Farmers' Valuation and Adoption of New Genetically Modified Corn Seeds

Permanent Link: http://ufdc.ufl.edu/UFE0024899/00001

Material Information

Title: Farmers' Valuation and Adoption of New Genetically Modified Corn Seeds Nitrogen-Fertilizer Saving and Drought Tolerance Traits
Physical Description: 1 online resource (189 p.)
Language: english
Creator: Jaramillo, Paul
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009


Subjects / Keywords: adoption, biotecnology, bounded, contingent, double, drought, fertilizer, genetically, modified, nitrogen, pay, sample, saving, selection, state, technology, tolerance, trait, transgenic, valuation, willingness
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation


Abstract: We used a Double Bounded (DB) Contingent Valuation (CV) methodology on a sample of 345 farmers from Minnesota and Wisconsin to obtain estimates of farmers Willingness-to-Pay (WTP) for corn seeds with Fertilizer Saving and Drought Tolerance traits. Two versions of each trait were presented for farmer valuation one obtained by genetic modification (GM) and one obtained via traditional breeding (nonGM). This allowed comparison of factors affecting WTP for a nonGM version with those affecting WTP for a GM version. The objective was to study adoption potential and understand which farm and farmer characteristics affect farmers WTP for these traits. In total, four traits were considered: Fertilizer Saving GM, Fertilizer Saving nonGM, Drought Tolerance GM, and Drought Tolerance nonGM. The estimated mean WTP s, in dollars per acre, were $17.25, $19.72, $18.73, and $20.87, respectively. In both Fertilizer Saving and Drought Tolerance traits, farmers are willing to pay more for the nonGM version compared to the GM version $2.47 and $2.14, respectively. At prices lower than $20 per acre, nonGM versions showed better adoption potential. Farmers specialized in nonGM corn showed willing to pay less for GM versions. Otherwise, factors affecting adoption of nonGM and GM versions were similar. The largest positive effect was observed for early adopters whom showed willing to pay between $9.61 and $14.96 per acre more than other adopters. Also, farmers were willing to pay 6 cents more for the Fertilizer Saving traits for each extra dollar they spend on fertilizer. Seed traits (GM or nonGM) seem to be regarded by farmers as imperfect ex ante substitutes (i.e., less than dollar per dollar). Results also suggest farmers see the Drought Tolerance trait as a substitute for crop insurance. Being a nonGM specialist also asymmetrically affected (reduced) the probability of participation in the Contingent Market exercise a larger non-response rate was observed for CV questions regarding the GM versions. We estimated a DB-Dichotomous Choice Sample Selection model and found no evidence of sample selection. Therefore, we hypothesize the reason for a lower participation rate was lack of familiarity with GM crop production practices and subsequent inability to construct valuations.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Paul Jaramillo.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Useche, Maria Del Pilar.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024899:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024899/00001

Material Information

Title: Farmers' Valuation and Adoption of New Genetically Modified Corn Seeds Nitrogen-Fertilizer Saving and Drought Tolerance Traits
Physical Description: 1 online resource (189 p.)
Language: english
Creator: Jaramillo, Paul
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009


Subjects / Keywords: adoption, biotecnology, bounded, contingent, double, drought, fertilizer, genetically, modified, nitrogen, pay, sample, saving, selection, state, technology, tolerance, trait, transgenic, valuation, willingness
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation


Abstract: We used a Double Bounded (DB) Contingent Valuation (CV) methodology on a sample of 345 farmers from Minnesota and Wisconsin to obtain estimates of farmers Willingness-to-Pay (WTP) for corn seeds with Fertilizer Saving and Drought Tolerance traits. Two versions of each trait were presented for farmer valuation one obtained by genetic modification (GM) and one obtained via traditional breeding (nonGM). This allowed comparison of factors affecting WTP for a nonGM version with those affecting WTP for a GM version. The objective was to study adoption potential and understand which farm and farmer characteristics affect farmers WTP for these traits. In total, four traits were considered: Fertilizer Saving GM, Fertilizer Saving nonGM, Drought Tolerance GM, and Drought Tolerance nonGM. The estimated mean WTP s, in dollars per acre, were $17.25, $19.72, $18.73, and $20.87, respectively. In both Fertilizer Saving and Drought Tolerance traits, farmers are willing to pay more for the nonGM version compared to the GM version $2.47 and $2.14, respectively. At prices lower than $20 per acre, nonGM versions showed better adoption potential. Farmers specialized in nonGM corn showed willing to pay less for GM versions. Otherwise, factors affecting adoption of nonGM and GM versions were similar. The largest positive effect was observed for early adopters whom showed willing to pay between $9.61 and $14.96 per acre more than other adopters. Also, farmers were willing to pay 6 cents more for the Fertilizer Saving traits for each extra dollar they spend on fertilizer. Seed traits (GM or nonGM) seem to be regarded by farmers as imperfect ex ante substitutes (i.e., less than dollar per dollar). Results also suggest farmers see the Drought Tolerance trait as a substitute for crop insurance. Being a nonGM specialist also asymmetrically affected (reduced) the probability of participation in the Contingent Market exercise a larger non-response rate was observed for CV questions regarding the GM versions. We estimated a DB-Dichotomous Choice Sample Selection model and found no evidence of sample selection. Therefore, we hypothesize the reason for a lower participation rate was lack of familiarity with GM crop production practices and subsequent inability to construct valuations.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Paul Jaramillo.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Useche, Maria Del Pilar.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024899:00001

This item has the following downloads:

Full Text




2 2009 Paul E. Jaramillo


3 To my family: Dad, Mom, Patty, Ivancho, Ferni, Belen, Chivi, Joaquin, Pau, and Nanitos; and to my grandfather Papa Guci and uncle Paco, may they rest in peace


