIJOL and DICTA funded a survey of bean farmers (N=160) in
Mideastern Honduras (Viana, Rodriguez, and Escoto, 1997). This study reported
adoption rates for Dorado that were quite high3 relative to those of Martel, possibly due to
the heavy concentration of the sample in areas targeted by the National Bean Program
(DICTA).

Drawing on the empirical results of these two studies, a study of BGYMV
incidence (Morales, 1994), as well as experimental trial data, Johnson and Klass (1999)
used a climate-based GIS statistical model to predict (map) BGYMV incidence over time

in order to estimate the magnitude of the production losses that were averted as a result of

 

3 For the Mideast region, Martel reported an adoption rate for Dorado of 27% in
1993, whereas Viana et al. reported an adoption rate of 50% in the same region and year.

10

BGYMV-resistant varieties. In addition, they used GIS to document a link between RV
adoption and poverty alleviation (as reported by a recent poverty survey).
3.2 Current research

In January-February 2001, the CRSP and PROFRIJOL implemented a random
sample survey of Honduran bean farmers (N =2 1 0), in collaboration with Zamorano and
DICTA. This survey was designed to generate data required to carry out an ex post
economic impact assessment of bean research in Honduras, as well to provide Zamorano
and DICTA with information about the characteristics of adopters and disadopters of
improved bean varieties, and farmers’ experience and opinions regarding various
agronomic, market, and consumption aspects of the improved and traditional varieties
that they planted. More specifically, the survey was designed to estimate the adoption
rates of RVs, as well as their yield and price performance relative to (TVs). The survey
targeted the Mideast (El Paraiso and Francisco Morazan departments) and Northeast
(Olancho) regions, which together account for about one-half of annual bean production

in Honduras.4

 

4 In each of the three departments, 70 farmers were selected using the following
methodology. Using bean area and production data from the 1993 Agricultural Census, a
list of villages was constructed for each department which represented a cumulative of 80
percent of bean area in that department. This list was then divided into deciles by
cumulative bean area, and one village from each decile was selected at random. Care was
taken to ensure that the cumulative number of villages selected from a given municipality
did not exceed its share of the Department’s total bean area. Selected villages beyond a
municipality’s share were replaced by the next random selection within the decile. Thus,
this sample selection method was constructed to focus upon “area” adoption rates for
benefit-cost analysis, yet to ensure that village selection remained representative of the
department’s bean area in terms of village geography and size.

11

The characteristics of our respondents (Table 2) are quite similar to those in
Martel’s 1993 survey (in the same two regions), suggesting that our sample is
representative of bean farmers in these two regionss. Bean producers in Honduras are
predominantly small farmers, the majority of whom use fertilizer and insecticide, market
at least some of their bean production, and depend upon beans as a significant important
portion of their household income (Martel, 1995).

Table 2. Characteristics of Sample Farmers, Mideast/Northeast Honduras, 2000

 

 

 

 

 

 

 

 

 

 

 

Characteristic Farmer Average, 2000
Primera Postrera

Bean area (ha) 1.52 2.13
Average yield (kg/ha) 652 424
Applied Compound Fertilizer and/or Urea (% farmers) 74 68
Applied Insecticide (% farmers) 73 65
Sold beans harvested from this season (% farmers) 75 63
Of those who sold beans, % of production sold (%) 65 80
Sample size (N = 170) (N = 202)
Source: CRSP/PROFRIJOL Bean Farmer Survey, January-February 2001.

 

 

4. Adoption of disease-resistant varieties
Honduran farmers typically plant two crops during the year. In the primera

(May/June to July/August), the rainy season, maize is the principal crop, and beans are

 

5 Our respondents’ bean area is similar to those of Martel’s 1993 sample, yet our
respondents have a stronger market orientation and are more likely to use chemical
inputs. These differences are likely due to the continuing commercialization of bean
production in these two regions, as well as differences in sampling strategy; Martel
stratified by valley vs. hillside farmers, while we stratified only by geographic location.

12

either intercropped with maize or monocropped". The postrera (September/October to
December/January) is a drier season, during which beans are almost exclusively
monocropped7. The primera accounts for approximately 33% of annual national bean
production, while the postrera accounts for 67% (1995-99 average; SAG, 1999). Because
the whitefly population is often highest under drier conditions, BGYMV is most frequent
and severe in the postrera.

In Primera 2000, 46 % of the respondents planted 3 RV, 62 % planted a TV
(Table 3), and 13 % of the farmers grew both a RV and a TV. Typically, farmers who
grow both an RV and a TV plant a small proportion to a TV, which is intended for home
consumption — due to preferred consumption and culinary characteristics - and a larger
area to an RV, intended for the market.

In Postrera 2000, 41 % of the respondents planted a RV, 76 % of farmers planted
a TV, and 22 % planted both a RV and a TV (Table 2). While BGYMV pressure is
typically greater in the postrera, the lower RV adoption rate in this season is likely related
to the increase in sample size, the lower rate of adoption in the Northeast region (in which
beans are principally grown in the postrera alone), and the fact that Dorado has a longer
crop cycle than preferred by farmers, especially in the drought-prone postrera (Martel,

1995).

 

6 In the following analysis, intercropped area is converted to “effective monocrop
bean area” given information from each respondent on his/her intercropping system. For
example, a farmer with an intercrop of four rows of beans to each row of maize is
assumed to have 80% “effective” area in beans for use in yield calculations.

7 While farmers in Olancho use relay cropping for beans in the postrera, for the
purposes of this paper, we assume the area to be the same as for monocropping.

13

Table 3. Farmer Varietal Use, Mideast and Northeast Honduras, 2000

 

 

 

 

 

 

 

 

 

 

Variety Primera Postrera
(% respondents) (% respondents)

Tio Canela“ 10.6 11.4
Dorado“ 30.0 25.7
Don Silvio“ 4.1 4.0
Total Resistant 46.2" 41.1"
Total Traditional 62.4" 76.2* *
Sample Size N=170 N=202
Source: CRSP/PROFRIJOL Farmer Survey, 2001 (N=210)
* BGYMV-resistant variety
"”" As some farmers planted more than one variety, the total is greater than 100%.

 

Defining small farmers as those having < 3.5 ha of total farm area, RV adoption in
2000 was neutral with respect to farm-size. Resistant variety adoption was also neutral
with respect to market orientation (for both primera and postrera 2000), defining market
orientation by whether or not each respondent sold beans in a given season.8 In addition,
the value of on-farm reproducibility of RVs is demonstrated by the informal nature of the
bean seed system: of the RV seed planted by our respondents in 2000, 59% was farmer-
saved seed; 27% was obtained from a neighbor; and only 14% was obtained from

government extension, an NGO, a trader, or an input supplier (Table 4).

 

8 However, probit results from essay #3 of the factors influencing adoption
indicate that market orientation is not neutral with respect to Dorado and Don Silvio; in
fact, market participation is a statistically significant (10% level) and negative factor in
use of these older RVs. The neutral result here is likely due to the fact that market
participation is not a statistically significant factor in Tio Canela use, and the test in the
current paper combines all RV users together.

14

However, RVs are not adopted neutrally with respect to the production
environment. Using altitude, as well as historical knowledge of BGYMV-prone areas as
a proxy for BGYMV pressure (Rosas, 2001), there is a statistical difference (a=0.04)
between the adoption of RVs in “BGYMV-prone” areas (typically with altitude <1 ,000

Table 4. Farmer Bean Seed Source: Improved Varieties. Honduras, 2000

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Seed Source Seed (Germplasm) Planted in Original Source of this
Postrera 2000 (%) Variety

(%)
Farmer-Saved Seed 59 na
Neighbors 27 57
Traders 6 10
DICTA / SAG 0 l6
Zamorano 3 3
CIAT Seeds of Hope 0 2
NGO 0 0
Input Dealer 2 4
Local Market 0 3
Artisan Seed 0 0
Other 2 5
Source: CRSP/PROFRIJOL Farmer Survey, 2001

 

 

 

m) and “non-BGYMV-prone” areas in both primera and postrera 2000. However, as
farmers in the “non-BGYMV” areas (mountains) often live in quite remote communities,
it is unlikely that they have the same level of access to information about RVs and the

seed itself, as do farmers in the valleys. Zamorano/DICTA have targeted higher altitude

15

farmers with Catrachita", a non-resistant, high-yielding variety released in 1987.
5. Estimation of the farm-level impact of improved bean varieties
5.1 Limitations of experimental trial and survey data

Conventional ex post impact assessment methods have typically focused on
productivity enhancement technologies, whose benefits are estimated as the yield
difference between traditional and improved technologies as observed in either
experimental trials or a cross-section of survey farmers. However, both of these data
sources have limitations in constructing the counterfactual situation used to assess the
farm-level benefits of productivity maintenance technologies. First, experimental trials
often do not well-approximate farmer conditions, especially considering that disease
frequency and intensity are fixed in experimental trials, yet under farmer conditions they
may vary spatially by weather patterns, altitude, and crop management practices. Second,
the presence of selection bias in farmer survey yield data will tend to underestimate the
real benefits of disease-resistant technologies (Johnson and Klass, 1999). Because
farmers in areas of low disease incidence are likely to grow TVs, and farmers in areas of
high disease incidence are likely to grow RVs, then observed survey yields of TVs will be
higher than what would have been observed in the absence of RVs. Thus, Johnson and
Klass (1999) argue that the appropriate comparison is between the yields of TVs and RVs

under disease pressure in experimental trials, which control for sample selection bias.

 

9 Analysis of the adoption and impact of Catrachita is presented elsewhere
(Mather et al. , forthcoming).

l6

Unlike previous research, we use a combination of experimental trial and farm-
level survey data within an expected utility framework to estimate the farm-level benefits
of RV adoption. Given that RVs historically have received price discounts in the market
(Martel, 1995), gains in profitability due to yield loss averted are tempered by losses in
profitability due to the RV’s poorer market characteristics. In an expected utility
framework, we use survey data on varietal price discounts and disease frequency,
combined with a range of estimates of yield loss averted from experimental trials, to
compute the equivalent income that RV adopters gain due to avoiding the risk of yield
losses to TVs under disease pressure.

5.2 Farm-level price discounts

Given the dispersion of farm-level bean sale prices over time and space,
regression analysis was used to control for time of sale, region, and remote areas by
season. The results of various specifications show that the Dorado and Don Silvio price
discount (relative to all TVs) for the primera 2000 (N=147) is in the range of 15% to
20%, while for postrera 2000 (N=147) the discount is in the 10% to 15% range (Table 5).
The Tio Canela price differential with respect to TVs is not significant in any of the
model specifications, perhaps due to small sample size (N=23). We thus compare the
sample means of Tio Canela and TVs, and find a 4% discount for Tio Canela in primera
2000 and a 9% discount in postrera 2000. Given these results, we take the average
between the primera and postrera for each variety and thus assume a 16% price discount

for Dorado and Don Silvio, and a 7% discount for Tio Canela.

17

Table 5. Farmers’ Bean Prices, Mideast and Northeast Regions, Honduras, 2000

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Primera 2000 Postrera 2000
Variety _ .
Farmer N CV % price Farmer N CV % price
Price (%) discount Price (%) discount
($US/kg) for MV" ($US/kg) for MV
Tio Canela $0.47 1 1 31 - 4.1 $0.47 10 20 - 9.6
Dorado $0.38 41 26 - 22.4 $0.46 27 27 - 11.5
Don Silvio $0.32 4 14 - 34.7 $0.47 5 20 - 9.6
Catrachita $0.52 9 32 6.1 $0.57 5 17 9.6
All Traditional $0.49 80 27 na $0.53 98 23 na
Varieties
All Varieties $0.46 147 30 na $0.51 147 25 na

 

 

Source: CRSP/PROFRIJOL Farmer Survey, 2001
* Compared with TVs
CV = coefficient of variation, na = not applicable

 

5. 3 Experimental and survey yields

In yield performance trials conducted on farmers’ fields (N=53) across

Honduras”, Tio Canela averaged 1,200 kg/ha, compared to an average of 850 kg/ha for

the local check—which represents a yield improvement of 41% above the local check

variety (Rosas and Varela, 1996). In experimental trials under severe BGYMV pressure,

RVs have yielded an average of 50% more than TVs (Rosas, 2001). In farmers’ trials,

Dorado yielded 20% more than the local check, and 50% more than the TV under

BGYMV pressure in experimental trials.

 

‘0 Farmer trials included a local check, typically a TV, and the farmers’ choice of
fertilizer application rate and management practice (the same for both Tio Canela and the

TV).

18

 

However, while the mean RV yield was typically higher than that for TVs, the
coefficient of variation (CV) in yield for all varieties in all years was high (ranging from
63 to 112%) (see Appendix Tables A-1 and A-2). Therefore, we could not find evidence
that RVs have statistically higher yields than TVs under farmer conditions. Multi-variate
regression analysis was used to test for yield differentials between RVs and TVs,
controlling for numerous factors besides variety which influence yield (fertilizer use,
cropping system, altitude, season, region, etc). However, yield regressions using OLS
also did not support the hypothesis that RVs are higher-yielding than TVs, most likely
because endogeneity of varietal choice may lead to biased estimates of the variety dummy
coefficient. Testing for and correcting selection bias econometrically is beyond the scope
of this paper. Computing intra-farrn yield differentials between RVs and TVs
demonstrated some evidence of RV yield gains (7 to 8% for some seasons). However,
the number of farmers growing both varieties was small in all seasons (N <30), and this
method is still subject to potential selection bias which could underestimate yield loss
averted. While the observed performance of any technology under farmer conditions
typically falls short of the experimental trial results, the fact that the RV survey yields and
coefficients of variation are quite similar to those of TVs — given that price discounts
exist, most farmers market beans, and adoption rates are fairly high — suggests that

selection bias is present.”

 

" Assuming that we only observe TV yields in areas of low disease pressure, then
we won’t observe low TV yields that would result from growing TVs in disease intense
areas. Thus, the observed TV mean is higher than the true TV mean, because the TV
yield distribution is truncated from below.

19

5.4 Disease frequency

To obtain a farm-level estimate of disease frequency, we showed respondents in
Francisco Morazan and Olancho pictures of the six principal bean diseases in Honduras
(without corresponding names), and asked them to identify the two diseases which have
caused the most damage to their bean crop in the past five years. Thirty-one percent of
the respondents listed the BGYMV picture as one of their two principal diseases”. Those
who indicate problems with a disease were then asked in which season (primera, postrera,
or both) the disease was most frequent, and how many times it had occurred in their fields
in the past five years. One-half of these respondents claimed that BGYMV was a
problem in both seasons, and 27 % said only in the postrera. The average frequency
reported by these respondents was four out of five years. For our base scenario, we
conservatively assume that BGYMV only occurs in the postrera, although we include
sensitivity analysis to test this assumption.
5.5 Expected utility framework

We assume that farmers maximize profit and have risk preferences as defined by
Constant Relative Risk Aversion with a risk coefficient R=1 .1 This assumption is not very
strong, given that relative risk coefficients over income for developing country farmers

typically fall between 0.3 and 1.7 (Binswanger, 1980), and that this coefficient is often

 

'2 It should be noted RV growers are unlikely to see symptoms of BGYMV on
their bean plants, even though the virus may be present (though it’s presence in a RV
does not result in yield 1055). As we would expect, of those respondents who claimed that
the virus was one of their two principal disease problems, only 33 % grew RVs.
However, this suggests that even in areas with lower disease pressure (where TVs are still
grown), the virus is present in four out of five years.