4 ACKNOWLEDGMENTS For m e, this journey began five years ago, fo r my parents it began long before when they decided to dedicate their lives to afford their kids the opportunity to fulfill their dreams. I wish to thank them first and most, not only for their word s and acts of support in these years, but more so, for their unconditional love and willingness to make my dreams theirs. This is not my success, it is ours. From my parents I have lear ned, among many other things, the value of perseverance, hard work, and honesty; but more importantly, they have taught me the value of family. I can confidently say my sister Patty and my brother Ivan share my views on this. My brother and sister are my best friends. As they say the appl e doesnt fall far from the tree and in our case, Patty, Ivan and myself, are three apples that fell close and have always and will always remain close. Ivan and Patty have always been an insp iration and an example for me; I owe them much, and wish to thank them and their families immens ely for all their support in these past years. I would especially like to thank Fernando, Be len, Maria Paz, Joaquin, and Ana Paula. I wish to thank one very special person that has been my cornerstone in this journey, my soul mate and my very best friend, Mi Nanos. I am grateful to her for her support in most difficult moments, for always helping me stay in track, for her example in life, for endless hours of lending her ears to my extended expositions of new ideas for this research, for her patience, for believing in me, for giving me the drive to follow through by helping me dream a future together, and most of all for giving me th e joy of loving someone that I admire. I would also like to thank tw o other very important people that had much to do in motivating me and inspiring me. I would like to thank my grandmother and grandfather, Mama Ine and Papa Guci, for teaching me with their example and words of wisdom, at a very young


5 age, about the value of knowledge and the responsibi lity one has for using a ll the talents that God gives us. They have much to do with who I am how I see myself, and where I want to go. I would like to thank my supervisory committ ee chair, Pilar Useche, and members, Lisa House, Walter Bowen, Charles B. Moss, Renata Serra and Alfonso Flores-Lagunes. I am indebted to them for countless valuable commen ts and collaborations. Thanks are also due to professors Brad Barham and Jeremy Foltz from th e University of Wisconsin-Madison; this work would not have been possibl e without their support. Finally, I would like to thank all the faculty and staff in the Department of Food and Resource Economics at the University of Florida for their teachings and guidance. Last but not least I wish to thank my friends here at Gain esville for filling these years with good memories.


6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES ...........................................................................................................................8LIST OF FIGURES .........................................................................................................................9ABSTRACT ...................................................................................................................... .............10CHAPTER 1 STRUCTURE OF THE THESIS ............................................................................................122 GENETICALLY MODIFIED CROPS .................................................................................. 15From Nave Crop Selection to Biotech Crop Engineering .....................................................17Genetic Engineering and Genetically Modified (GM) Crops ................................................. 19Relevance, Potential Benefits, and Potential Risks ................................................................ 23Potential Benefits of Biotech Crops ................................................................................30Potential Risks of Biotech Crops ..................................................................................... 32Brief History of Early Commercialization of GM Crops .......................................................36Current Status of GM Crops ................................................................................................... 38Fertilizer Saving and Drought Tolerance Traits .....................................................................413 THEORY AND KEY CONCEPTS ........................................................................................ 47Willingness-to-Pay (WTP) and Willingness-to-A ccept (WTA) ............................................. 48Contingent Valuation (CV) ..................................................................................................... 51Elicitation Formats, Statistical E fficiency, and Starting Point Bias ................................ 52Statistical Models for Single Bounded and Double Bounded CV Data ..........................56Single bounded model: two approaches reconciled ................................................. 57Double bounded model ............................................................................................62Stated vs. Revealed Preferences ......................................................................................64Existence Value and Alternative Methods to Obtain WTP Measures of Non-Market Goods ......................................................................................................................... ..664 ESTIMATING PRODUCERS WT P FOR CORN SEED TR AITS NOT YET IN THE MARKET ........................................................................................................................ .......73Theoretical Model ............................................................................................................. ......79Willingness-to-Adopt ...................................................................................................... 81Empirical Application ......................................................................................................... ....84Data and Surveys .............................................................................................................85Factors Affecting Adoption and WTP ............................................................................. 86Farm and farmer characteristics ............................................................................... 89


7 Production data ......................................................................................................... 91Purpose of production ..............................................................................................92Early adopters and familiarity .................................................................................. 93NonGM farmers .......................................................................................................94Drought measures .....................................................................................................94Insurance costs .........................................................................................................95Results .............................................................................................................................96Fertilizer saving trait ................................................................................................96Drought tolerance trait ........................................................................................... 1005 TESTING FOR SAMPLE SELECTION BI AS IN OUR WTP ESTIMATES .................... 113Non-Response and Sample Selection in CV ........................................................................ 114Heckman Selection Model a nd Two-Stage Procedure ..................................................116Modeling Sample Selection in Single Bounded and Double Bounded CV .................. 119Issues with Data Availability and Surveying Strategies ................................................ 121Theoretical Model ............................................................................................................. ....123Initial Setup ...................................................................................................................123Double Bounded Mechanism Revisited ........................................................................ 124Sample Selection Model ................................................................................................128Empirical Application ......................................................................................................... ..132Data and Surveying Strategy: Some Highlights ............................................................132Results ...........................................................................................................................1346 FURTHER RESEARCH AVENUES ................................................................................... 143Uncertainty in Agri cultural Production ................................................................................144Strategy 1: Modifying the Envi ronment Surrounding the Plant ....................................145Strategy 2: Modifying the Genetic Material of the Plant Itself ..................................... 146State Contingent M odel of Production ................................................................................. 147Perfect Information, Perfect Timi ng, and Perfect Supplementation ............................. 149Perfect Supplementation and the State Contingent Approach ...................................... 151Seed Technologies .........................................................................................................151State-General and State-Specific Inputs ........................................................................153Genetically Modified Traits: Stat e Specific or State General? ..................................... 155Option Price ...................................................................................................................1587 CONCLUSIONS .................................................................................................................. 162APPENDIX A GM CROP REGULATION .................................................................................................. 170B RESULTS FOR AUXILI ARY ESTIMATIONS .................................................................176LIST OF REFERENCES .............................................................................................................177BIOGRAPHICAL SKETCH .......................................................................................................189


8 LIST OF TABLES Table page 2-1 Summary of genetically m odified (GM) crops approved for comm ercial growing in the U.S. in 1996 .................................................................................................................444-1 Traits considered for farmer valuation ............................................................................. 1034-2 Seed codes and variety descriptions ................................................................................ 1034-3 CV survey versions with first and second offers ............................................................. 1034-4 Description of variables selected from 2006 corn poll and 2007 corn poll ..................... 1044-5 Proportion tests and number of farmers by version of Contingent Valuation (CV) survey ........................................................................................................................ .......1054-6a Summary statistics and comparison of means (fertilizer saving trait) ............................. 1064-6b Summary statistics and comparison of means (drought tolerance trait) .......................... 1074-7 Definitions of drought measures ...................................................................................... 1084-8 Estimation results for fertilizer saving trait ..................................................................... 1094-9 Estimation results for drought tolerance trait ................................................................... 1105-1 Regions, indicators, and possible an swer sequences to DB-DC questions ...................... 1395-2 Types of respondents .......................................................................................................1395-3 Tabulation of responses to the fertilize r saving trait CV questions (nonGM and GM) ... 1395-4 Sample selection m odel estimation results ...................................................................... 140B-1 Estimation results for drought to lerance trait (insured only) ........................................... 176


9 LIST OF FIGURES Figure page 2-1 Estimated and projected population of the world by projectio n variant, 1950-2050.. .......452-2 Total fertility rates (T FR) in 2008 (estimates).. .................................................................452-3 Participation in total genetically modifi ed (GM) crop planted area, by country, 2006. .... 462-4 Global GM crop planted areas, for four ma jor GM crops, as a percentage of their global planted area. ............................................................................................................463-1 Total economic value and valuation techniques. ............................................................... 724-1 Contingent Valuation (CV) questio nnaire portion of the 2007 Corn Poll. ......................1114-2 Map of average percenta ge area affected by severe drought (D2) for counties in Minnesota and Wisconsin, 2006. ..................................................................................... 1114-3 Potential adoption curve for nonGM and GM fertilizer saving trait. .............................. 1124-4 Potential adoption curve for nonGM and GM drought tolerance trait. ............................ 1125-1 WTP regions based on the tradit ional double bounded (DB) model. .............................. 1415-2 Regions over the (iizz21,) plane implied by the double bounded dichotomous choice (DB-DC) sample selection model. ................................................................................... 1415-3 Possible changes in survey respons e behavior between the nonGM and GM questions. .................................................................................................................... .....1426-1 Representation of a state-general input and its transformation curve. ............................. 1616-2 Representation of a state-specific input and its transformation curve. ............................ 1616-3 Representation of a perfectly state-gene ral input and its transformation curve. .............. 161


10 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy FARMERS VALUATION AND ADOPTION OF NEW GENETICALLY MODIFIED CORN SEEDS: NITROGEN-FERTILIZER SAVING AND DROUGHT TOLERANCE TRAITS By Paul E. Jaramillo August 2009 Chair: Pilar Useche Major: Food and Resource Economics We used a double bounded (DB) Contingent Valuation (CV) methodology on a sample of 345 farmers from Minnesota and Wisconsin to obta in estimates of farmers willingness-to-pay (WTP) for corn seeds with fertilizer saving (FS) and drought tolerance (DT) traits. Two versions of each trait were presented for farmer valua tion one obtained by genetic modification (GM) and one obtained via traditional breeding (nonGM). This allowed comparison of factors affecting WTP for a nonGM version with those a ffecting WTP for a GM version. The objective was to study adoption potential and understand which farm and farmer characteristics affect farmers WTP for these traits In total, four traits were considered: FS-GM, FS-nonGM, DT-GM, and DT-nonGM. The estimated mean WTPs, in dollars per acre, were $17.25, $19.72, $18.73, and $20.87, respectively. In both FS and DT traits, farmers are willing to pay more for the nonGM version compared to the GM version $2.47 and $2.14, respectively. At prices lower than $20 pe r acre, nonGM versions show ed better adoption potential. Farmers specialized in nonGM corn showed willing to pay less for GM versions. Otherwise, factors affecting a doption of nonGM and GM versions were similar. The largest positive effect was observed for early adopters whom showed willing to pay between $9.61 and


11 $14.96 per acre more than other adopters. Also, farmers were willing to pay 6 cents more for the FStraits for each extra dollar they spend on fertili zer. Seed traits (GM or nonGM) seem to be regarded by farmers as imperfect ex ante substitutes (i.e., less than dol lar per dollar). Results also suggest farmers see the DT trait as a substitute for crop insurance. Being a nonGM specialist also asymmetrically affected (reduced) the probability of participation in the CV market exercise a larger non-response rate was observed for CV questions regarding the GM versions. We estimated a DB-Dichotomous Choice Sample Selection model and found no evidence of sample selection. Therefore, we hypothesize the reason for a lower participation rate was lack of familiarity with GM crop production practices and subsequent inability to construct valuations.


12 CHAPTER 1 STRUCTURE OF THE THESIS The f ocus of this dissertation is to stu dy farmer adoption decisions regarding new genetically modified (GM) seed technologies. In specific, we cons ider two new promising traits that are in the R&D pipelines of major seed comp anies, these are: (i) a drought tolerance (DT) and (ii) fertilizer saving (FS) trait. The dissertation is or ganized as follows. Chapter 2 presents an overview of the GM crop industry. The chapter is intended to provide the reader with the big picture su rrounding GM seed techno logies. The potential benefits of GM crops are outlined emphasizi ng the relevance of th e technology to utmost important welfare improvement goals set by wo rld consensus regardi ng poverty alleviation, reduction of world hunger, and environmental sust ainability. The controversial facets of the technology are not left out as we discuss several different concer ns voiced by different segments of the population. At the end of the chapter, the current situation of the GM crop industry is summarized, and the two GM traits that are th e focus of our study are introduced a FS and a DT trait. Chapter 3 reviews the relevant theory for the non-market approach used in this thesis to examine the adoption of GM traits. A non-market approach is necessary because the traits we study have not reached the market yet. The key concept of willingn ess-to-pay (WTP) and Contingent Valuation (CV) methods are discusse d in detail. The reader unfamiliar with CV methods should find this chapter very helpful. Chapter 4 builds on the theoretical concepts considered in the previous chapter and adapts them to the context of technology adoption, in orde r to empirically analyze the value of the GM traits to farmers and their adoption potential. A sample of farmers from the states of Minnesota and Wisconsin is used to obtain estimates of farmers willingness-to-adopt corn seeds with