20

assumed to be unity (Newberry and Stigliz, 1981). Valuation of the benefits of risk
reduction typically involves total household income, as the household may have a
portfolio of agricultural and non-agricultural activities, some of which are uncorrelated
with the returns to the risky income component in question (Walker, 1989). As we do not
have data on total household income, we only include bean income in this equation, and
thus use a rather low risk coefficient.

We use the mean value from experimental trials (RVs yield on average 50% more
than TVs under severe BGYMV stress) as an upper bound for “yield loss avoided.” We
assume that TV yield loss to BGYMV is 0%, 15%, 25%, 35%, or 50%, each with a
probability of 0.20. '3 Thus, we assume that disease frequency is four out of five years,
based upon the survey data, and that the average yield loss is 25 %." We solve the
following equation for F to estimate what adopters are willing to pay to avoid the yield
risk of the TV:

1n [(1 - F)Yo] = 0.2 In [Yo] + 0.2 In [0.85Yo] + 0.2 In [0.75Yo] +
+ 0.2 In [0.65Y0] + 0.2 In [0.5Y0]

where: Y0 represents bean income, and F represents the value to farmers of an RV
in terms of a percentage of bean income

 

'3 We assume that these yield loss amounts and frequencies are constant over
time, which implies that BGYMV pressure remains constant.

'4 The yield losses and frequencies are chosen based on the farmer survey results
that disease losses occur 80% of the time, that breeders found that TVs experienced yields
50% below those of RVs in experimental trials under conditions of severe incidence, and
so they result in an expected yield loss of 25% that is arbitrarily set to one-half that of
experimental trials. Other yield loss scenarios also meet these criteria. In the absence of
survey data on disease intensity and yield loss at the farm-level, we thus make
conservative assumptions of yield loss to disease.

21

The left-hand side of the equation represents the expected value of an RV, while
the right-hand side represents the expected value of a TV, given our assumptions of yield
losses and frequencies. This equation assumes that the yield of an RV and a TV are equal
in the absence of disease pressure, and that each receives the same price. Solving for F
yields 0.27, which indicates that a risk-averse farmer would be willing to pay up to 27%
of bean income to avoid the risk represented by the TV. Thus, given that Dorado receives
a 16% price discount, this implies that Dorado adopters gain the equivalent of 11% in net
bean income due to the yield loss averted. Because Tio Canela’s price discount is only
7%, Tio Canela adopters gain the equivalent of 20% in net bean income due to the yield
loss averted. Of the 27% benefit, 2% can be attributed to our assumed nature of risk
(relative) and the coefficient of risk-aversion (one).

5. 6 Incremental farm-level costs

For the following benefit-cost analysis, we assume that RV adopters incur no
incremental costs per hectare. The vast majority of adopters obtained their seed from
other farmers (uncertified seed), and even if they pay their neighbor a markup above what
they would pay for a TV, the cost of seed is not a large proportion of input costs
(Tshering, 2002). Although RV adopters may be able to save both labor and input costs
from reduced pesticide applications, the farm-level health and financial benefits from

reduced pesticide use associated with RV adoption are not considered here.

22

6. Economic surplus analysis
6.] Aggregate benefits

A small open-economy economic surplus model is used (Alston et al., 1995), to
estimate the downward shift of the supply curve. In this model, the supply curve is linear,
and its shift is parallel, which is a reasonable assumption given that RV adoption is scale-
neutral. The yield gain associated with RVs comes from the farm-level benefit of yield
loss and risk avoided, with the price discount already deducted. No market price effects
are assumed in this model, as incremental volumes of production are assumed to be
exported (El Salvador imports Honduran beans regularly). The choice between an open-
or closed-economy model had a small effect on the total surplus generated, based on
sensitivity analysis of our model and data.

Area adoption curves were constructed for each RV using a logistic function fit to
data points from the Martel’s (1995) survey in 1993 and our 2001 surveys in the Mideast
and Northeast regions — weighting the annual RV bean area by the seasonal and regional
shares of the annual total of the two regions from the year of varietal release until 2010
(Figure 1; see Appendix). The area adoption curves are quite similar to “farmer”
adoption curves, given the relative homogeneity of farm-size in the sample.

For Tio Canela, we have only four years of farmer adoption behavior, as the
variety was released in 1997. However, it is clear that it is quickly being adopted-~both
by former growers of Dorado or Don Silvio and by TV users. Therefore, we assume that
its potential ceiling is equal to the area planted to Dorado in 1998 (37% in the primera,

31% in the postrera for the Mideast; and 34% in the primera, 22% in the postrera for the

23

Northeast)”. Thus, we project that Tio Canela will replace Dorado and Don Silvio by
2007, given Tio Canela’s price advantage, relative to both of these varieties. For Don
Silvio, we assume that the 2000 adoption level of 4% (average of primera and postrera
2000) is its ceiling in each region, and that growers gradually disadopt (switching to Tio
Canela) until the rate falls to 0% in 2007.

The discount rate is assumed to be 10%, and the supply elasticity 0.7, given that
the short-run and intermediate supply responses of a semi-subsistence crop are generally
assumed to be inelastic. The model also uses historic data on bean production by season
and region from 1987-1999 (SAG, 1999). We assume that future (2000-2010) production
levels by season and region will be the same as the 1996-1999 average levels (excluding
postrera 1998 due to Hurricane Mitch). The bean price series used for economic analysis
is the Honduras farmgate price from 1987-1999. We assume that bean prices remain
constant until 2010 at the 1999 level, which was among the lowest in the past decade.
We use the farmgate price series rather than an import parity price, given that the Mideast
and Northeast regions are close to the main export market (El Salvador and Nicaragua, a
shadow market, as per Martel (1995)). Thus, we assume that the farmgate price is the
best approximation to export opportunities.
6.2 Research costs

Each of the improved varieties was developed, tested, released, and extended by

Zamorano and DICTA, supported by the Bean/Cowpea CRSP, PROF RIJOL, and CIAT.

 

'5 Given that the Post-Hurricane Mitch seed distribution could have influenced

adoption rates in 1999-2000, we use 1998 as a year more representative of the typical
demand for RVs.

24

The CRSP and PROFRIJOL and have provided funding and training, while CIAT has
provided germplasm, training, and collaboration with both DICTA and Zamorano in
conducting regional field trials. Costs associated with the development of the RVs
(Appendix Table A-3) begin in 1984 (the initial developmental stages of Dorado) and
continue to 1997 (when Tio Canela was released).

CIAT’s role in the development of Dorado (the cross, nursery, and on-station trial
stages from 1984 to 1986) is approximated by estimating the cost per released variety of
CIAT’s bean program (during the 1980-85 period)”. The CRSP supports the Zamorano
bean program’s research activities, as well as training. This analysis assumes that 60% of
the CRSP support to the Zamorano bean program budget and 60% of training investments
during the 1986-1997 period can be attributed to the development and dissemination of
RVs. Financial support from PROFRIJOL to Zamorano’s bean program is included at
the 100% level. Zamorano’s variable cost contribution to the bean program — 50% of the
director’s salary (research) — is included, while the fixed costs of buildings and the other
50% of the director’s salary (teaching) are not included.

DICTA’s national bean program began its work with Dorado in multi-locational
trials in 1986 and 1987. We include 100% of DICTA’s estimated bean research and
extension costs for the Danli and Olancho experiment stations during the period from
1986-1997. Financial support from PROFRIJOL to DICTA is also included at the 100%

level.

 

'6 CIAT varietal development costs for this time period provided by Nancy
Johnson, CIAT Economist.

25

Since structural adjustment in the late-19805, improved bean varieties have
primarily been extended through DICTA’s artisan seed projects, Zamorano-based pilot
projects, the sole private commercial seed supplier (Hondugenet, located in the capital,
Tegucigalpa), commercial sales of certified seed by Zamorano, and governmental and
non-govemmental agricultural development projects, which typically obtain their seed
from Zamorano. Various NGOs distribute improved seed to participants in their projects
and support the development of local farmer seed banks across the country, although they
are less active in the Mideast and Northeast than in other regions. The most prominent
seed initiative has been the USAID-funded Honduras Post-Mitch Agricultural Recovery
program, which multiplied and distributed improved bean seed via NGOs to farmers
across the country, following hurricane Mitch in October, 1998. However, the NGOs
distributing the seed focused their efforts on other regions of Honduras, which were hit
harder (and are generally poorer) than Mideast and Northeast Honduras.

The extension efforts of NGOs and the Post-Mitch project are not included as
costs in this analysis for two reasons. First, project records do not indicate how much
seed was delivered to each region, although it is clear that the general location of the
NGOs’ operations prior to Mitch placed a minor emphasis on the two regions of analysis.
Second, only 9% of our sample received seed directly from an NGO following Mitch, and
we do not know how many other farmers may have received seed indirectly.

6.3 Rate of return
For our base scenario, we assume that BGYMV causes yield losses to TVs only

during the postrera season. Under this assumption, and those discussed above regarding

26

the farm-level benefits of RV adoption and RV adoption rates, the ex post economic rate
of return to breeding disease-resistant beans in Honduras during the period 1984-2010 is
41.2 %, and the Net Present Value at a discount rate of 10% is US$296 million.

This analysis provides strong evidence that the return to disease-resistant bean
research in Honduras has been profitable in aggregate. We have assumed that average '
yield losses to BGYMV are one-half that of experimental trials, and that the current rate
of RV adoption is the ceiling rate. In addition, we have assumed near-maximum variable
costs of bean varietal development and dissemination. Furthermore, although data are not
available to document the spillover impact in non-surveyed areas, key informants report
that many farmers in other regions of Honduras, as well as in El Salvador and Nicaragua,
have adOpted both Tio Canela and Dorado. In addition, during the period of interest,
Zamorano developed additional varieties (Catrachita in Honduras; Bribri in Costa Rica),
for which benefits associated with their release are not considered here. Thus, this
analysis charges Zamorano and DICTA’s bean research expenditures against only the
RVs that have been released in Honduras.

6. 4 Sensitivity analysis

Sensitivity analysis demonstrates that the key parameters in the rate of return
analysis relate to disease pressure by season, elasticity of supply, and the average
expected yield loss to BGYMV (Table 6). For example, assuming that BGYMV causes
yield losses to TVs during both the primera and postrera seasons, the economic rate of
return increases to 48.1 %, and the Net Present Value at a discount rate of 10% increases

to US$462 million. The actual situation is likely somewhere between these two

27

scenarios, as BGYMV is known to be a problem in the primera, but with less frequency
than that of the postrera. With respect to elasticity of supply, assuming a higher elasticity
of 1.0 % reduces the IRR to 35 %, whereas a lower elasticity of 0.4 % increases the IRR
to 55 %. However, we use 0.7 % in the baseline scenario because semi-subsistence crops
are generally assumed to have an inelastic elasticity of supply. The high sensitivity of our

Table 6. Sensitivity of returns to bean breeding research to changes in key

 

 

 

 

 

 

 

 

 

 

 

 

 

arameters

Elasticity of Supply: IRR Net Farm-level Value of Yield Loss IRR
(%) Averted as % of Total Bean Income (%) (%)

1.0 % 34 Dorado/D.Silvio 6%; Tio Canela 15% 29

0.7 % * 41 Dorado/D.Silvio 11%; Tio Canela 20% * 41

0.4 % 55 Dorado/D.Silvio 16%; Tio Canela 25% 50

BGYMV pressure by IR Dorado Area in 2000 Replaced by IRR

season: (%) Tio Canela by 2007? (%)

Postrera only * 41 Yes * 41

Primera and Postrera 48 No, area of both remain at year 2000 level 41

Discounted Benefits IRR

Extended to: (Year) (%)

2000 40

2005 41

2010 * 41

* Value used for the baseline scenario. Source: calculated by the authors

 

 

 

IRR to the assmned elasticity of supply concurs with the recent work of Oehmke and
Crawford (2002). With respect to farm-level benefits, changes in the net farm-level value
of yield loss averted would occur if we change our assumptions of either the average

expected yield loss to BGYMV (itself a function of disease intensity and/or frequency) or

28

the price discount by variety. Even at a considerably lower net farm-level value for
Dorado and Tio Canela, the IRR of 29% is still quite high. We also do sensitivity
analysis on the extension of benefits over time to examine the scenario in which the net
value of yield loss averted by RVs decreases in the future due to either lower whitefly
populations or diminished genetic resistance to the virus over time. We also test the
implication of our assumption that Tio Canela (which has a higher net-farm-level value)
completely replaces Dorado by 2007. Analysis of both these sets of assumptions shows
that the rate of return is not very sensitive to assumptions beyond the year 2000.
6.5 Distribution of benefits

The distribution of benefits appears to be widespread across producers, as even in
the valleys of Honduras, beans are grown predominantly by farmers with less than three
hectares of land, and RV adoption is neutral with respect to farm-size and market
orientation. However, the distribution by growing environment is not as equitable. It is
clear from the widespread adoption of RVs and the declining use of Catrachita (a non-
resistant improved variety targeted to mountain farmers) that farmers in lower altitudes
have gained the most from bean research since the mid-19805 (Mather et al. ,
forthcoming). However, Zamorano’s current breeding priorities are focused on
developing varieties with improved drought tolerance — one of the principal production
constraints of high-altitude bean farmers — as well as varieties with improved heat-
tolerance (especially for the lowlands on the Atlantic Coast). In addition, Zamorano is

continuing to improve the market acceptance traits of their disease-resistant lines.

29

7. Conclusion

This paper provides evidence that investment in breeding disease-resistant beans
in Honduras has been profitable. Under the base scenario assumptions, the ex post
economic rate of return to disease-resistant bean research in Honduras during the period
1984-2010 is 41.2 %, well above the assumed 10% opportunity cost of capital.
Moreover, the adoption of RVs in Honduras has been widespread and neutral with respect
to farm-size. This paper also demonstrates an approach that uses farm-level and
experimental trial data within an expected utility framework to measure the farm-level

benefits of RV adoption.

References

Alston, J ., Norton, 6., Pardy, P., 1995. Science Under Scarcity: Principles and Practices
for Agricultural Research and Priority Setting. Cornell University Press, Ithaca,
NY.

Binswanger, H., 1980. Attitudes towards Risk: Experimental measurement evidence
from rural India. American Journal of Agricultural Economics 71 :774-84.

Johnson, N., Klass, J ., 1999. The Impact of Crop Improvement Research on Rural
Poverty: A Spatial Analysis of BGYMV-Resistant Bean Varieties in Honduras.
Paper prepared for the workshop: Assessing the Impact of Agricultural Research
on Poverty Alleviation, San Jose, Costa Rica, September 14-16, 1999.

Mather, D., Bernsten, R., Rosas, J.C., Viana, A., Escoto, D. Forthcoming. The
Economic Impact of Bean Research in Honduras. Staff Paper. Department of
Agricultural Economics, Michigan State University, East Lansing, MI.

Martel, P., Bernsten, R., Weber, M., 2000. Food Markets, Technology, and the Case of
Honduran Dry Beans. Michigan State University International Development
Papers, Working Paper No.78. Department of Agricultural Economics, Michigan
State University, East Lansing, MI.

30

Martel, P., 1995. A Socio-Economic Study of the Honduran Bean Subsector: Production
Characteristics, Adoption of Improved Varieties, and Policy Implications. Ph.D.
Dissertation, Department of Agricultural Economics, Michigan State University,
East Lansing, MI.

Morris, M.L., Dunin, H.J., Pokhrel, T., 1994. Returns to wheat breeding research in
Nepal. Agric Econ 10:269-282.

Morales, F.J., Anderson, P.K., 2001. The Emergence and Dissemination of Whitefly-
Transmitted Geminiviruses in Latin America. Arch. Virol 146:415-441.

Newberry, D.M., Stigiz, J .E., 1981. The Theory of Commodity Price Stabilization: A
Study of the Economics of Risk. Oxford Clarendon Press.

Oehmke, J .F ., Crawford, E.W. 2002. Sensitivity of Returns to Research Calculations to
Supply Elasticity. American Journal of Agricultural Economics 84:366-69.

Rodriguez, F., Diaz, 0., Escoto, D., 1994. Situacion actual del mosaico dorado del fiijol
en la America Central: Honduras. In: Morales, F. J. (Ed.), El Mosaico Dorado del
Frijol: Avances de Investigacion.

Rosas, J .C., 2001. Personal communication. Director, Programa de Investigaciones en
F rijol, Escuela Agricola Panamericana, Zamorano, Honduras.

Rosas, J .C., Varela, 0., 1996. Propuesta de Liberacion de la Nueva Variedad de Frijol
Tio Canela-75. Programa de Investigaciones en F rijol, Escuela Agricola
Panamericana.

Secretaria de Agricultura y Ganaderia (SAG). 1999. Compendio Estadistico.
Departamento de Inforrnacion Agricola, Unidad de Plantearniento y Evaluacion de
Gestion, Tegucigalpa, Honduras.

Smale, M., Singh, R.P., Sayre, K., Pingali, F., Rajaram, S., Dubin, H.J., 1998. Estimating
the economic impact of breeding nonspecific resistance to leaf rust in modern
bread wheats. Plant Dis. 82:1005-61.

Tshering, C., 2002. Profitability Analysis of Bean Production in Honduras. M.Sc.

Thesis, Department of Agricultural Economics, Michigan State University, East
Lansing, MI.

3]

Viana, Abelardo, Federico Rodriquez and Danilo Escoto, 1997, “Adopcion de la
Variedad Dorado Region Centro-Oriental de Honduras,” Direccion de Ciencia y
Tecnologia Agricola, Programa Regional de Investigacion en Frijol (DICTA-
PROF RIJOL), Honduras.

Walker, T., 1989. Yield and Household Income Variability. In: Anderson, J ., Hazel], P.
(Eds), Variability in Grain Yields: Implications for Agricultural Research and

Policy in Developing Countries. The Johns Hopkins University Press, Baltimore,
pp. 309-319.

32

Appendix

Table A-1. Bean Yields by Variety, Primera 1998-2000, Honduras

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Primera 2000 Primera 1999 Primera 1998
Variety Yield CV N Yield CV N Yield CV N
(kg/ha) (%) (kg/ha) (%) (kg/ha) (%)

Tio Canela 869 73 18 753 75 12 575 l 13 5
Dorado 724 66 50 923 63 50 785 72 49
Don Silvio 556 80 7 561 70 7 285 51 2
Catrachita 458 77 l l 436 77 11 588 94 18
All TVs 632 65 104 701 62 102 595 84 89
Source: CRSP/PROFRIJOL Bean Farmer Survey, 2001

CV=coefficient of variation (standard deviation divided by the mean)

 

 

Table A-2. Bean Yields by Variety, Postrera 1998-2000, Honduras

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Postrera 2000 Postrera 1999 Postrera 1998

Variety Yield CV N Yield CV N Yield CV N

(kg/ha) (%) (kg/ha) (%) (kg/ha) (%)
Tio Canela 537 93 23 791 53 14 190 200 7
Dorado 369 92 52 624 90 58 268 139 55
Don Silvio 552 74 8 796 46 8 389 141 2
Catrachita 21 1 68 10 483 80 12 165 176 19
All TVs 446 80 151 612 63 133 239 118 124
Source: CRSP/PROFRIJOL Bean Farmer Survey, 2001
CV=coefficient of variation (standard deviation divided by the mean)

 

 

33

 

Table A-3: Costs of Improved Bean Varietal Development, Honduras, 1984-1997

 

EAP a EAP training 0 PROFRIJOL° DICTA“ CIAT° TOTAL
Year (sue) ($05) ($06) (sue) ($05) ‘ (sue)
‘W 11 , ,
1985 112,000 112,000
1986 33,985 8,022 22,314 41,407 112,000 217,728
1987 48,379 13,101 15,000 35,387 111,867
1988 59,153 19,578 15,000 56,830 150,558
1989 66,662 24,767 15,000 45,525 151,954
1990 63,777 18,346 15,000 45,559 142,682
1991 85,486 20,817 15,000 50,562 171,864
1992 64,270 22,864 15,000 48,331 150,465
1993 83,776 12,093 15,000 23,354 134,223
1994 76,696 10,282 15,000 19,009 120,986
1995 66,339 0 15,000 18,469 99,808
1996 60,181 0 15,000 17,959 93,140
1997 65,294 0 15,000 17,499 97,793

 

" Represents 60% of CRSP contribution to EAP bean program for research;
includes $30,000/year for breeder‘s salary
° Represents 60% of CRSP funding of EAP bean program for training;
c Represents 100% PROFRIJOL assistance to EAP and DICTA bean programs
‘1 Represents 100% of DICTA's bean research and extension costs for Danli and Olancho
° Estimate of the cost of CIAT's bean improvement program from 1980-85
per released variety

 

34

 

 

 

Cumulate % area planted to RV

20%

10%

40%

Fig. 1: Resistant Variety Adoption
Mideast & Northeast Honduras

 

30%

 

 

 

 

C3
o\
I

/r '\
,-
‘l‘ /
o”
/
1
p
/
- /
f
/
/
7
/
1 . v ><7i——x
1987 1989 1991 1993 1995 1997 1999
Year

 

 

 

—o— Dorado —x— Tio Canela—x— Don Silvio

 

 

35

 

 

IMPACT ASSESSMENT OF MAINTENANCE RESEARCH TECHNOLOGIES: THE
CASE OF DISEASE-RESISTANT BEANS IN HONDURAS

ESSAY TWO

36

1. Introduction

Conventional ex post impact assessment methods have typically focused on
productivity enhancement technologies, whose benefits are estimated as the yield
difference between traditional and improved technologies as observed in either
experimental trials or a cross-section survey of farmers. However, both of these data
sources have limitations in constructing the counterfactual situation used to assess the
farm-level benefits of productivity maintenance technologies. Various authors have used
experimental trial data to estimate yield loss averted through use of a disease-resistant
cultivar (Smale et a1 1998, Morris et a1, 1994), but experimental trials often do not well-
approximate farmer conditions. This is especially true for measuring disease resistance
technologies, considering that disease frequency and intensity are fixed in experimental
trials, yet under farmer conditions they may vary spatially by weather patterns, altitude,
and crop management practices. However, the presence of selection bias in farmer survey
yield data will tend to underestimate the real benefits of disease-resistant technologies, to
such an extent that Johnson and Klass (1999) recommend the use of experimental trials
over survey data, even with the limitations of the trial data.

This essay proposes methodological advances that allow for the use of farm
survey data to address the questions of yield loss avoidance and returns to research. The
point of departure is Otsuka et al (1994), who use farm-level survey data to estimate the
impact of RV (disease-resistant variety) rice in the Philippines. Otsuka et al apply a
modified Heckman two-step procedure (adapted from Lee, 1978) to correct for selection

bias. Lee’s procedure gives unbiased and consistent estimates of parameters of the

37

equations of interest (traditional variety (TV) yield and RV yield), which Otsuka uses to
compare the mean rice yields of TVs and RVs. However, these equations represent the
yield response of a given subsample of the population — TV growers in one case, and RV
growers in the other. In other words, they compare the expected RV yield (conditional on
RV use) with the expected TV yield (conditional on TV use). However, the more
appropriate counterfactual scenario for impact analysis is unobservable - the yield that
RV users would have obtained in the absence of RVs.

Extending Otsuka et al, this paper constructs the counterfactual to RV yields as
what yields RV users would have obtained had they continued to grow TVs (i.e. the TV
yield conditional on RV use). Constructing the appropriate counterfactual is especially
poignant in the case of RVs (and other technologies that mitigate adverse circumstances),
since it is anticipated that the TV yield profile differs significantly between RV users and
TV users. Specifically, farmers adopt RVs to avoid yield losses from disease. Farmers
who choose to continue to plant TVs must not experience significant yield losses,
otherwise they would adopt the RV. Using farm-level survey data on disease-resistant
bean yields in Honduras. a modification of Lee’s two-step procedure is employed to
obtain selection-corrected bean yield equations for TV and RV users. As demonstrated in
the labor supply literature (Lee, 1978; Duncan and Leigh, 1980), estimation of these
selection-corrected equations enables the prediction of imputed yields for each
individual’s unobserved varietal choice, conditional on his/her own observed
characteristics. That is, the selection-corrected yield equations account for the

endogenous farmer decision to adopt or not adopt RVs, and allows for meaningful

38

prediction of TV yields (including yield loss) that those farmers adopting RVs would
have gotten had the RVs not been available. Thus, the predicted TV yields of RV users
can be statistically compared with predictions of RV yields to estimate the farm-level
impact of disease-resistant bean varieties in Honduras, and the impact of the research that
developed these varieties.
2. Model

Our principal interest in this paper is to estimate the impact of RVs by predicting
the counterfactual yield of RV users -- the TV yield that RV adopters would have gotten
in the absence of RVs -- and comparing this with predictions of RV yields for the same
subgroup.I In the following model, each farmer’s observed yield depends on his varietal
choice, represented by equation (2). Yield YT,- is observed if the farmer grows a TV,
while YR, is observed if the farmer grows an RV. An implicit assumption of the model is

that farmers are knowledgeable about the disease pressure that they face.

11' = To + Yizi+ei (1)
We observe:
YR, when I, =1 if farmer grows a RV

YTi when I, = 0 if farmer grows a TV, but never both

 

' Comparing predicted TV yields with predicted (instead of actual) RV yields
ameliorates the influence of measurement error on the actual yields.

39

lnYRi = BRO + BRIXRI + uRi (2)
lnYTi = Bro + BTIXTI + “rt (3)
where lnYRi and lnYT, are the logs of yield for individual i, Xi is a vector of
personal and farm-level characteristics, and uRi and uTi are residuals. This is a
switching regression model with endogenous switching, as first developed by
Goldfeld and Quandt (1973) and later modified by Lee (1978).
3. The sample selection problem
Consider a population of farmers who can be identified as either adopters or non-
adopters of RVs. Further assume that beans are grown in areas of high and low disease
pressure, that high disease pressure leads to significant yield losses for TVs but none for
RVs (Table 1), and that all farmers (at least those in high disease areas) have access to
RVs. A random sample survey of this population will result in two subsamples of

farmers: one with observations on RV yields, the other with observations on TV yields.

Table 1. Hypothetical ielriprofiles by variety and disease pressure

 

 

 

 

 

 

 

Disease Pressure Scenario ,
Low # 1: TV yield = 800 kg/ha I # 2: RV yield = 800 kg/ha
High # 3: TV yield = 600 kg/ha ‘ # 4: RV yield = 800 kg/ha
TV = traditional variety; RV = disease-resistant variety

 

If consumers prefer the culinary characteristics of TVs, then farmers in low disease areas
could continue to grow TVs (Scenario # 1).2 However, farmers in areas of high disease
pressure will endogeneously select RVs (Scenario # 4) in order to avert the yield losses of

TVs (Scenario # 3). For the purposes of impact assessment, the most appropriate

 

2 We assume no price discounts for these scenarios. However, in the case that
follows, RVs face a price discount in the market.

40

counterfactual for RV users is not the sample mean of TV users (Scenario # 1), but what
RV users yields would have been in the absence of RVs (Scenario # 3). Therefore, failure
to account for this endogeneous choice may lead to bias in the comparisons of sample
mean yields by variety.

The same may be said of predictions of varietal yields from regressions, yet
whether (or not) sample selection creates a statistical problem depends upon whether (or
not) there are differences in both the observable and unobservable characteristics of TV
and RV growers. For example, once we assume that decision to adopt an RV is not
random, then the RV and TV subsamples potentially have different characteristics.
Sample selection bias in this case occurs when some component of the varietal choice
decision is relevant to the yield determination process; that is, when some of the
determinants of the varietal decision also influence yield. Yet, when the relationship
between the varietal decision and the yield determination is purely through the
observables, then appropriate variables in the yield equation can control for this. Thus,
sample selection bias will not occur simply because of differences in observable
characteristics (Vella, 1998).

When one further assumes that unobservable characteristics (such as disease
pressure) affecting the farmer’s varietal decision are correlated with the unobservable
characteristics affecting the yield determination process, then a relationship exists
between the varietal decision and the yield determination process. If these unobservable
characteristics are correlated with the observables, then the failure to include an estimate

of the unobservables will lead to incorrect inference regarding the impact of the

41

observables on yields. Thus, under these circumstances, a bias will be induced due to
sample selection (ibid, 1998).

In the case of our survey data from Honduras, we do not observe RV (TV) yields
for some farmers due to the outcome of another variable - varietal choice. Because
disease pressure — and perhaps other unobservables which influence varietal choice — is
not distributed randomly throughout the sample and is correlated with yield, then E(u,- |
varietal choice at 0). If some part of the yield distribution is cut off, then estimated
residuals from this model are truncated normal, and OLS estimates of yield equations are
inconsistent.

To see the effects of sample selection bias more clearly, consider the following

data generating process for yield:

yieldi = BO + B,* rainfalli + BJ‘fertilizeri + B3"‘RVi + deisease pressure,- + ui (l)
where: RV = 1 if farmer i grows a resistant variety, and RV=0 if not; rainfall and
fertilizer use are observed variables; and disease pressurei is an omitted variable
In the absence of a farm-level measure of the true intensity, timing or frequency of

disease pressure, and assuming that disease pressure is not distributed randomly
throughout the sample, then the results of the yield regression (1) will be biased because
the error term is actually u,- + B,*disease pressure. Thus, the mean error will not be
zero. In addition, we will encounter a further source of bias in that the omitted variable is

likely correlated with the RV dummy.

42

Roy (1951) provided an early discussion of the problem of self-selectivity,
discussing the problem of individuals choosing between two professions, hunting and
fishing, based on their productivity in each activity. The observed distribution of incomes
of hunters and fishermen was determined by these choices. The later discussion of the
econometric consequences of self-selection bias began with studies of womens’
participation in the labor force and implications for the estimation of wage equations
(Gronau, 1974; Lewis, 1974). The basic idea of the two-stage estimation procedure, as
first pr0posed by Heckman (1976) (and later extended by Lee (1976)), is to use
information from a probit function of labor force participation to construct a regressor for
the wage equation which serves to adjust the wage equation error term so that its expected
value will be zero.

4. Estimation Procedure

The two-stage estimation procedure for each yield equation proceeds as follows.

First, using all n observations, we estimate by ML 3 probit model of Z, on I, to obtain

estimates of y, and 7,.

1* = 70 T Yiz,i+ei* (4)

Conditional on RV use, the RV yield equation is:

lnYRi = BRO + BRiXRi T oRei )‘Rn + VR.‘ for H = 1 (5)

43

where E(vRI | Ii = 1) = 0. The term KR, = [-flw,) / F (w,)] is known as the inverse Mills’
ratio (IMR), where f is the density function of a standard normal random variable, and F
is its cumulative distribution, and where w, = 7,, + y,Z’,. The IMR term models the
truncation effect on yields associated with sample selectivity, and thus enables E(vR, | Ii =
1) = 0. The coefficient on the IMR term, oRc, is the covariance between e, , the error term
from the RV selection equation, and um, the error term of the RV yield equation. The
actual test for selection bias in the RV yield equation is simply a t-test of whether Ho: em,

= 0 or not. Rejection of the null indicates the presence of selection bias.