13 FSand DT traits. Farmers were asked to consider two different versions of the same trait one obtained via conventional selective breeding (i.e., nonGM) and another by recombinant ribonucleic acid (RNA) techniques (i.e., GM). This amounts to a total of four traits: (i) nonGM, (ii) FS-GM, (iii) DT-nonGM, and (iv) DT-GM. One question we ponder is whether farmers value the agronomic aspects of the trait or focus more on the nonGM vs. GM aspect, and whether we can disentangle both aspects of th e valuation. The results obtained show the differences in adoption potential between traits. They also provide estimates of the effects of farm and farmer characteristics on the WTP for each trait. Valuable insights are gained by comparing different versions of a same trait. Th e results should prove valuable both for policy makers interested in overseeing desired adoption rates and for seed companies interested in developing efficient pricing programs. Chapter 5 explores the possibility that sample selection bias may be affecting the estimates of farmers WTP for the GM traits that we obt ained in Chapter 4. The application of a sample selection model was motivated by the observation in our sample, of a considerably larger number of non-respondents to the CV questions concerning the GM version of a trait compared to the number of non-respondents to CV questions concerning the nonGM version of the same trait. This observation led us to consider the possibility that failure by the farmer to respond to GM questions could be systemati cally related to his WTP. If th at were the case, it would imply underlying sample selection problems. Testing for sa mple selection in our results from Chapter 4 is important because if detected it would s uggest we have inconsistent estimates for the parameters of the population init ially targeted. On the other hand, if no sample selection is detected, it would provide more reliability to our findings.


14 Chapter 6 gives an analysis of GM crops under the state contingent paradigm initiated early by Arrow (1953, 1964) and Debreu (1952). The uncertainty inherent in GM crop adoption has traditionally been oversimp lified and sometimes even neglected. Analogously, the analysis of crop production in general, models uncertainty using a stochastic production function which is simply a production function with a random error term added to it. The state contingent model of production developed by Chambers and Quiggin (2000) as an alte rnative to the traditional stochastic production framework, presents a different and intuitively appealing view of farming under uncertainty. While the chapter is not inte nded to fully develop and implement a new theory, the discussion hopes to shed some light on the GM crop farmer adoption scenario.


15 CHAPTER 2 GENETICALLY MODIFIED CROPS Crop plants are constantly being im proved to ma intain a secure and sustainable supply of food in order to meet demand from current and future generations (OECD 1993). A most valid recent example of this is the use of selective breeding techni ques to develop the high yield varieties (HYVs) (or miracle seeds) that formed the basis for the Green Revolution a large scale technological diffusion pheno mena that sprouted dramatic increases in food production and averted widespread famine in As ia during the 1940s and 1960s. The interplay between food production and global population has intrigued many brilliant minds for centuries. In the 1700s, Malthus hypothesized that hunger and famines would ceaselessly plague humanity as population growth, driven by fertility rates, would inevitably exceed food production growth. Gregory Clark (2007) has recently formalized Malthus ideas in a mathematical model showing technological adva nces can, at best, suppor t larger quantities of individuals, but cannot improve the long run level of material wellbeing. If fertility rates run amuck, showing a large and sustained increase, the short run material wellbeing may drop drastically and very high death rates may ensue. Many places across the gl obe today, particularly in poor developing countries, presen t fertility and death rates that are not far from this scenario. One of the most important challenges for the coming decades will be to find ways to achieve a sustainable increase in food production in order to a lleviate world hunger and eliminate extreme poverty (United Nations 2008), while at the same time recognizing: (i) that in the most likely scenario human population will continue to grow until it reaches a supposed peak of 9.2 billion in 2050; (ii) that most of this increase in population will happen in developing countries which are already the most affected by hunger, ma lnutrition, poverty, and disease; and (iii) that, given the state-of-the-art, the current scale of economic and agricultural activity is already


16 causing important destruction and changes in the e nvironment that threaten our ability to achieve an increase in food production that is sustainable. A growing consensus exists in the global de velopment and scientific community that the current state-of-the-art based on conventional br eeding technologies will not suffice to achieve the increase in food production needed to meet fu ture demand in 2050. In his lecture as a Nobel Peace Prize recipient, Norman Borlaug known as one of the fathers of the Green Revolution has expressed his view that the Green Revolutio n has only won us a temporary success in mans war against hunger, one that may be sufficien t only for the three decades succeeding it. Genetically modified (GM) crops1, improving upon conventional breeding techniques in terms of precision, scope, and speed of seed improvement s, present themselves as a promising new technology that could complement selective bree ding techniques in our search for future sustainable food security. GM crops, however, like many other innovative technologies, are not without controversy. Many uncertainties surround the GM crop topic. Opponents of biotechnol ogy do not discard the claimed benefits of GM crops, but argue that not enough is known about the environmental safety and health risks associat ed with these foods and that they should be more rigorously controlled (U.S. General Accoun ting Office 2002). Other opponents ha ve ethical concerns about manipulating the genetic material of living beings. A last vari ety of opponents, believes in the benefits of GM crops, but holds some doubts about how these should be produced, by whom (private or public), and what the net effects will be for poor developing nations. In spite of all controversies, many stil l believe biotechnology and GM crops will be essential in augmenting our chance s of securing adequate levels of food production in the future 1 Also known as biotech crops or transgenic crops.


17 while at the same time minimizi ng further stress to the environm ent. Most proponents of biotech crops cite enhanced crop yields, more envi ronmentally friendly food production, and more nutritious food as the major benefits and reasons to move forward (U.S. General Accounting Office 2002). Borlaug calls for the need of a se cond Green Revolution and endorses the use of biotechnology to develop the new miracle seed s that will be necessary (Monsanto 2005; Schattenberg 2009). This chapter provides the reader with an overview of GM crops: their discovery; their definition and the concepts from biology used to describe them; their relevance, potential benefits, and controversial facets; the motives fo r regulating their development and the different observed approaches to regulatio n; their history as a new and growing industry; the current situation of such industry; and a glimpse into tw o particularly important biotech traits that may start being marketed in the near future and that are the focus of this dissertation a fertilizer saving (FS) trait and a drought tolera nce (DT) trait (in corn seeds). From Nave Crop Selection to Biotech Crop Engineering The invention of agriculture som e 10,000 year s ago was followed by a long period of nave crop selection that lasted thousa nds of years. The first experime nters in crop selection had to actually separate edible crops from poisonous and non-edible ones. As sad a picture as this might be, early hunter-gathere rs and proto-farmers probably relie d on trial and error methods to do this.2 The second wave of crop selection, after sort ing out all edible crops, had the objective of selecting crops with higher yields and resistance to disease and pests. For centuries, nave trial and error remained the exclusive method. Early farmers would plant different seeds and hand2 In practice this brought some advantages, specifically, tr ial and error practices helped early hunter-gatherers and proto-farmers develop an extensive knowledge about the qua lities and potential uses of the local flora. This point has been argued by Jared Diamond (1 999) in his Pulitzer Prize winning book Guns, Germs, and Steel (who supported his argument on several studies in the area of Ethnobiology).


18 pick those which resulted in individuals with the most desirable traits. Improvements in crop genetic traits during this period happened at a very slow pace. It was not until the turn of the 20th century, when the experiments on inheritance of an Austrian monk named Gregor Mendel (1822-1884) were rediscovered by the German botanist Karl Erich Correns (1864-1933), that crop selection took its fi rst big turn. Understanding the rules of inheritance allowed for a more systematic appro ach to breeding based on the probabilistic processes th at were found to govern the passing of genes from one generation of crops to the next. Simple trial and error was substituted by calculated trial and error. With this, genetic improvements could be obtained in a fa ster manner by design and analysis of plant breeding experiments. The discovery led to the era of modern pl ant breeding characterized by the production of thousands of improved varie ties at a relatively accelerated pace. The importance of such innovation to global stability and human advancement cannot be overly emphasized. The Green Revolution, led by the Rockefeller Foundation along with the Ford Foundation and other major agencies duri ng the 1940s and 1960s, was largely driven by the use of selective breeding for the rapid development of HYVs. The Green Revolution was mainly a diffusion phenomena, it spread technologies that already existed in the industrialized world (in particular pesticides, irrigation technologies, and fertil izer) into the developing world. HYVs were better suited than existing indigenou s varieties to exploi t the large potential increases in productivity of using these technologies. HYVs had a hi gher nitrogen-absorbing capacity compared to indigenous varieties but were less resilien t to other pressures of the indigenous ecosystems and so they had to be bred and adapted to tailor the particularities of each environment modern selective breeding was th e key. The novel technological innovation of the Green Revolution were these so called mir acle seeds. Today, the consensus among many


19 scientists is that the Green Revolution allowed food production to keep pace with global population growth and thus it helped us avoid the Malthusian catastrophe.3 By the end of the Green Revolution in the late 1960s, Nobel Prize sc ientists Francis Crick and James Watson discovered the molecular structure of deoxyr ibonucleic acid (DNA). Their groundbreaking research and findings led to the development of biotechnology which allows scientists to identify desirable genes in one liv ing organism and transfer it to another without mating them. The application of this knowledge to crops and foods resulted in the birth of biotech crops also known as GM crops. Biot echnology tremendously in creased the potential benefits of crop improvement compared to sele ctive breeding, taking a huge leap not only in terms of speed but also in the scope and pr ecision with which crop improvement could be obtained. Prior to biotechnology, when only conventiona l selective breeding was possible, only intra-species genes could be transferred from one generation to the next via natural reproduction (including sexual and vegetative). W ith current biotechnology-based techniques, not only can desirable genes from one species be inserted into other unrelated species but the process is also sped up by artificial means of producti on such as gene guns or Agrobacterium tumefaciens. In biotechnology, crop selection has evolve d to its current form more accurately described as a crop engineering or crop production process rath er than crop selection per se. Some have termed this post-Green Revolution era the Gene Revolution. Genetic Engineering and Genetically Modified (GM) Crops GM crops4 are produced from natural crops that ha ve been conferred a specific desired trait by introduction of pre-isolated gene(s) via genetic engineering me thods. It is well 3 However, some critics of the Green Revolution argue that while impeding a food shortage in the short run, it has set us on a long-run path of population growth that is inconsistent with the Earths carrying capacity.


20 understood, in genetics, that different genes (g enotype) are responsible for the expression of different physical characteristics or traits (phenotype) in all livi ng organisms. Gene transference is possible due to the ubiquity of DNA in a ll living cells. The DNA mol ecule stores all the genetic material and information necessary for correct metabolic functioning of the organism. Genes are segments of DNA that encode neces sary information for production of a specific protein. Produced proteins then function as cat alytic enzymes regulating metabolism, or as storage units, and contribute to the expression of a specific trait. Decoding of the information in DNA by the cell requires its transcription into messenger ribonucleic acid (mRNA) molecules which then produce proteins by a process called translation. These two processes (transcription and translation) are regulated by a complex set of mechanisms so that production of a particular protein is activated only when and where it is needed (Department of Soil and Crop Sciences 2004). Thus, traits are physical characteristics co nferred, in part, to the organism by its genetic material. The term genetic engineering applies to scient ific laboratory methods that involve direct manipulation of an organisms genes. The engi neering process generally involves five basic steps: (i) isolation, (ii) insertion, (iii) tr ansfer, (iv) transformation, and (v) selection. In GM crop engineering, a final (vi) breeding step is added to produce market able seeds and evaluate gene stability. During the isolation stage a natu rally occurring gene of interest is identified that is responsible for the expression of a desired trait in a donor organism. The engineer then proceeds to insert the isolated gene into an ad equate vector such as a plasmid (i.e., Agrobacterium tumefaciens). Once the vector is obtained, it can be re produced to produce copies of the gene and 4 Alternative commonly used names include bioengineered or biotech foods (or crops) or transgenic foods (or crops).