Conditional on TV use, the TV yield equation is:

lnYTi = Bro + BTixTi + OTei kn + VT. for H = O (6)

where E(vT, | Ii = 0) = 0. The term 79,, = [ f(w,) / (l - F(\y,))].

The error term e, in the selection equation is assumed to be distributed standard
normal, and also independent of Z, (and therefore X,). In addition, X, must be a strict
subset of Z,, and we must have some elements of Z, that are not also in X,. This means
that we need a variable that affects selection but does not have a partial effect on yields

(Wooldridge, 2000)’. Yield equations in switching regressions are commonly estimated

 

3 If 2 = x, then it, can be highly correlated with the elements of x,, and such
multicollinearity may lead to very high standard errors for the 3,. Thus, without this
exclusion restriction, it is extremely difficult to distinguish sample selection bias from a
misspecified functional form for yields (Wooldridge, 2000).

44

either jointly (pooled) (Otsuka, 1994) or separately (Lee, 1978; Duncan and Leigh, 1980).
This paper estimates the two yield equations pooled together to gain efficiency in
parameter estimation. An RV dummy serves as an intercept shifter for RV users, and is
interacted with some of the regressors. Thus, the expected yield for an RV user is

represented by (7) while that of a TV user is represented by (8).

lnYR, l R, = BTo + BR0*RV + BT,XR, + BR,XR,*RV + oRc, )tR, + vR, for RV=1; Ii = l (7)

lnYT, | T, = BT0 + + BT,XT, + + on, 79,, + VT, for RV=0; Ii = 0(8)

In the case of disease-resistant beans in Honduras, TV yields are expected to be
truncated from below due to selection bias. That is, low TV yields caused by disease
pressure are unobserved if farmers in high-disease areas switch to RVs. Thus, assuming
that TV yields depend upon the level of disease pressure, then unobservable
characteristics (disease pressure) will likely influence both the varietal decision (with no
disease pressure, farmers prefer the TV due to its better price) and the yield determination
of TVs. However, it is not clear whether (or not) the presence (or not) of disease pressure
would affect RV yields. That is. RV yields under farmer conditions are expected to be
about the same with or without disease pressure, ceteris paribus. On the other hand,
given that RV yields are not often observed in non-disease areas (higher elevations) due
to their price discounts relative to TVs, and that RVs are likely better suited for the
conditions of valley farmers. there may well be a part of the RV yield distribution that is

not observed.

45

Because the error term of the pooled regression has a heteroskedastic structure, we
apply Huber/White’s procedure to obtain robust standard errors. The final step is to
calculate predictions of RV users’ RV yields using the corrected RV equation (9). The
counterfactual prediction is RV users’ TV yields, which are calculated using the

characteristics of RV users evaluated with the coefficients from the TV equation (10).

ln§Rl | RI: fiTO + BRO*RV '1' fiTlle T f3121')(141*R\/ + 8am )‘Ri fOFRV=1;1i=1(9)

lnYT, | R,= 8,, + + 8,,XR, + + 311.14... for RV=O; Ii= 1 (10)

5. Background and Data

Honduras’ two bean research programs, which are implemented by Zamorano
(Programa de Investigaciones en Frijol, Escuela Agricola Panamericana) and DICTA
(Direccion de Ciencia y Tecnologia Agricola, the Honduran National Bean Program),
work in collaboration with three organizations: the USAID-funded Bean/Cowpea
Collaborative Research Support Project (CRSP); the Programa Cooperativo Regional de
Frijol para Centroamerica, Mexico, y El Caribe (PROFRIJOL), funded by COSUDE, the
Swiss Agency for Development and C00peration; and CIAT (Centro Intemacional de
Agricultura Tropical). Bean Golden Yellow Mosaic Virus (BGYMV) is the principal
bean disease in Honduras, and one of the main production constraints in Honduran

valleys (Martel, 1995). The virus is transmitted by the whitefly species Bemisia tabaci

 

which is normally found below 1,000 m in most growing regions of Honduras and more

frequently in the drier postrera season (September/ October to December/January).

46

Zamorano and DICTA have developed several varieties with resistance to BGYMV
(RVs), the first of which was released in 1990. However, the economic gains from yield
loss averted through RV use are at least partially offset by price discounts in the market
from 7 to 15 percent (Martel, 1995; Mather et al, 2003).

This analysis utilizes data collected through a farm-level survey of bean farmers
(N=210), which was implemented by the CRSP and PROFRIJOL in January-February
2001, in collaboration with Zamorano and DICTA. The survey targeted the Mideast (El
Paraiso and Francisco Morazan departments) and Northeast (Olancho) regions, which
together account for about 60 % of Honduras’ annual bean production. Bean producers in
Honduras are predominantly small farmers, the majority of whom use fertilizer and
insecticide,~market at least some of their bean production, and depend upon beans for a
major portion of their household income (Martel, 1995; Mather et a1, 2003). Honduran
farmers typically plant two crops during the year. In the primera (May/June to July/
August), the rainy season, maize is the principal crop, and beans are either intercropped
with maize or monocropped". The postrera (September/October to December/January) is
a drier season, during which beans are the major crop and are almost exclusively
monocropped. Because the whitefly population is often highest under drier conditions,
BGYMV is most frequent and severe in the postrera.

Most of the sample farmers (70 percent) live in villages which have faced

BGYMV pressure in the past (Table 3), and also (58 percent) live in villages which have

 

" In the following analysis, intercropped area is converted to “effective monocrop

bean area” given information from each respondent on his/her intercropping system.

47

received village-level extension support, including promotion, demonstration, and/or
access to RVs. RV adoption is widespread, and mean RV yields tend to be higher than
those of TVs (Table 2). Yet, because of the high variance in both RV and TV yields,
yields by varietal type are not statistically different (at the 5 % level).

The impact analysis of varietal development involves the estimation of the
incremental change in benefits received by the farmers who have adopted the variety. In
our case, we want to estimate the current yields of RV users, as well as their yields if they
had not adopted an RV. However, as mentioned above, using the current TV yields of
TV users as the RV users’ counterfactual ignores the potential for selection bias to
underestimate the benefits to RVs. Therefore, assuming that selection bias is present, the
appropriate counterfactual to current RV yields is unobservable: the yields these RV
growers would obtain if they were to use TVs instead. The fact that the RV survey yields
and coefficients of variation are quite similar to those of TVs — given that significant
price discounts exist, most farmers market beans, and adoption rates are fairly high —
suggests that selection bias is indeed present, and that using current TV yields of TV

users would underestimate the benefits of RVs to RV users.

48

Table 2. Adoption of RVs and Bean Yields by Season, 1999-2000, Honduras

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variety Primera Postrera
1999 2000 1999 2000
Varietal Use: RVs (% farmers) 45 45 41 42 3
TVs 64 62 67 76
BothanRVandaTV 9 7 8 18
Sample size (farmers) N=l64 N=170 N=202 N=202
Yield: RVs (mean kg/ha) 857 769 667 446
TVs (mean kg/ha) 678 632 615 459
Disaster (%) b 4 6 6 14
Rainfall; first month of season (avg mm) ° 126 171 314 206
Sample size (parcels) d N=188 N=203 N=242 N=268

 

b Farmer with yield < 80 kg/ha

 

Source: CRSP/PROFRIJOL Farmer Survey, 2001 (N=210)
’ As some farmers planted more than one variety, the total is greater than 100%.

° Source: Depto. de Servicios Hidrologicos y Climatologicos, Direccion General de Recursos
Hidricos, Secretaria de Recursos Naturales, Honduras

d Parcels aggregated at the farmer and variety level

 

49

 

Table 3. Characteristics of sample farmers, Honduras, 2000

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Characteristic Primera Postrera
Bean area (mean ha) 1.30 1.58
Altitude (m) 945 913
% Farmers: in Area of Bean Golden Mosaic Virus a (%) 70 70
Used formula fertilizer (%) 55 48
Used Urea (%) 49 45
Used both formula and urea (%) 29 25
Intercropped (%) 14 15
Ever planted a resistant variety (PRV) (%) 67 68
% farmers living in village with extension (by type): b
DICTA artisan seed program (%) 22 19
Dept. of Natural Resources (%) 35 32
Zamorano (%) 16 14
Catholic Church (%) 13 17
No extension at village level (%) 42 45
Sample size N=l 70 N=202

 

 

improved varieties

Source: CRSP/PROFRIJOL Farmer Survey, 2001 (N=210)
“ defined by Dr. Juan Carlos Rosas, Director, Programa de Investigaciones en Frijol,
Zamorano b extension which includes promotion, demonstration, and/or extension of

 

6. Estimation results of varietal choice function

We estimated the probit RV adoption function (Table 4) using data from four

consecutive seasons: the primera and postera seasons of 1999 and 2000. Most of the

elements of the adoption function are factors expected to influence bean yield such as

fertilizer ‘use, rainfall, cropping system, season, etc., as well as village dummies which

help to capture non-observable agronomic factors common at the village level. Farmer

50

 

socioeconomic factors such as age, education, and farrnsize are not included here due to
their insignificance.

Formula fertilizer use is defined as zero if the farmer used no formula fertilizer in
a given season, and is defined as one if the farmer used formula in that season’. Variables
for urea fertilizer use and combined formula and urea use are similarly constructed.
Rainfall represents the total village rainfall (mm) in the first month of the given season.
While rainfall during the first month of a bean season is undoubtedly an important factor
in yields, the explanatory power of this rainfall data is questionable for several reasons.
First, only nine rainstations across the three departments had complete data for the four
seasons of production data. Thus, each of the 30 villages was assigned the value reported
at the nearest rainstation to represent village rainfall. Second, the timing and distribution
of rain within a month is not captured by the monthly total. Intercrop refers to any
cropping system which is not monocrop (but does not include the relay system common
in Olancho in the postrera, which is essentially monocrop). Disaster is a dummy to
capture the influence of yields less than 80 kg/ha.

The remaining factors in the adoption function serve as exclusion restrictions —

variables that affect selection but do not have a significant partial effect on yields‘. These

 

5 We only have fertilizer use data for primera and postrera of 2000. Since this
variable measures only use and not fertilizer level, and since many farmers regularly use
fertilizer, we assume that fertilizer use in 1999 was the same as that observed in 2000.
This assumption does not affect the test for selection bias, nor the significance or
magnitude of key variables in the corrected yield equation such as RV and fertilizer use.

6 These exclusion restriction variables were included in an OLS yield regression
and determined to be insignificant at the 0.10 level.

51

include a dummy variable for villages in which BGYMV has been a problem in the past7,
village altitude (lower altitudes experience higher BGYMV pressure), village altitude
squared, and a dummy variable for farmers who had planted an RV at some time in the
past (PRV).

While a measure of a farmer’s information about and access to RVs may be well-
captured by village-level extension variables, these variables are not ideal exclusion
restriction variables since extension presence in a village -- which in this case includes
more than just information about and access to RVs — would be expected to influence
(increase) yields. BGYMV is only a rough indicator of disease pressure, since more
significant the intensity of disease attack and its frequency is more significant in the
farmers’ adoption decision. Altitude is also a rough indicator of disease pressure, as
lower altitudes tend to have higher BGYMV pressure. PRV is an ideal exclusion
restriction variable because it is intimately related to adoption, yet does not predict
adoption perfectly due to disadoption (67 percent of the sample has planted an RV at
some point, while 45 percent continue to grow an RV) and is not correlated with yields.

The estimation results (Table 4) show that the coefficient on the BGYMV dummy
is positive, significant and of relatively large magnitude, which confirms the expectation
that the principal benefit of RVs is expected to be the yield loss averted due to their
disease resistance, and that farmers outside the BGYMV area would not likely adopt the

older RVs which have significant price discounts. Also as expected, the coefficient on

 

7 BGYMV area classification as designated by Dr. Juan Carlos Rosas, Director,
Programa de Investigaciones en Frijol, Zamorano.

52

the PRV dummy is positive, significant and of sizeable magnitude. Altitude is also
significant, yet it’s sign is contrary to expectation as the probability of adoption is
hypothesized to decrease as altitude increases.8 The factors influencing adoption which
are of principal interest to this paper are the exclusion restriction variables, none of which

are significant when tested in the yield regression.

 

8 While village altitude is negatively correlated (pairwise) with RV adoption, it’s
sign becomes positive when village dummies are included in the probit regression.

53

Table 4. Probit Estimates of Resistant Variety Adoption Equation, Primera and

Postrera, 1999-2000

 

Explanatory Variables 8

RV adoption
(RV=1, TV=0)

 

Constant

Season = Postrera

Year = 2000

Intercrop

Disaster (yield < 80 kg/ha)

log of Rainfall in first month of season (mm)
Used formula fertilizer (0=no; l=yes)

Used urea fertilizer (0=no; l=yes)

Used both formula & urea (0=no; 13165)

4.09 (2.79)"
-0055 (-120)
-0023 (-0.61)
0.044 (0.63)
0.075 (1.01)
0.014 (0.28)
-0.108 (.212)M
0.104 (2.15)"
0.0005 (1.65)*

 

Village-level Extension: DICTA
Zamorano
Dept of Natural Resources
Catholic Church

BGYMV area

Altitude (m)

Altitude2 (m)

Ever Planted Resistant Variety (PRV)

0.729 (1.75)
-0432 (345)"
-0.946 (.442)M

-0042 (-019)
0.258 (1 .80)*
0.006 (2.08)"
-00000 (-1.82)*
0.355 (7.81)"

 

a Does not include 30 village dummies
(z stats in parentheses)

** significant at the 0.05 level

"' significant at the 0.10 level

 

 

N = 900

Log Likelihood = -381 .47

Pseudo R2 = 0.343

Coefficients calculated as dF / dX

 

7. Estimation results of bean yield function

To estimate the yield function, we pooled the data over the primera and postrera

seasons of 1999 and 2000 (Table 5). We assume that many factors affecting yields, such

54

 

as year, cropping system, village dummies, etc., have the same effect on both TVs and
RVs. However, we assume that the productivity of TVs and RVs are differentially
affected by factors such as fertilizer use and season. The regressors in the yield function
are nearly the same as those used in the probit function, the difference being the inclusion
of the inverse Mills’ ratios (IMR) and the omission of the exclusion restriction variables
from the probit function.

As expected, the IMR coefficient for TVs is positive, significant at the 0.05 level
and of considerable magnitude, thus indicating the presence of selection bias. The
positive sign on this coefficient indicates positive selection which means that observed
TV yields are higher than what would be observed (positive truncation) if a farmer
randomly chosen from the whole sample were to plant a TV. The reason for this is that a
farmer selected at random may be in a disease-intensive area, and thus would get lower
yields, whereas such a farmer would not, in reality, plant a TV.

We did not have an a priori expectation for the IMR coefficient term for RVs,
although it is found to be positive, significant at the 0.10 level and of smaller magnitude
to that of the TV. However, because the sign on the IMR term IR, = [- f(\y,) / F(\y,)] is
negative, while its corresponding coefficient is positive, this implies that negative
selection occurs for RV growers. That is, observed RV yields are lower than they would
be in the absence of the unobservables which drive RV use. An explanation for this
could be that the unobservable BGYMV pressure (which may well be correlated with
other unobserved disease and insect pressures to which the RVs are not resistant) which

drives a farmer to experiment with RVs (PRV) tends to lower the RV yields from what

55

they would be in disease-free areas. In addition, we do not observe many RV growers
(and, thus, RV yields) in disease-free areas for two main reasons. First, farmers in areas
with low to negligible BGYMV pressure will not likely grow RVs due to their market
price discounts. Second, areas above 1,000 m are more remote and thus have less access
to markets and extension - both of which condition access to RV seed and fertilizer.

The results of the yield function estimation corrected for selection bias compared
with uncorrected OLS results are most distinctively different for the RV dummy and the
constant (Table 4). The RV dummy coefficient is highly significant and large in the
selection corrected estimation, whereas it is insignificant and small in the uncorrected
estimation. Thus, the selection-bias correction increases the significance and magnitude
of the RV dummy. All other coefficients in the yield model are essentially of the same
significance and magnitude. This means that an RV farmer has higher expected yields
than a TV farmer with similar characteristics.

Both corrected and uncorrected estimation show that RVs are more responsive to
urea fertilizer than TVs, although the R V*urea coefficient is not significant at the 10%
level in either equation. However, given the relative magnitude of the RV dummy and
those for fertilizer response, it is clear that the principal benefit of RV use is disease
resistance, not improved response to fertilizer. The coefficient for season (postrera) is
negative and significant as expected, given that the postrera season is drier, has more
disease pressure, and historically produces lower yields than the primera. While we
would expect rainfall to increase yield, the coefficient on village rainfall (mm) is

insignificant, perhaps due to the presence of village dummies and the nature of the

56

rainfall data (explained above). While intercropping systems can result in higher bean
yields, the. significant and positive coefficient on intercrop is likely driven by two factors:
most intercropping occurs in the primera (when yields are higher), and the conversion of
intercropped bean area to “effective monocrop” bean area may underestimate the
effective area. The inclusion of village dummies improves the explanatory and thus
predictive power of the regression, as well as the estimation precision of nearly all the
coefficients, although our interest in coefficients relates principally to those for the RV

dummy, fertilizer use, and the IMR terms.

57

Table 5. Adjusted and Unadjusted OLS Estimates of Yield Equation, Primera and

Postrera, 1999-2000

 

 

 

RV * Both Formula & Urea Fertilizers

-0.0006 (-0.51)

Explanatory Variables ‘3 Corrected OLS Uncorrected OLS
ln(yield) ln(yield)
Constant 5.630 (14.74)“ 5.718 (14.82)"
Season = Postrera - 0.220 (-2.97)** -0.238 (-3.23)M
Year = 2000 - 0.192 (-2.85)** -0.193 (-2.85)**
Intercrop 0.245 (2.52)M 0.269 (2.81)"
Disaster (yield < 80 kg/ha) -3.369 (44.80)" -3.36 (44.69)"
log of Rainfall in first month of season (mm) 0.021 (0.29) 0.026 (0.35)
Used formula fertilizer (0=no; l=yes) 0.206 (2.14)" , 0.151 (1.64)
‘ Used urea fertilizer (0=no; l=yes) 0.060 (0.76) 0.098 (1.27)
Used both formula & urea (0=no; l=yes) - 0.0004 (-1.53) - 0.0005 (-1 .46)
RV dummy 0.552 (2.73)" 0.129 (1.10)
RV * Season = Postrera - 0.175 (-1 .52) - 0.177 (-1.56)
RV * Year = 2000 - 0.020 (-0.l8) - 0.015 (-0.l4)
RV * Disaster - 0.554 (-1.48) - 0.554 (-1 .47)
RV * Formula Fertilizer - 0.025 (-0.16) - 0.013 (-0.08)
RV "‘ Urea Fertilizer 0.155 (1.09) 0.151 (1.07)

-00007 (-054)

 

Inverse Mills’ Ratio (TVs)
Inverse Mills’ Ratio (RVs)

0.342 (2.51)”
0.236 (I .69)*

 

" Does not include 30 village dummies
(t stats in parentheses)

*" significant at the 0.05 level

* significant at the 0.10 level

 

N = 900
_ R2 = 0.680
F(46, 854) = 16.07

 

 

N = 900
R2 = 0.678
F(44, 856) = 16.36

 

8. Counterfactual Predictions

Estimation of the selection corrected yield function enables the prediction of

imputed yields for each individual’s unobserved varietal choice, evaluating each

58

 

 

individual’s own observed characteristics at the coefficients from the alternate subsample.
Finally, the predicted TV yields of RV users can be statistically compared with
predictions of RV yields to estimate the farm-level impact of disease-resistant bean
varieties in Honduras.

The significant coefficient on the RV dummy imply that the expected yield of RV
users is higher than that of TV users, ceteris paribus. However, the more appropriate
counterfactual scenario for impact analysis is the predicted yield that RV users would
have obtained in the absence of RVs — the TV yield of an RV grower. The conditional
counterfactual estimates the yield differential between predicted RV and TV yields for a
farmer chosen at random from the RV subsample. This conditional counterfactual9 is
computed for RV growers as the sum of the selection corrected constant (common to both
RV and TV growers) plus the relevant IMR coefficient multiplied by the corresponding

IMR term, plus the characteristics of RV growers multiplied by the TV

 

9 The conditional counterfactual assumes that the grower has already selected into
the RV subsample and includes the IMR term. Unconditional counterfactual would not
assume that the grower had already selected into the RV subsample, and would not
include the IMR term in the computation of predictions. While predicted yield levels
vary given which counterfactual is computed (Table 6b), the percentage differential
between the predicted RV yield and the predicted RV counterfactual is not affected by
this choice. The reason for this is that since these differentials are calculated using
predictions from the same subsample, it doesn’t matter whether or not the IMR term is
included. However, if we were to compute a differential between predicted RV yields
(for RV users) and predicted TV yields (for TV users) - a differential across subsamples
— then the counterfactual choice makes a large difference. For example, the differential
between conditional predictions across subsamples are almost identical to those from
OLS (605 vs. 541 kg/ha is 11%), while differentials between unconditional predictions
are much larger (714 vs 476 kg/ha is 33%). Conditional differentials would thus be
expected to be smaller than unconditional estimates, in which the varietal choice has not
yet been made.

59

 

.638 2:8 05 3 33.3.3 .mfizcoLobE

58.5 no .82: 05 we wean—:28 .atcoaotmc :31 .2: 6 _ >M \ A3. _ no: - >5. ewe .8 ”ES “mac 2: 3 poetic 3:828 mac—om.»
e86€2a :35 65 :85 228605 23% 58:. 29:8 .8953 coconut? 05 we can. 65 we noun—=23 fl Eta—Bot? c3 1....
«Swa— _mm n 2262. >.~. 3&9— Vmc u «Lo—Ea >._. 35w: man u >._. ”me—om» 29:8 :88 32820

fitme— mmm H 80.88 >m 36w: o; n 20th >M map—ha— mno u >m 3263 29:3 .82: noEoBO

338%258 .1. mo .1. 2:8 MS: 23 353508 52— ops—2: 26:23.5 36280 _._

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

car: a:
2 a 5; 8m 2 , a ma. :5 ”$12 >m 858.. 5.265 .s a :63. s E
2 2 :4 8m 2 2 E. 86 memuz 8256a ” Q a :65. a, E
o m - «S N8 6 m - :6 NS £12 62:5 u Q t :62 s E
S m am new a m N: :5 ”$12 Eur: 665 Q a :65. s E
a a - am 8m 2 - 6m - c8 .3 ”$12 9m 5.25 2 a :63. a, E
m - w - N 5 Sn 6m - S - a: .3 «$12 sari. :>a 5263; a :65. e, Z
a - _ - am as. «N 3” :m E. $.12 8258 n a. :65 s >m
S 2 68 NE S a. a z. E 3.12 86E... n z. :6: s B.
o a E 36 mm s. an m8 312 t a. :62. s >m
Ase a; \ was as a: \ we:
5.26;
Etcoaoba eosguoi :32 6...... Ezeeoba c8235 :32 8% 63.52.35 .3
035mm maocgvoi :32
cotaESmm mAO 382825 Eases—tam mAO 38280 ham mo cemtanU

 

 

 

 

88-83 £262 :65 B852: Ea 685260 62. 526: 63.. >a E: B. .6 226:5...— 5 63:.

60

 

coefficients estimated from the TV subsample.

After obtaining predictions from each regression, the mean values of log(yield)
are scaled to levels”). The mean predictions (Table 6) are similar to actual means,
although the predicted RV yields are lower in general than actual RV yields. Two
differentials are presented (Table 5); the first is calculated as the ratio of the difference
between sample mean yield predictions (from the mean predicted yields columns) divided
by the first term; for eg., [(RV - Tch | R, )/ RV ] * 100; while the second is calculated as
the mean of farmer differentials, calculated by the same ratio, but for each farmer.

Given that we found negative selection for RV yields (mean RV yields are lower
than they would be if RVs were grown in areas of no disease pressure) and positive
selection for TV yields (TV yields are higher than they would be if TVs were grown in
areas of high disease pressure), we would expect that differentials calculated from the
selection-corrected yield equation would be larger than those from the uncorrected OLS
equation. As expected, comparison of differentials by estimation technique shows that
those from the corrected model are significantly larger than those from the uncorrected
model (Table 5); the corrected model predicts the mean farmer differential to be 42% in
the primera, and 28% in the postrera, while the uncorrected model predicts a differential

of 12% and -9%, respectively for these two seasons.

 

'0 The conversion of predicted log yield to level yield is exp(scale) *
exp(ln(yield__hat)) where scale is computed as the coefficient B0 of the regression
Yield_hat = B0 exp(ln(yield_hat)) with no constant.

61

Furthermore, due to market price differentials by variety, yield differentials must
be adjusted in order to more accurately reflect the net gain to farmers of varietal choice.
For example, the valuation of an RV grower’s counterfactual (using a TV) must account
for the fact that while the RV may yield more than the TV, the market price of an RV is
15% less relative to a TV1 ‘. Even after adjusting the differentials for market price
discounts, the corrected model predicts that RV growers enjoy net income gains of 27%
in the primera and 13% in the postrera compared to what they would have earned
growing TVs.

However, adjusting the uncorrected model differentials for the market price
discounts yields counterintuitive results: farmers growing RVs would lose 3% in net
income in the primera and lose 24% in net income in the postrera relative to what they
would have earned growing TVs. This result implies that RV growers do not enjoy any
yield gain (yield loss averted) to compensate for the price discount (and actually lose net
income in both seasons), yet continue to grow RV nevertheless. By contrast, RV
differentials from the corrected model more than offset the price discount in both seasons.
The conflicting results from these two models highlight the sensitivity of impact analysis
of maintenance technologies to the method of econometric measurement of farm-level net
benefits: using farm-level net benefits (differentials) from the corrected model yields

positive aggregate benefits, whereas using differentials fiom the uncorrected model yields

 

“ Dorado and Don Silvio, released in the early 19905, represent 75% of current
RV users. The mean price discount for these varieties is 15%. However, Tio Canela,
released in 1997 and representing 25% of RV users, has a smaller farm-level price
discount of 7% (Mather et al, 2003).

62

negative aggregate benefits.

Counterfactuals are also computed for TV growers as a means to test the model,
by estimating a predicted TV yield and adding the RV coefficient and the RV’fertilizer
coefficients (for those TV farmers who use fertilizers). While we would expect predicted
differentials for TV growers to be positive -- or if negative, to not be larger than the price
discount -- the model in fact predicts that TV growers would see sizeable net gains from
switching to RVs. As indicated by Table 6, these results are driven by the RV coefficient,
which is added to the TV counterfactual. Given that access to RVs is not likely a critical
factor in the farmers’ adoption decision, this result may be explained by the fact that the
RVs were targeted for valley farmers, whose growing conditions are quite different from
higher-altitude farmers. That is, the RV dummy represents a yield effect enjoyed by RV
adopters in disease-prone areas (which may well have other disease and pest pressures
beyond simply BGYMV) over and above BGYMV resistance, that the average TV
grower will not enjoy. If this assumption is correct, then the more appropriate
counterfactual for TV growers is #3 (Table 6), wherein we compare the mean predicted
TV yield with the mean predicted TV yield which includes the fertilizer effects of RV
growers. These counterfactual values appear more reasonable as predicted RV yields are
not high enough to offset the RV price discount. Another explanation for the large TV
differentials could be that the model consists primarily of dummy variables, and thus we
may not have enough variation among the characteristics (regressors) of TV farmers to
capture the differential response of TV and RV varieties to different farmer input

quantities and qualities.

63

9. Implications for Impact Analysis

Using the economic surplus model and costs of research as outlined in paper #1,
rates of return to disease-resistant bean research in Honduras are calculated for the
postrera seasons 1982-2010 (Table 7), using differentials from the corrected and
uncorrected models, as well as the differential derived from an expected utility
framework (paper #1; Mather et a1, 2003). The results demonstrate the significant
implications of the method described in this paper for the construction of counterfactual
scenarios for use in the analysis of the impact of maintenance technologies. Use of
uncorrected OLS results would lead an analyst to conclude that RV use generated
negligible or negative returns, while the corrected OLS results generate a high ROR. The
expected utility framework provides a minimum estimate of the net farm-level benefits of
RV use, which appear to have underestimated the benefits in comparison with the

corrected OLS approach using sample mean differentials.

64

Table 7. Rates of Return to Disease-Resistant Bean Research in Honduras, Postrera
Seasons 1982-2010

Method of Calculating Incremental Incremental net farrn-Ievel Economic Rate
Farm-level net benefits of RV use benefit of Dorado use in of Return * (%)
the postrera season,
adjusted for price
discount (%)

 

 

 

 

 

 

 

 

Expected Utility Framework ** 11 . 41.2
Farmer Mean Corrected Model: 13 45.0
Differential Uncorrected Model: - 24 -
Sample Mean Corrected Model: 19 54.5
Differential Uncorrected Model: - 16 —

 

 

* Calculations use the economic surplus model as outlined in first essay.
*"' Framework and benefits as presented in first essay.

 

 

10. Conclusions

This paper demonstrates a method for using farm-level survey data in the
construction of counterfactual scenarios for use in impact assessment of maintenance
research. The method uses a modification of Lee’s (1978) two-step procedure to correct
for selection bias, the presence of which in farm-level survey data will likely lead to
underestimation of the benefits of maintenance research (Johnson et al, 1999). The
method is applied to test for selection bias and estimate yield differentials between RV
and TV bean yields in Honduras by constructing the counterfactual to RV yields as what
yields RV users would have obtained had they continued to grow TVs (i.e. the TV yield
of RV users). The corrected yield model predicts that RV growers enjoy net income
gains of 13 to 19 percent compared to what they would have earned growing TVs, while

the uncorrected OLS model predicts that RV growers see either no income gain or even

65

 

the uncorrected OLS model predicts that RV growers see either no income gain or even
losses by growing RVs (relative to TVs). This application demonstrates that the
implications of this method are significant for the assessment of maintenance research

impacts, both at the farm-level and in terms of the ROR to research investments.

References

Duncan, G., D. Leigh. 1980. Wage Determination in the Union and Nonunion Sectors:
A Sample Selectivity Approach. Industrial and Labor Relations Review 34:24-34.

Goldfeld, S.M., Quandt, RE. 1973. The Estimation of Structural Shifts by Switching
Regressions. Annals of Economic and Social Measurement 2:475-485.

Gronau, R. 1974. Wage Comparisons — A Selectivity Bias. Journal of Political
Economy 82:1119-43.

Heckman, J. 1976. The Common Structure of Statistical Models of Truncation, Sample
Selection, and Limited Dependent Variables and a Simple Estimator for Such
Models. Annals of Economic and Social Measurement 5:475-492.

Heckman, J. 1979. Sample Selection Bias as 3 Specification Error. Econometrica
47: 153-161.

Johnson, N., Klass, J ., 1999. The Impact of Crop Improvement Research on Rural
Poverty: A Spatial Analysis of BGYMV-Resistant Bean Varieties in Honduras.
Paper prepared for the workshop: Assessing the Impact of Agricultural Research
on Poverty Alleviation, San Jose, Costa Rica, September 14-16, 1999.