21 then used to transfer the gene and transform the target organism. Alternatively, the use of gene guns eliminates the need for vectors by transferri ng the genes directly into the target organism. The final step in the engineering process entails the selection and separation of the successfully genetically modified organisms fr om those that failed to take up the gene. In the case of GM crops, the resulting GM organism is usually cros sed into other crop lines (via regular crosshybridization methods) that have desired commercial traits. Besides its obvious objective of producing a final seed product, this final step allows for the engi neer to evaluate the genetic stability of the newly introduced gene. Genetic engineering provides three main adva ntages over traditiona l selective breeding increased precision, increased speed, and incr eased possibilities. Fi rst, the engineering techniques avoid one of the major problems en countered by traditional crop breeders that use cross-hybridization no unwante d genes are introduced. Seco nd, sexual reproduction used in conventional breeding requires the plant to reach sexual maturity before each cross is made. Genetic engineering works with individual cells a nd does not require mature plants. Thus genetic engineering increases the speed at which a new variety with the de sired traits can be obtained. Finally, genetic engineeri ng increases the pool of po ssible traits that can be transferred since it allows for gene transfer between unrelated species. While it remains impossible to account for all po ssibly transferable traits and genes; we can present categories for those traits that have shown the most popularity among researchers and developers (James and Krattiger 1996), these ar e: bacterial resistance (BR), fungal resistance (FR), insect resistance (IR), herbicide toleran ce (HT), marker (M) genes, male sterility (MS), quality characteristics (Q), and virus resistance (VR). Promot er (P) genes should also be considered. Most of these categ ories are self-explanat ory. Quality traits include physical and


22 agronomic characteristics of the plant or fruit, like for example, delayed ripening (tomato), or production of a wanted chemical substance (bet a-carotene in rice), or pigment production (in flowers). The use of M genes and P genes is il lustrated in the following subsection by using the development of Bt-crops as an example.5 Bacillus thuringiensis commonly known as Bt, is a gram-positive bacteria that occurs naturally in soil. Early in the 20th century, the entomologist Ernst Berliner isolated a novel bacterium that had the capacity to kill the la rvae of Mediterranean flour moths in the German province of Thuringia (Pueppke 2001). The new bacterium was dubbed Bacillus thuringiensis or Bt. Bacteriologists then became aware that different strains of Bt are able to kill different insects. Bt produces proteins (kno wn as Cry proteins or -endotoxins) which, when ingested by the insect, adhere (bind) to its gut and disrupt its digestive system resulting in eventual starvation and death. Research has helped in determining which strains affect which insects. Cry proteins have been categorized (Hfte and Whitley 1989) into major classes with their respective susceptible insect families (e.g., CryI: Lepidoptera, CryII: Lepidoptera and Diptera, CryIII: Coleoptera, CryIV: Diptera ) No binding sites for Cry proteins exist in the intestines of mammalian species, therefore, livestock and humans are not susceptible. Bt had long been used as a pesticide before it was even considered for use in GM crops. Commercially produced first in 1927, Bt based pesticides were popularized and rel eased for large scale sale by Sandoz Corp. as Thuricide (Feitelson 2001). Bt was re gistered for use in the U.S. in 1961. With the discovery of genetic engineering, it became possible to develop insect resistant crops by transferring the gene that produces the Bt t oxin. First, a Bt strain that attacks the desired 5 Some material is drawn from Krattiger (1997)


23 insect is identified and the implicated cry gene6 isolated. Expression of the desired protein is usually low in these genes. P-ge nes are usually included as part of the transformation package in order to obtain desired expressi on levels. P-genes act like switche s that turn on the production of the desired protein. M-genes, used to identify successfully modifi ed cells, are also part of the transformation of plants at the tissue culture phase. These M-genes can confer modified cells with resistance to antib iotics (e.g., kanamycin) or herbicides or express certain chemicals for visual identification. For example, the plant cel ls in tissue culture m odified with kanamycin resistance can be treated with an tibiotics resulting in death of n on-modified cells and survival of only transformed ones. As mentioned earlier the gene transformation can be done via Agrobacterium tumefaciens or with biolistic methods su ch as gene guns. Finally, once successfully transformed cells are identified, they are grown into full plan ts for seed production and further testing. Major Bt genes include cryIA(b) ,cryIA(c) and cryIII(a) Relevance, Potential Benefits, and Potential Risks This sec tion intends to illuminate some of th e key issues in the benefits versus risks debate surrounding GM crops. It does not attempt to resolve such debate, but rather to inform the reader about the reasons why GM crops and the ensuing debate are rele vant to poverty reduction, hunger reduction, environmental conservation, huma n health, and different human religious and ethical beliefs. The subsections on potential benefits and risks presented in this section are also overviews of the issues, as such, they do not pr esent a comprehensive evaluation of the benefits and risks of GM crops which is beyond the scope of this chapter and c ould only be adequately pursued in a case by case basis (per trait, per region of the world). We start by discussing the 6 Cry is used to denote the protein, while cry is used to denote the gene. Cry stands for crystalline reflecting the appearance of the -endotoxins.