Lee, L.F. 1976. Estimation of Limited Dependent Variable Models by Two-Stage
Methods. Ph.D. dissertation. University of Rochester.

Lee, L.F. 1978. Unionism and wage rates: A Simultaneous Equations Model with
Qualitative and Limited Dependent Variables. International Economic Review
19:415-433.

Lewis, H.G. 1974. Comments on Selectivity Biases in Wage Comparisons. Journal of
Political Economy 82(6):]145-55.

66

Mather, D., Bernsten, R., Rosas, J.C., Viana, A.R., Escoto, D. 2003 (forthcoming). The
Economic Impact of Disease-Resistant Bean Research in Honduras. Agricultural
Economics.

Martel, P., Bernsten, R., Weber, M., 2000. Food Markets, Technology, and the Case of
Honduran Dry Beans. Michigan State University International Development
Papers, Working Paper No.78. Department of Agricultural Economics, Michigan
State University, East Lansing, MI.

Martel, P., 1995. A Socio-Economic Study of the Honduran Bean Subsector: Production
Characteristics, Adoption of Improved Varieties, and Policy Implications. Ph.D.
Dissertation, Department of Agricultural Economics, Michigan State University,
East Lansing, MI.

Otsuka, K., F. Gascon, S. Asano. 1994. Second-generation MVs and the evolution of the

Green Revolution: the case of Central Luzon, 1966-90. Agricultural Economics
10:283-295.

Roy, AD. 1951. Some Thoughts on the Distribution of Earnings. Oxford Economic
Papers 3:155-146.

Smale, M., Singh, R.P., Sayre, K., Pingali, P., Rajaram, S., Dubin, H.J., 1998. Estimating
the economic impact of breeding nonspecific resistance to leaf rust in modern

bread wheats. Plant Dis. 82:1005-61.

Vella, F. 1998. Estimating Models with Sample Selection Bias: A Survey. The Journal
of Human Resources. 32:127-169.

Wooldridge, J. 2000. Introductory Econometrics: A Modern Approach. South-Westem
College Publishing.

67

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5088.. none .8 :5 .23.. 2.85 05 .3 v0.3.6.3 .5_a_.:o.ot€
.252 .6 .62.. 2.. a 3.5.3.8 5. 3.6262... 2.5.... .8. .. . >5 . . a. :63 - >6. .5... .6. 2:2 .2... 2.1.. 62.5... .2628
5203 38:55 SEE o... :55 5:26an 263 .82: 29:8 5222. oozeotE 05 we 3.8 o... 55 3.23:... 5_ 33:255.“. to..— ..
.mBoafioESo .1. .3 a
a5».— .mm .1. 89:505. >.~. mashe— Vme n flog... >.—. mag—E.— Nwm u >._. .520.» 29.85 :88 3.5.0530
fitme— mmm u 9.2.565. >M mafia.— Sw u «BEE >M 35».— m3 u >¢ .526.» 2an5 :35 33655.0
€2.55 .2.
N. 5 55.. :5 N. . 5 55.. 5:. 55512 >5 5....2. 5...... 3.. a :63. 5> E
5. 5. .9. 55.. a. 5. :5 5. 5.512 6.2.8.. . 5.. t :63. 5., >..
c 5 - :5 NS .. 5 - $5 5.5 5512 22...... u 5... a :63. 55 >5
5 5 5.5 :5 5 5 c5... 5:. 55512 €2.55 52.... 5.. a :63. 5., Z.
55 - 55 - 85 :5 N5 - 55 - 55. 5:. 55512 .3. 2...... N... a :63. 5., >2
5 - .5 - a... :5 55 - .5 - a: 53. 55512 €2.55 .3. 525... a :65. 5.. Z.
55 .5 5.5 5:. 55 .5 555 555 $.12 22.8.. u e. :65 5., >2
5.. 5.. 5 .6 55. 5.. 5.. 2:. N... 5212 22...... u a. :63 5., 3.
55 a... .55 58 55 3. 5. 5... 5.5.1.2 . a. _..6>... 55 3.
es. .2. . w... as. .2. . 5..
FEEDER. 5.8625. :32 .. 33:25.85 526; 3.6525. :32
cum . 25.86535 .3
as: 0383 55:662.. :32
.23 cotafimmm mAO 38280 as: 3223 cot—.8251". mAO unsetoo .95 we saga—=00

 

 

 

 

 

8.555. .2252 22> .2825 .22....50 6.... .22....525 82. 5......N 5.2. >5 .2... >2 .6 58.5.3... .55 2...;

68

FACTORS INFLUENCING THE ADOPTION OF DISEASE-RESISTANT BEAN

VARIETIES IN HONDURAS

ESSAY THREE

69

1. Introduction

Improved bean varieties are one of the best ways of improving small farm
household income and food security in Central America — they are scale-neutral,
reproducible on-farm, and do not require complementary physical or human capital inputs
(Martel et al, 2000). Previous research by Martel (1995) found that the adoption of
disease-resistant bean varieties (RVs) in Honduras was neutral with respect to farm size,
yet biased with respect to fertilizer use, administrative region, and topographical region.
The first RVs released, Dorado (in 1990) and Don Silvio (in 1993), were more widely
adopted in valleys, where disease pressure is greater, yet received market price discounts
of 10 to 15 percent due primarily to their darker color. The latest RV released, Tio
Canela (in 1997), purportedly has better market characteristics and thus is expected to
replace Dorado and Don Silvio, as well as to be adopted in areas of lower disease
pressure.

This paper builds on Martel’s work by using a more recent (2001) survey to
investigate the factors influencing adoption of the RVs. This survey provides a more
recent snapshot of the factors influencing RV use; after ten years of extension and farmer
adoption/disadoption of Dorado, and three years after the release of Tio Canela. In
addition, this paper investigates the role of three additional factors in the farmer’s
adoption decision not included in Martel’s analysis: extension, market participation, and
ambient adoption rates. These additional variables can inform policy makers and
breeders in new ways about the kinds of farmers adopting or not adopting RVs, and

especially regarding whether or not Tio Canela’s better relative market characteristics are

70

borne out by farmer adoption by both Dorado/Don Silvio growers and previously non-RV
growers.

Literature on seed systems and technology adoption in developing countries offers
various explanations for non-adoption of improved crop varieties by farmers (Feder et al,
1985; Tripp, 1997). Two of the more common explanations which are relevant to
improved bean varieties include: lack of information about improved varieties and/or lack
of access to improved seed; and secondly, inappropriate or unprofitable technology (i.e.
experimental results are not representative of farmer agronomic and/or economic
conditions). Access factors include extension programs, which affect farmers’
information and access to improved seed, as well as farmer-to-farmer exchange, which is
represented by the local RV adoption rate. Because Honduras’ bean seed system is highly
informal, access to information about new varieties and seed itself could be a constraint
to more widespread adoption of RVs. Although farmer seed exchange is common, it may
take considerable time to achieve extensive coverage, and between and even within
communities, seed exchange mechanisms may not be equitable (Tripp, 1997). Breeding
factors affect the agronomic and economic performance of the RVs under farmer
conditions and include the level of disease resistance and the market acceptance of the
RVs.

In this paper, farm survey data from Honduras are used in probit analysis to
investigate the significance and magnitude of “access” factors vs. “varietal/breeding”
factors in the adoption of disease-resistant bean varieties in Honduras. The results of the

probit analysis have implications for whether factors under the control of breeders or

71

policy-makers would have more leverage to alleviate remaining constraints to more
widespread adoption of improved bean varieties in Honduras, or whether a combination
of actions is most complementary to greater adoption.
2. Background
2.] Organizations

Honduras’ two bean research programs, which are implemented by Zamorano
(Programa de Investigaciones en Frijol, Escuela Agricola Panamericana) and DICTA
(Direccion de Ciencia y Tecnologia Agricola, the Honduran National Bean Program),
work in collaboration with three organizations: the USAID-funded Bean/Cowpea
Collaborative Research Support Project (CRSP); the Programa Cooperativo Regional dc
Frijol para Centroamerica, Mexico, y El Caribe (PROFRIJOL), funded by COSUDE, the
Swiss Agency for Development and Cooperation; and CIAT (Centro Intemacional de
Agricultura Tropical). This paper utilizes data from a farm-level survey of bean farmers
(N=210) that was implemented by the CRSP and PROF RIJOL in January-February 2001,
in collaboration with Zamorano and DICTA. The survey targeted the Mideast (El Paraiso
and Francisco Morazan departments) and Northeast (Olancho) regions, which together
account for about one-half of annual bean production in Honduras.
2.2 BGYMV-Resistant Bean Varieties

Bean Golden Yellow Mosaic Virus (BGYMV) is the principal bean disease in
Honduras, and one of the main production constraints in Honduran valleys (Martel,
1995). The virus is transmitted by the whitefly species Bemisia tabaci. which is normally

found below 1,000 meters in all growing regions of Honduras and more frequently in the

72

drier postrera season (September/October to December/January). Zamorano and DICTA
have developed several varieties with resistance to BGYMV, the first of which was
released in 1990 (Dorado), the second (Don Silvio) in 1993. However, the economic
gains from yield loss averted through use of Dorado or Don Silvio are at least partially
offset by price discounts in the market of approximately 15 percent (Martel, 1995; Mather
et al, 2001). The newest RV (Tio Canela), released in 1997, faces a smaller market price
discount of about 7 percent (Mather et a1, 2001). While two other RVs include DICTA
113 and DICTA 122, their adoption rates to date are negligible (ibid, 2002).
2.3 Seed Distribution Channels

RV bean seed is disseminated through various channels, although 57 % of RV
adopters first obtain seed from neighbors (Mather et al, 2001). While DICTA supported
an artisan seed program in the early 19905, this program was relatively small in scale, and
focused on Dorado. Both Zamorano and Hondugenet sell certified seed directly to
farmers, which is packaged in 50 lb bags and sold exclusively from Zamorano’s campus
and Hondugenet’s facilities in Tegucigalpa (the capital). Zamorano and the Department
of Natural Resources each have additional extension services in some villages which
involve RV promotion, demonstration, and/or access to seed. Zamorano’s extension
work has primarily promoted Tio Canela. In addition, various NGOs, such as the
Catholic Church, distribute improved seed to project participants and support the
development of local farmer seed banks across the country, although they are typically

less active in the Mideast and Northeast than in other regions.

73

2.4 Post-Mitch Seed Distribution

Following the devastation caused by hurricane Mitch in October 1998, the
USAID-funded Post-Mitch Agricultural Recovery for Honduras project represented the
most significant extension of improved bean technology since the public extension
service was eliminated in the late 19805. With funding from USAID and the British
Department for International Development, Zamorano and 10 independent farmers with
irrigation and seed production experience were contracted to multiply a total of 97 mt of
RV bean seed (Tio Canela, Dorado, and Don Silvio—approximately equal amounts of
each variety) in the verano season (December to May), which was then distributed in 10
lbs bags by 41 NGOs to small- and medium-size farmers for planting in 1999 during the
primera (May/June to July/August), the rainy season, and postrera (September/ October
to December/January), a drier season (EAP, 2000; Mainville, 2000). A separate
component of this USAID project also funded demonstration plots throughout the country
in 1999-2000, which helped to introduce farmers to RVs and Zamorano’s recommended
management practices and complementary inputs. While this project may have exposed
many farmers to improved varieties for the first time, only 9% of our sample received RV
seed from Zamorano or an NGO after Mitch. However, this is not surprising since much
of the NGO assistance was targeted to other regions of Honduras, which were hit harder
(and are generally poorer) than Mideast and Northeast Honduras.
2.5 Survey Farmers

Bean producers in Honduras are predominantly small farmers, the majority of

whom use fertilizer and insecticide, and market at least some of their bean production

74

(Tables 1a and 1b). Most of the sample farmers (70 percent) live in villages which have
faced BGYMV pressure in the past (Table 1b), and also which have benefitted from
extension activities including promotion, demonstration, and/or access to RVs (58
percent). These data suggest that the characteristics of RV and TV growers are quite
similar. Honduran farmers typically plant two crops during the year. In the primera,
maize is the principal crop', and beans are either intercropped with maize or
monocropped.2 During the postrera, beans are almost exclusively monocropped.

Because the whitefly population is often highest under drier conditions, BGYMV is most
frequent and severe in the postrera. The adoption analysis that follows only considers the
postrera season, as this is the primary bean production season and nearly all of the sample

farmers planted in the postrera.

 

' Typically, farmers plant beans in the primera in order to multiply seed for their
postrera planting.

2 In the following analysis, intercropped area is converted to “effective monocrop
bean area” given information from each respondent on his/her intercropping system.

75

Table la. Characteristics of Sample Bean Farmers, Postrera 2000, Honduras

 

 

 

 

 

 

 

 

 

 

 

RV Adopters ‘3 Nonadopters All farmers
Characteristic (N=79) (N=123) (N=202)
( Mean or % )
Farm size (ha) 12.8 10.8 11.5
Bean area (ha) 2.35 2.28 2.14
Education (years in school) 3.2 2.9 3.01
Age (years) 49 49 49
Altitude (m) 332 b 964 b 913
Expected September rainfall (mm) 216 212 214
Used formula fertilizer (%) 44 50 48
Used urea fertilizer (%) 53 ° 40 ° 45
Used both formula and urea (%) 28 23 25
Used no fertilizer (%) 31 33 32
Used insecticide (%) 75 b 64 b 68
Used fungicide (%) 17 d 34 d 20
Intercropped (%) ' 23 d 11 d 15
Source: CRSP/PROFRIJOL Farmer Survey, 2001 (N=210)
a 19% of farmers used both an RV and a TV (N=38)
b significantly different (5% level) across RV use using two-sided t-test
° significantly different (10% level) across RV use using two-sided t-test
d significantly different (10% level) across RV use using one-sided t-test‘

 

 

76

Table lb. Characteristics of Sample Bean Farmers, Postrera 2000, Honduras

 

 

 

 

 

 

 

 

 

 

RV Adoptersa Nonadopters All farmers
Characteristic (N=79) (N=123) (N=202)
( Mean or % )

Marketed beans (%) 56 63 60

Sold 1 to 50% of harvest (%) 21 21 21

Sold > 50% of harvest (%) 34 42 39

% farmers in village with BGYMV b (%) 84 ° 62 ° 70
% farmers in village with extension: °

DICTA artisan seed program (%) 24 g 15 3 19

Dept. of Natural Resources (%) 32 32 32

Zamorano (%) 16 12 14

Catholic Church (%) 18 16 17

Received RV seed from Post-Mitch org (%) 13 ° 2 c 6

No extension at village level (%) 46 40 42

Planted an RV, Postrera ‘96 49 ° 7 ° 24

Mean village RV adoption rate, Postrera ‘95 d 29 ° 15 ° 20

Mean village RV adoption rate, Postrera ‘98 d 37 ° 19 ° 26

 

 

Source: CRSP/PROFRIJOL Farmer Survey, 2001 (N=210)
’ 19% growers used both an RV and a TV (N=3 8) b defined by Dr. J .C. Rosas, Director,
Programa de Investigaciones en Frijol, Zamorano c extension which includes promotion,

demonstration, and/or extension of improved varieties d calculated for each farmer as the
ratio of RV users to non-RV users in his/her village (N=7), not including the farmer

° significantly different (5% level) across RV use using two-sided t-test
f significantly different (10% level) across RV use using two-sided t-test
3 significantly different (10% level) across RV use using one-sided t-test

 

77

 

3. Adoption Model

In the following probit analysis, adoption of an RV is defined as planting an RV
in postrera 2000. The adoption model builds from Martel’s (1995) logit model
specification of Dorado adoption, which included such variables as: age, education, farm
size, fertilizer use, administrative and topographical region. This paper not only uses a
more recent farrnlevel survey, but also three additional variables in the varietal decision
model: market participation, extension, and the village RV adoption rate. In addition, this
adoption analysis will differentiate between the older RVs (Dorado and Don Silvio) and
the newest RV (Tio Canela) because of the expected differential influence on farrners’
varietal choice with respect to market participation, extension, and village RV adoption
rates.

A majority of survey farmers market at least some of their postrera harvest (Table
2), with two-thirds of postrera bean sellers selling more than half their harvest (Mather et
al, 2001). Market participation may affect a farmer’s varietal choice depending upon
his/her expected yield loss to disease as well as his/her risk preferences. For example,
given that sizeable price discounts for RVs exist in the market, farmers in low-disease
pressure areas who sell beans would not likely grow RVs because the potential yield loss
averted through RV use would not compensate for the income lost to the market price
discount. In this case, participation in the market (and the level of participation) would
be expected to negatively affect a farmer’s decision regarding RV use. In addition,
according to our survey results, current RV users (and RV disadopters) prefer the culinary

characteristics of the traditional varieties (Mather et al, 2001). Therefore, farmers who

78

don’t sell beans — net bean buyers or subsistence farmers — would only grow RVs if their
expected disease losses put the household below their threshold consumption
requirements. On the other hand, assuming that the expected yield losses in disease-
intense areas are greater than the market price discount, then market participation by
farmers in these areas would not be expected to have a negative effect on adoption
(although we cannot say whether the expected effect for these farmers would be
negligible or positive).

We use two terms to measure the effect of market participation on adoption. The
first, a dtunmy for “low participation” sellers who sold between 1 to 50% of their postrera
2000 harvest, represents those farmers who are likely selling unexpected surplus from
their production over and above that intended for home consumption. The second, a
dummy for “high participation” sellers who sold more than 50% of their postrera 2000
harvest, represents farmers who produce and market beans as a cash crop, or farmers who
have no choice but to sell at harvest to repay input or other loans (and must later purchase
beans for home consumption on the market).

We also include village-level dummies for those villages which received
extension including promotion, demonstration, and/or access to RVs. DICTA extension
is expected to have a positive influence on Dorado/Don Silvio3 adoption, whereas
Zamorano extension is expected to have a positive influence on Tio Canela adoption. To

capture the effect of the Post-Mitch seed distribution, we include a dummy for those

 

3 From this point on, “Dorado” will refer to both Dorado and Don Silvio. These
varieties were both released in the early 19903, promoted by DICTA and face similar
market price discounts.

79

farmers who received RV seed from a Post-Mitch program NGO. In place of Martel’s
topographical region variable (valley vs. mountains), we use a village dummy for areas
which have been known to have faced BGYMV pressure in the past. We also add
village-level altitude as a separate measure of disease pressure.

In the Dorado probit, village RV adoption rates for postrera 1995 are used as yet
another measure of historical BGYMV pressure in the village. The village RV adoption
rate in 1995 should be a good indicator of long-term expected disease pressure, since
farmers would not likely continue to grow the old RVs — regardless of market
participation status — unless their expected disease losses were in excess of the price
discount (or in excess of subsistence requirements). Thus, villages with higher adoption
rates of these old RVs in 1995 very likely had high levels and frequency of BGYMV
pressure that same year.

The village RV adoption rate is computed for each farmer as the ratio of RV
adopters in a given village to non-adopters (not including the farmer). Given this form of
calculation, this variable also serves as a proxy measure of the given farmer’s informal
access to both RV seed as well as information about RV’s agronomic and market
performance under local conditions. Thus, the village RV adoption rate proxies the
extent to which farmer-to-farmer exchange of information and seed influences RV
adoption. The informal nature of the bean seed system in these regions of Honduras is
highlighted by the fact that in postrera 2000, 89% of the sample farmers either planted
saved seed (59%) or obtained seed from a neighbor (29%), while only 14% was obtained

from government extension, an NGO, a trader, or an input supplier together (Mather et al,

80

2001). Because the DICTA artisan seed program was most active in the early 19903, and
because many current Dorado growers had adopted the variety by 1996, we chose the
postrera 1995 season for the village-level RV adoption rate to estimate the relative
influence in 1995 of extension (DICTA), village-level farmer-to-farmer information and
seed access, and village-level disease pressure (both proxied by the village-level RV
adoption rate, postrera 1995) on current Dorado use.

As just mentioned, current Dorado use is likely to be highly correlated with prior
use (i.e., most Dorado growers in 2000 are not new adopters). Thus, we include a dummy
for those farmers who grew an RV in postrera 1996 (RV_96), in order to capture
unobservable farmer characteristics which may have influenced RV adoption prior to
1996.

In the Tio Canela probit, village RV adoption rates for postrera 1998 are used as
another measure of historical BGYMV pressure in the village. Given that adoption of
RVs up to postrera 1998 involved primarily Dorado and Don Silvio, this use of the
village adoption rate is principally as a measure of historical disease pressure. We chose
the postrera 1998 season because this planting occurred one month before Hurricane
Mitch hit Honduras (in October, 1998). This timing gives us a measure of RV use by
village prior to the Post-Mitch RV seed distribution, and also prior to widespread Tio
Canela dissemination (Tio Canela was released in 1997). To capture the effect of the
Post-Mitch RV seed distribution in 1999, we include a separate dummy in the regression

for farmers who received RV seed from the Post-Mitch program.

81

If Tio Canela has better agronomic and/or market performance than the older
RVs, we would expect Tio Canela to be adopted both in areas with lower village RV
adoption rates, as well as in historically disease-intense areas (as a replacement for the
older RVs). Thus, we would expect that Tio Canela adoption would be less influenced by
market participation and disease pressure than that of Dorado. For a proxy of farmer-to-
farmer exchange of Tio Canela seed and information, we compute a village-level Tio
Canela adoption rate for postrera 1999.
4. Estimation of adoption equations
4.1 Dorado

We estimate two probits for both Dorado and Tio Canela; one with the RV_96
dummy and one without. The following discussion of estimation results refers to the
specifications which include the RV-96 dummy, unless otherwise indicated. Martel’s
(1995) logit model, based on a 1993 survey, found Dorado adoption to be neutral with
respect to education and farm size, positively correlated with fertilizer use, the mideast
region (El Paraiso and Francisco Morazan departments), and valley topography, and
negatively associated with age. The results of our probit models4 (Table 2) show that the
adoption of Dorado (N=52) and Don Silvio (N=8) (all referred to as Dorado from here
on) is neutral with respect to farm size, which concurs with Martel (1995). Another point
of concurrence is that our dummy indicator for disease pressure (BGYMV) is positively

correlated with Dorado use, yet our continuous indicator (altitude) is surprisingly positive

 

" We apply Huber/White’s procedure to obtain robust standard errors for each
probit.

82

Table 2. Probit Estimates of Dorado Adoption, Postrera, 2000

 

Factors

Dorado ' Adoption (Dorado=l , other=0)

 

Department = F.Morazan
Department = Olancho

Tio Canela
Dorado a

Education (years)

0.140 (1.10)
0.029 (0.22)

-0.234 (-2.74)**"‘

0.027 (2.14)"

0.070 (0.58)
0.063 (0.46)

- 0.231 (~2.98)*"

0.021 ( l .69)*

 

 

Age (years) 0.002 (0.60) 0.001 (0.37)
In (Farm size) (ha) 0.019 (0.81) 0.005 (0.21)
Formula Fertilizer (0=no; l=yes) -0.290 (2.76)"* - 0.287 (-2.78)*"”"
Urea Fertilizer (0=no; l=yes) 0.017 (0.18) - 0.033 (-0.35)
Both Formula and Urea (0=no; l=yes) 0.132 (0.85) 0.236 (1.41)
Insecticide (# of applications) 0.073 (3.32)"* 0.071 (3.38)*"”"
Fungicide (# of applications) - 0.020 (-0.63) - 0.021 (-0.68)
Intercropped (0=no; l=yes) 0.186 (1 .63)‘ 0.175 (1.51)
In (Expected September rainfall) (mm) 0.351 (1 .92)"”'l 0.265 (1.43)
ln (Altitude) (m) - 0.060 (-0.35) 0.062 (0.34)
BGYMV area (0=no; l=yes) 0.168 (1.54) 0.238 (2.50)"*
Sold 1 to 50% of harvest (0=no; l=yes) - 0.031 (-0.39) - 0.052 (-0.67)
Sold 51 to 100% of harvest (0=no; l=yes) - 0.155 02.09)" - 0.141 (1.83)*
Village-level Extension: DICTA 0.278 (2.44)" 0.169 (1.54)

Zamorano 0.092 (0.81) 0.077 (0.61)

Dept of Natural Resources -0.134 (-1.64)* - 0.122 (-1 .63)*

Catholic Church 0.204 (1.50) 0.139 (1.05)
Received Post-Mitch RV seed ‘99 0.057 (0.40) - 0.032 (0.24)
Village RV adoption rate, Postrera ‘95 (%) 0.775 (4.51)""' 0.531 (2.97)"'"
Planted an RV, Postrera ‘96 — 0.465 (5.00)"*
’ includes Don Silvio (N=8) N = 202 N = 202

*** significant at the 0.01 level
" significant at the 0.05 level

"' significant at the 0.10 level
Coefficients calculated as dF / dX
(z stat in parentheses)

 

Wald chi2(20) = 68.8
prob > chi2 = 0.0000
Log Likelihood= -
82.1

Psueudo R2 = 0.322

 

Wald chi2(20) = 99.2
prob > chi2 = 0.0000
Log Likelihood= -
71.7

Psueudo R2 = 0.408

 

83

 

Table 3. Probit Estimates of Tio Canela Adoption, Postrera, 2000

 

Factors

Tio Canela Adoption (T.Canela=l , other=0)

 

Department = F.Morazan

Department = Olancho

Tio Canela
Dorado ‘

Education (years)
Age (years)

In (Farm size) (ha)

Formula Fertilizer (0=no; l=yes)
Urea Fertilizer (0=no; lacs)
Both Formula and Urea (0=no; l=yes)

Insecticide (# of applications)
Fungicide (# of applications)

Intercropped (0=no; l=yes)
ln (Expected September rainfall) (mm)

In (Altitude) (m)

BGYMV area (0=no; l=yes)

Sold 1 to 50% of harvest (0=no; l=yes)
Sold 51 to 100% of harvest (0=no; l=yes)
Village-level Extension: DICTA

Zamorano

Dept of Natural Resources

Catholic Church

Received Post-Mitch RV seed ‘99

Village RV adoption rate, Postrera ‘98(%)
Village T.Canela adopt. rate, P05 ‘99 (%)
Planted an RV, Postrera ‘96

0.125 (1.52)
0.132 (2.10)"

- 0.069 (-2.68)*"

0.007 (1.83)*
- 0.000 (-0.26)

0.006 (0.61)

0.019 (0.52)
0.098 (2.11)"
- 0.061 (4.83)“

0.015 (2.08)"
0.013 (1.72)*

0.062 (1.14)
-0051 (-O.83)
0.086 (1.60)

0.016 (0.46)
0.014 (0.40)

- 0.016 (-051)
0.005 (0.16)
0.098 (1.32)

- 0.038 (-152)
- 0.049 (-1.56)
0.278 (2.97)*"
0.106 (1 .92)*
0.298 (1.93)"

0.117 (1.47)
0.138 (2.21)"

- 0.074 (-3.13)***

0.007 (1.79)"I
- 0.000 (-0.34)

0.004 (0.37)

0.016 (0.46)
0.094 (2.05)"
- 0.058 (-1 .82)"

0.016 (2.26)"
0.011 (1.58)

0.066 (1.22)
-0.066 (-1.08)
0.090 (1 .70)*

0.014 (0.43)
0.009 (0.29)

- 0.018 (-0.64)
0.001 (0.02)
0.113 (1.46)

- 0.042 (-1.79)*
- 0.049 (-1.67)"‘
0.177 (2.33)m

0.100 (1 .88)*

0.260 (1.76)*
0.064 (1.69)‘

 

 

' includes Don Silvio (N=8)

*** significant at the 0.01 level
** significant at the 0.05 level

* significant at the 0.10 level
Coefficients calculated as dF / dX

(z stat in parentheses)

 

N = 202

Wald chi2(20) = 59.7
prob > chi2 = 0.0001
Log Likelihood= -
46.1

Psueudo R2 = 0.356

 

N = 202

Wald chi2(20) = 56.8
prob > chi2 = 0.0003
Log Likelihood= -45.2
Psueudo R2 = 0.369

 

84

 

and insignificant.

In contrast to Martel, our results indicate that there is no longer a regional
disparity in Dorado adoption (northeast region adoption no longer significantly lags that
of the mideast), age is not significantly correlated with adoption (and its sign is positive),
and education has a significant and small positive effect on adoption. A more interesting
contrast is that our results show that formula fertilizer use has a significant and negative
effect on adoption (Martel did not distinguish between different types of fertilizer).
However, the effect of urea use is insignificant and of much smaller magnitude, while
there is a positive (yet insignificant) effect of combined urea and formula use on
adoption.

The coefiicient on a dummy indicating Tio Canela use is unsurprisingly
significant, large, and negative. If Tio Canela does outperform the old RVs in agronomic
and/or market acceptance terms, we would expect to see old RV users switch to Tio
Canela. However, Tio Canela was only recently released, 19% of current old RV users in
our sample had not heard of Tio Canela, and 48% indicated that they did not have access
to seed or sufficient information about the variety.’ The coefficients of both market
participation dummies are negative, as expected, and the magnitude of the “high” market
participation dummy is larger than that of the “low” market participation dummy.

However, only the “high” participation coefficient is statistically significant (10% level),

 

5 “No access” was defined as those growers who indicated that they: a) had heard
of Tio Canela; and b) responded to the question of why they hadn’t planted Tio Canela by
saying either that they didn’t have access to the seed or didn’t know enough about the
variety.

85

and the magnitude is not large compared with other variables. This implies that market
participation has served as a significant disincentive to Dorado adoption.

The RV-96 dummy itself is highly significant and of large magnitude, as
expected. This dummy captures the influence of unobserved variables related to the
farmer’s characteristics and production environment in 1996 that were associated with his
decision to use an RV in that year. The inclusion of the RV-96 dmnmy has little effect on
the variables mentioned above — such as education, age, farm-size, fertilizer use, etc. —
which lends credibility to the interpretation of these coefficients as either their marginal
effect on new RV adoption since 1996 or their effect on continued RV use.

The coefficient on the village RV adoption rate (postrera 1995) is positive,
significant at the 1% level, and is also the largest in magnitude. In addition, the dummy
for historical BGYMV pressure is also significant and of considerable magnitude.
Together, these results demonstrate that the older RVs are more likely to be used in areas
of historical BGYMV pressure. The fact that the coefficient on the BGYMV dummy is
of much smaller magnitude than that on village RV adoption rate in 1995 demonstrates
the impreciseness of using one category to describe disease pressure that varies by
frequency, intensity, and timing over space and time.