24 relevance of GM crop technologies to hunger reduction, poverty reduction, and environmental conservation. The magnitude of current and future human impact on the environment depends on three main factors: (i) population, economic activity, a nd technological state of the art. In the following we argue that sustaine d population growth and increasing human economic activity are likely to increase human impact on the environmen t. The most viable avenue for reducing such impact comes from the development and a doption of highly sustainable technologies. The human impact on the environment can be summarized in the I-PAT equation: PxAxTI In this equation I represents the magnit ude of human impact, P represents total population, A represents income (or output) per person, and T represents the technological state-of-the-art measured by the environmental impact per dollar of income. A high T implies the current state-of-the-art is not environmentally friendly. Highly su stainable technologies may be defined as those showing high values of S, where TS /1 The I-PAT equation clearly shows that, given our preceding discussion about the pr esence of pent up economic growth in the world, there are two viable strategies for averting further damage to the environment: (i) population controls and (ii) the development and diffu sion of new technologies that are high-S. The relevance of GM crops, whether we consid er ourselves advocates or skeptics, becomes evident when we identify that one of the most important challenges for the coming decades will be to find ways to achieve a sustainable increas e in food production in order to alleviate world hunger and eliminate extreme poverty (United Nations 2008), while at the same time recognizing: (i) that in the most likely scenario human population will continue to grow until it reaches a supposed peak of 9.2 billion in 2050; (ii) that most of this increase in population will happen in developing countries which are alr eady the most affected by hunger, malnutrition,


25 poverty, and disease; and (iii) that given the state-of-the-art, th e current scale of economic and agricultural activity is already causing important destruction and changes in the environment that threaten our ability to achieve an increas e in food production th at is sustainable. The global human population in 2007 is esti mated at 6.6 billion (World Bank 2007). Under the assumption that fertility rates continue to decline, the United Nations (UN) Population Division estimates world population will reach 9.2 billion in 2050 (United Nations 2007). The UN Population Division puts forward four variants of the TFR evolution and the corresponding population forecasts (See Figure 2-1) Population forecasts developed by the UN Population Division depend mostly on the evolution of total fertility rates (TFR) the average number of children per woman during her reproductive years. A TFR=2 means the population stabilizes and stops growing.7 The medium-decreasing variant of TFR is considered the most likely.8 All four TFR variants published by the UN Population Division, except the constant variant, assume that the TFR converges toward the replacement rate as we move towards 2050.9 7 A TFR of 2 is called the replacement rate. At a TFR of 2 each woman has two children, most likely one boy and one girl. The girl will grow up to give birth to a boy and a girl (in average) and thus replace herself and her brother in the population. 8 In addition, a slowly-decreasing fertility variant and a rapi dly-decreasing fertility variant are published. The fourth and final variant assumes the TFR remains unchanged (i.e., business as usual). In the medium variant, world population reaches its peak of 9.2 billion in 2050. It is cons idered a peak since in this scenario all nations reach a TFR 2 by midcentury. 9 The current TFR rate for the world as a whole is estimated at around 2.58 (CIA 2008). If the TFR remains constant instead of converging to the replacemen t rate, we would reach a population of 11.8 billion by 2050. On the other hand, even if a miracle event were to instantly drop world TFR down to the replacement rate, we would still observe an increase in population of about 1 billion, reaching 7.5 billion in 2050. While it may seem contradictory, an increase in population even after reaching the replacement rate is warranted by what is called population momentum. Population momentum happens because of the pyramidal (triangular) age structure of populations that have maintained a high TFR pr ior to reaching the replacement rate. In these nations the ma jority of the population is of young age with only a small percentage of elderly peopl e. Even if the young grow to only replace themselves in the population (TFR=2) this would increase population because such replacement would mean an increase compared to the small number of elderly people leaving the popula tion (being deceased). For a detailed explanation see Sachs (2008).


26 Besides driving growth rates, the TFR also influences the age stru cture in the population. Nations with high TFR (i.e., TFR>2) will replac e each aging individual with more than one young individual thus gradually concentrating the age structure at young age categories. On the other hand, nations that maintain the TF R at the replacement rate or lower (TFR2) will tend to have aging populations with higher concentrations of older people. A generally recognized and observed pattern is that most developed nations have TFR 2 while many developing nations have TFR>3 (See Figure 2-2). Thus aging populations and low TFRs in developed nations mean those populations are stable and w ont contribute to the expected increase in population. This means mo st of the 2.6 billion in crease in population will happen in developing nations which are also the most affected by poverty, disease, malnutrition and hunger. The World Bank estimates show about 1.4 bill ion people living below the international poverty line of $1.25 a day in 2005 (Chen and Ravalion 2008). Even though the globes population is steadily urbanizing, the largest shar e of poor still live in ru ral areas and depend on agriculture. According to new evidence 75 percent of the poor, living on less than $1 per day in developing nations, still re side in the countryside (Ravallion, Chen, and Sangraula 2007). Their livelihoods depend largely on th e natural provision of res ources by the environment. Experts argue that the current scale of human economic activity is already causing environmental destruction at unprecedented le vels. Deforestation and other human-related impacts now threaten and affect almost every sing le ecosystem in the globe. Ocean fisheries are being depleted of fish. Water is becoming more and more scarce: for both drinking and irrigation. Climate change, among other important impacts, is causing huge changes in weather and agricultural production: reducing rainfall levels in many places and increasing rainfall


27 variability in general. The le vel of human economic activity is only expected to grow. A simple exercise making the reasonable assumption that a ll regions of the globe join the convergence club10 in the coming years is developed in Sachs (2008), it is shown th ere that global Gross Domestic Product (GDP) is expected to rise by a factor of six times by 2050. More than a century ago, in his famous 1798 publication An Essay on the Principle of Population, Thomas Malthus warned us about th e imminent danger of human population growing faster than food produc tion. He argued that human p opulation growth, depending only on humans will to reproduce, has no bounds and ha ppens at a geometric rate. On the other hand, food production growth is bounded by the stock of natural resources and by the technological state-of-the-art, and happens only at something closer to an arithmetic progression. In his time, Malthus could not have foreseen the improvements man would achieve in the efficiency of food production due to incredible advances in agricultura l production technologies. About two centuries after Malthus presented hi s catastrophic hypothesis, two German chemists named Fritz Haber and Carl Bosch, discovered what is known as the Haber-Bosch process. This process is used to fixate nitrogen from the air for industrial scale production of nitrogen fertilizer (N-fertilizer) esse ntial for plant growth. Bo th scientists were awarded the Nobel Prize (1918 and 1931) for their breakthrough. In 1970, Norman Borlaug, known as the father of the Green Revolution, received the Nobel Peace Prize for his work in the devel opment and diffusion of HYVs of dwarf wheat in India and Pakistan. The extensive development of HYVs and the earlier discovery of the HaberBosch process formed the technological basi s for the Green Revolution a large scale 10 Convergence describes the process by which the poorer co untries (regions) catch up with the richer countries. Convergence occurs when per capita income of a poorer country grows more rapidly in percentage terms than per capita income in the richer regions.