The inclusion of the RV-96 dummy is more pronounced on the RV seed and
information access variables: the village RV adoption. rate for 1995 and DICTA
extension. In the model without RV-96, the coefficient on DICTA is 0.278 and
significant at the 5% level, while that for the village RV adoption rate for 1995 is 0.775

and significant at the 1% level. However, the DICTA coefficient in the model with RV-

86

96 is not significant" and is nearly half the size at 0.169, while the village RV adoption
rate in 1995 is still significant at the 1% level but is smaller at 0.531. The reason for this
change is that the effect of DICTA extension on current Dorado use is diminished once
we account for the effect of DICTA extension on RV use in 1996 (RV_96). Thus, we are
modeling two channels of influence on current Dorado use. First, DICTA extension in
the early 1990s influenced early adoption (RV_96), which has resulted in continued
Dorado use for some farmers. Second, DICTA extension increased RV seed information
and access in some villages, which, over time, increased the potential for farmer-to-
farrner exchange both in those and other villages. Likewise, the effect on current Dorado
use of a village’s RV adoption rate in 1995 is diminished once we account for the effect
that this village RV adoption rate has on RV use in 1996 (RV_96).

These results suggest that while farmer-to-farmer exchange has a larger effect on
farmer adoption than does extension, extension plays a role in helping to jump-start the
dissemination process. This is consistent with the descriptive statistics cited above
concerning the informal nature of the bean seed system in Honduras. Finally, the effect
of the Post-Mitch program (a farmer-specific, not village level variable) is not significant
for Dorado adoption because most of our sample farmers who received RV seed from a
Post-Mitch program received Tio Canela, and those who received Dorado seed were

already Dorado growers.

 

6 Although the coefficient on DICTA extension is not significant once RV_96 is
included, RV_96 and DICTA are jointly significant at the 1% level in a LR test.

87

4.2 T in Canela

The following discussion of Tio Canela estimation results refers to the
specifications which include the RV-96 dummy, unless otherwise indicated. The results
of our probit model for Tio Canela (Table 3) show that its adoption (N=22) is neutral
with respect to farm size, education, and age.7 In contrast with Dorado, there is a slight
regional disparity in Tio Canela adoption (adoption is 14% more likely in Olancho). In
addition, Tio Canela adoption shows nearly the opposite correlation with fertilizer use
from that of Dorado. The coefficient on formula fertilizer use is of negligible magnitude
and is insignificant, the effect of urea use is significant (5% level) and positive, and that
of combined urea and formula use on adoption is significant (10% level) and negative,
but of small magnitude.

The coefficient of the dummy indicating continued Dorado use is negative, as
expected, although only half the magnitude of the corresponding coefficient (for Tio
Canela) in the Dorado adoption function. As mentioned above, nearly two-thirds of
current Dorado growers do not have sufficient information about and/or seed access to
Tio Canela. The coefficients of both market participation dummies are highly
insignificant and small in magnitude. This implies that, in contrast to Dorado, the smaller
market discount for Tio Canela has not served as a significant deterrent to its adoption.

A more significant contrast to Dorado is seen in the coefficients on disease

pressure variables. The coefficient on the historical BGYMV pressure dummy is

 

7 Although the coefficient on education is significant at the 10% level, its
magnitude is negligible.

88

insignificant and of negligible magnitude (while the same coefficient in the Dorado probit
is 0.238 and significant at the 1% level), which indicates that Tio Canela is being adopted
outside of traditionally disease-intense areas. Even more striking, the coefficient on the
village RV adoption rate for 1998 is significant (10% level) but of much smaller
magnitude than that of Dorado (0.10 for Tio Canela compared with 0.53 for Dorado).
This offers further evidence that farmers in lower-disease pressure areas are adopting Tio
Canela, while also suggesting that some Dorado growers are switching to Tio Canela
Further contrast between Dorado and Tio Canela adoption is seen in the magnitude of the
RV_96 variable, which is 0.47 for Dorado and 0.064 for Tio Canela. Thus, while some
Tio Canela adopters were previously Dorado users, the unobserved factors that explained
Dorado use in 1996 are still very important in explaining current Dorado use, yet have
little influence on Tio Canela adoption.

The variables which measure RV seed access and information - Post-Mitch RV
seed, Zamorano extension, and village-level Tio Canela adoption rate in postrera 1999 -
are correlated in various ways. First, Zamorano multiplied the seed for the Post-Mitch
NGO dissemination, and many of these NGOs work in the same villages where Zamorano
has previously offered extension. Second, as was the case with DICTA extension and
Dorado use in villages, we would expect that extension would lead to higher village
adoption rates. In addition, our Post-Mitch dummy only captures the direct effect of this
program — those farmers who received seed direct from a Post-Mitch NGO in 1999. Yet,
other farmers may have received Post-Mitch program RV seed in 2000 indirectly through

program participants (in 1999). Thus, the village Tio Canela adoption rate in 1999 may

89

also capture an indirect effect of Post-Mitch seed dissemination.

Unsurprisingly, the effect of the Post-Mitch program is significant (5% level)
because several of the Tio Canela users in our sample received the variety for the first
time from the Post-Mitch program.8 The effect of Zamorano village-level extension is
positive, of a magnitude similar to that of the Post-Mitch program, yet is insignificant.’
However, Zamorano extension is correlated (0.41) with the variable for village-level Tio
Canela adoption rate in 1999, and these two variables are jointly significant at the 1%
level in a LR test. '0 Given that the market participation and disease pressure variables
are either insignificant or small in magnitude, the fact that the largest coefficient in the
model is the village-level Tio Canela adoption rate in 1999 suggests that the main

constraint to increased Tio Canela adoption is increased seed information and access.

 

8 It is not clear why the magnitude of the Post-Mitch dummy coefficient
diminishes from 0.278 to 0.177 when the RV_96 dummy is included in the regression
(when no other variable coefficient in the model experiences notable changes). This
implies that some unobservable factors that explain RV use in 1996 also explain how or
why the farmer has access to the Post-Mitch NGO program.

9 It is not surprising that the Zamorano extension magnitude is smaller (0.11) than
that of the Post-Mitch program (0.177) because the latter is a farmer-level variable
(indicating whether or not the farmer himself received RV seed) whereas the former is a
village-level variable (indicating whether or not the village received RV demonstration,
information, and/or seed access).

'0 When the village Tio Canela 1999 adoption rate is not included in the model,
the coefficient on Zamorano extension is much larger at 0.220 and is significant at the 1%
level. However, the Post-Mitch dummy actually becomes insignificant and declines in
magnitude to 0.108.

90

4.3 Breeding Factors

With respect to breeding factors, the influence of market price discounts for
Dorado are reflected in the negative effect of market participation on Dorado adoption.
In addition, the large magnitude of the coefficient for the 1995 village RV adoption rate,
and the smaller but significant coefficient on the BGYMV dummy, indicate that Dorado
is primarily used in areas of historically high disease pressure due its market price
discount. By contrast, there is no significant effect of market participation on Tio Canela
adoption, the BGYMV dummy for Tio Canela is insignificant, and the effect of the 1998
village RV adoption rate for Tio Canela adoption is much smaller, all of which
demonstrate that Tio Canela is being adopted both in areas of high and low disease
pressure.
4. 4 Access Factors

With respect to access to Dorado, DICTA extension presence in a farmer’s village
has a much smaller effect on the probability of Dorado adoption (about one-fourth) than
the village RV adoption rate in postrera 1995. However, Zamorano extension presence
has only half the effect on the probability of Tio Canela adoption compared with the
village Tio Canela adoption rate in postrera 1999. These results are not surprising given
that Dorado was released over 10 years ago, and extension typically has a larger effect
early in the dissemination process (as is seen with Tio Canela). At this point in time,
access to Dorado is not a principal constraint to increased Dorado impact as 61% of our
survey respondents have planted Dorado at one point, and only 16% of those who haven’t

planted it indicate that they never did so due to lack of seed and/or information (Mather et

91

al, 2001). On the other hand, Zamorano extension would be expected to have a large
effect on adoption (relative to farmer-to-farmer exchange) at this early stage of its
dissemination process. Tio Canela has only been planted by 21% of the survey
respondents, and 47% of those who haven’t planted cite lack of seed and/or information
as the reason. Given that the coefficients of the market participation and disease pressure
variables in the Tio Canela probit are either insignificant or small in magnitude, the main
constraint to increased Tio Canela adoption seems to be increased seed information and
access.
5. Conclusions

In this paper, farm survey data from Honduras are used in probit analysis to
investigate the significance and magnitude of “varietal/ breeding” factors vs. “access”
factors in the adoption of disease-resistant bean varieties in Honduras. The results
indicate that increased adoption of Dorado and Don Silvio are more constrained by poor
market characteristics (breeding factors) than lack of access to seed and information. Yet,
farmers’ continuing use of Dorado since the early 19905 — primarily in areas of
historically high disease pressure — demonstrates the positive farm-level benefits of yield
loss averted due to its disease resistance. However, the poor market characteristics of
these early-generation RVs, reflected in average price discounts of 15% relative to TVs,
has constrained adoption primarily to disease-intense areas. In addition, the market price
discounts have reduced the net income gains for Dorado users.

By contrast, the results of analysis of Tio Canela adoption (released in 1997)

indicate that the market acceptance (breeding) aspects of the variety appear to be

92

improved over those of Dorado, as Tio Canela is being adopted in areas of both
historically high and low disease pressure. However, the principal constraint to further
impact of Tio Canela is information and seed access. With increased access, current
Dorado users would be expected to switch to Tio Canela and enjoy larger net incomes
due to Tio Canela’s smaller price discounts. In addition, the. adoption of Tio Canela
outside of traditionally high-disease pressure areas implies that Tio Canela offers positive

net benefits for growers in low-disease pressure areas as well.

References

Escuela Agricola Panamericana (EAP/Zamorano), 2000, information received from the
Proyecto de Produccion y Distribucion de Semilla de Frijol para Pequenos y
Medianos Productores Afectados por Mitch.

Feder, Gershon, R. Just, D. Zilberman. 1985. Adoption of Agricultural Innovations in
Developing Countries: A Survey. Economic Development and Cultural Change
33:255-298.

Mainville, Denise, 2000, “Post-Mitch Seed Distribution Activity: Preliminary Analysis,”
unpublished draft of Bean/Cowpea CRSP Working Paper, Michigan State
University, East Lansing, MI.

Martel, P., Bernsten, R., Weber, M., 2000. Food Markets, Technology, and the Case of
Honduran Dry Beans. Michigan State University International Development
Papers, Working Paper No.78. Department of Agricultural Economics, Michigan
State University, East Lansing, MI.

Martel, P., 1995. A Socio-Economic Study of the Honduran Bean Subsector: Production
Characteristics, Adoption of Improved Varieties, and Policy Implications. Ph.D.
Dissertation, Department of Agricultural Economics, Michigan State University,
East Lansing, MI.

93

Mather, D., Bernsten, R., Rosas, J.C., Viana, A.R., Escoto, D. 2001. The Impact of Bean
Research in Honduras. Paper presented at the Conference on Impacts of

Agricultural Research and Development, San Jose, Costa Rica, February 4-8,
2002.

Tripp, R. 1997. The Institutional Conditions for Seed Enterprise Development. Working
Paper 105: Overseas Development Institute. London, UK.

94

CONCLUSIONS

1. Economic impact of disease-resistant bean research in Honduras

This research uses a recent farm-level survey (2001) in the two principal bean-
producing regions of Honduras to estimate the adoption rates of disease-resistant bean
varieties, the farm-level benefits of adoption, and the ex post rate of return to bean
research in Honduras from 1982-2010. The results demonstrate that investment in
breeding disease-resistant beans in Honduras has been profitable. Under the base
scenario assumptions, the ex post economic rate of return to disease-resistant bean
research in Honduras during the period 1982-2010 is 41.2 %. Survey results show that
adoption of RVs is widespread; 42% of sample bean farmers used an RV in the postrera
season of 2000. Moreover, the adoption of disease-resistant bean varieties in Honduras
has benefitted primarily small farmers. Therefore, these results support previous claims
that improved bean varieties are an ideal means by which to improve small farm
household income and food security in Central America — they are scale-neutral,
reproducible on-farm, and do not require complementary physical or human capital inputs
(Martel et al, 2000).
2. Methodological contributions

This research also demonstrates two methodological contributions for the use of
farm-level survey data in the construction of counterfactual scenarios in impact
assessment of maintenance research. The first approach uses farm-level and experimental

trial data within an expected utility framework to value the farm-level benefits of disease-

95

resistant varieties. This method estimates that adopters gain the equivalent of 7 to 16%
(depending on the variety) in bean income from the yield loss averted through RV use.
This calculation represents the minimum expected net gain for adopters, as it solves the
expected utility framework for the point at which TV users would switch to RV use. The
robustness of the high aggregate rate of return to bean research mentioned above is
supported by the use of this minimum expected net farmlevel gain for adopters in its
calculation.

The second method uses a modification of Lee’s (1978) two-step procedure to
correct for selection bias (Heckman, 1979; Goldfeld and Quandt, 1973), whose presence
in fannlevel survey data will likely lead to underestimation of the benefits of maintenance
research (Johnson et a1, 1999). The method is applied to test for selection bias and
estimate yield differentials between RV and TV bean yields in our survey data from
Honduras by constructing the counterfactual to RV yields as what yields RV users would
have obtained had they continued to grow TVs (i. e., the TV yield of RV users). The
corrected yield model predicts that RV growers enjoy net income gains of 13 to 19% in
the postrera season, compared to what they would have earned growing TVs; while the
uncorrected OLS model predicts that RV growers see either no income gain or even
losses by growing RVs (relative to TVs). This application demonstrates that the
implications of this method are significant for the assessment of maintenance research

impacts.

96

3. Factors influencing the adoption of RVs

Finally, this research also uses the farm survey data in probit analysis to
investigate the significance and magnitude of “varietal/ breeding” factors vs. “access”
factors in the adoption of disease-resistant bean varieties in Honduras. The results
indicate that adoption of Dorado and Don Silvio are more constrained by poor market
characteristics (breeding factors) than lack of access. Yet, farmers’ continuing use of
Dorado isince the early 19905 — primarily in areas of historically high disease pressure —
demonstrates the positive farmlevel benefits of yield loss averted due to its disease
resistance. However, its poor market characteristics, reflected in average price discounts
of 15% relative to TVs, results in little adoption —— and thus apparently no net income
benefits — for growers outside of these high-disease areas, as well as reduced net income
gains for Dorado users themselves.

By contrast, the results of analysis of Tio Canela adoption (released in 1997)
indicate that the market acceptance (breeding) aspects of the variety appear to be much
improved over those of Dorado, as Tio Canela is being adopted in areas of both
historically high and low disease pressure. However, the principal constraint to further
impact of Tio Canela is information and seed access (a policy aspect). With increased
access, current Dorado users would be expected to switch to Tio Canela and enjoy larger
net incomes due to Tio Canela’s smaller price discounts. In addition, the adoption of Tio
Canela outside of traditionally high-disease pressure areas implies that Tio Canela offers

positive net benefits for growers in low-disease pressure areas as well.

97

References

Goldfeld, S.M., Quandt, RE. 1973. The Estimation of Structural Shifts by Switching
Regressions. Annals of Economic and Social Measurement 2:475-485.

Heckman, J. 1979. Sample Selection Bias as a Specification Error. Econometrica
47 :153-161.

Johnson, N., Klass, J ., 1999. The Impact of Crop Improvement Research on Rural
Poverty: A Spatial Analysis of BGYMV-Resistant Bean Varieties in Honduras.
Paper prepared for the workshop: Assessing the Impact of Agricultural Research
on Poverty Alleviation, San Jose, Costa Rica, September 14-16, 1999.

Lee, L.F. 1978. Unionism and Wage rates: A Simultaneous Equations Model with
Qualitative and Limited Dependent Variables. International Economic Review
19:415-433.

Martel, P., Bernsten, R., Weber, M., 2000. Food Markets, Technology, and the Case of
Honduran Dry Beans. Michigan State University International Development
Papers, Working Paper No.78. Department of Agricultural Economics, Michigan
State University, East Lansing, MI.

98

|11111111111111l