28 technological diffusion phenomena that sprouted dramatic increases in food production and averted widespread famine in Asia during the 1960s. Malthus also could not have foreseen the currently slow but still apparent convergence (among high-income nations) in fertility rates to wards the replacement rate. As mentioned, at the replacement rate; zero popul ation growth is achieved. Skeptics of the Malthus hypothesis use these two factors (advances in technology and convergence in fertility rates) to argue that no such catastrophe is imminent. However, even if we agree with the relevance of these two factors, this is not to say that Malthus was wrong. It simply means that the basic Malthusian thesis, as such, applies to a world with a fixed level of technology and unchecked population gr owth rates. Therefore, the Malthusian catastrophe is in fact imminent unless we keep on with advances in technology and popula tion controls that are necessary to avoid it. Clark (2007) has recently formalized Malthus ideas in a mathematical model showing technological advanc es can, at best, support larger quantities of individuals but cannot improve the long run level of material wellbeing. This is because, in a dynamic world, larger quantities of better quality food and materi al living standards will give rise to healthier individuals and in particular he althier female populations which s hould increase fertility rates. The result is a decrease in death rates and an increase in population that drops material wellbeing back to its fixed long run level. If fertility rates run am uck, showing a large and sustained increase, the short run welfare may drop drastical ly and very high death rates may ensue. Evidence from recent human history seems to support Clarks argument that, given a fixed and exogenous birth rate, technolog ical advances may provide a s hort run increase in material well being, but in the long run it will only increase the numbe r of people we can support at


29 subsistence material wellbeing levels.11 As Borlaug (1970) puts it in his Nobel Prize Lecture, the technological advances diffused during the Green Revolution, have won us a temporary success in mans war against hunger [] it has given man breathing space [] sufficient for [] three decades [if fully implemented]. But [] hu man reproduction must also be curbed. More than three decades have passed since only a partially implemented12 Green Revolution took place. Despite major advances in agriculture and growth in food production, the twentieth century saw famine after famine kill millions of people across the world. It is evident that sizes of human populations in several places have reached a point where they have outstripped the local environment carrying cap acity, at least with currently deployed technologies (Sachs,2008). A growing consensus exists in the global de velopment and scientific community that the current state-of-the-art based on conventional br eeding technologies will not suffice to achieve the increase in food production needed to meet future demand in 2050. Many believe biotechnology and GM crops to be essential in augmenting our chances of reaching this objective and at the same time minimizing further stress to the environment.13 Development and deployment of biotechnology crops is becoming more and more relevant. Among those that support this view is Norman Borlaug, who calls for the need of a second Green Revolution and endorses the use of biotechnology to develop the new miracle s eeds that will be necessary (Monsanto 2005; Schattenberg 2009). If appropriately diffused, biotech crop s have the potential 11 Technology advancements may be able to increase materi al wellbeing in the long run if the fixed birth rate assumption is relaxed and we assume that birth rates may be brought down. 12 Many rural places across the globe, wh ere very rustic agricultural practices predominate, remain unreached by the technologies of the Green Revolution. See Duflo, Kremer, and Robinson. (2006) for an interesting view of why some of these communities have failed to adopt agricultural technologies. 13 The international scientific and development community now recognizes that doubling or tripling of world food, feed and fiber production by the year 2050 to meet the needs of an 11 billion global population cannot be achieved without biotechnology. (James and Krattiger 1996).


30 of improving food security acro ss the globe and minimizing impact on the environment through rapid improvements in local and global crop pro ductivities per unit of input (i.e., land, water, fertilizer, pesticides, la bor or other inputs). The following subsection presents an overview of the most cited benefits attributed to GM crops. After this we finish this section by presenting an overview of the most cited potential risks. Potential Benefits of Biotech Crops Scientists and researchers have the ability to genetically m odify agronomic traits and quality traits. Agronomic traits are traits that determine crops suitability to its specific environment. Scientists modify these traits to produce the following be nefits: (i) improvements in productivity and easiness14 of food production, (ii) lower produc tion costs, and (iii) achieving greater consistency of production (i.e., reducing uncertainty). Quality traits refer to those conferring specific physical or chemical desirable characteristics to the commercialized portion of the crop. Scientists modify quality traits to obtain benefits like improving the qualit y and nutritional contents of foods. With respect to agronomic traits, some of the different strategies being investigated to increase crop output per unit of input are: (i) attacking the caus es of crop losses, for example, pests (e.g., insects, virus, disease) or competito rs for soil nutrients (e .g., weeds); (ii) improving plants own efficiency in using inputs (e.g., Nfertilizer saving); and (iii) improving plants ability to grow under harsh condi tions (e.g., salinity, drought, fros t resistance) (P ew Initiative on Food and Biotechnology 2001). Examples of biotech agronomic traits are: 14 Ease of use may be particularly important for farmers th at have limited access to edu cation and extension services.


31 Resistance to pests and disease (e.g., Bt-corn) (marketed) Tolerance to chemical herbicides (e.g., Roundup Ready varieties) (marketed) Tolerance to drought (in last stages of research and development (R&D) pipeline) Improved absorption of soil nutrients (e .g., N-fertilizer saving) (R&D pipeline) Tolerance to adverse soil and other physical conditions (e.g., salinity, frost) The second type of beneficial traits co rresponds to those that improve quality and nutritional content. During its earl y years, most genetic modifications of plants were aimed at increasing or protecting crop yields. These early modifications, the so called first generation GM crops, usually involved parts of the plant not cons umed directly by humans (i.e., cornstalks). The beginning of the 21st century saw a new wave (second generation) of potential modifications involving changes in the composition of foods to enhance their quality an d nutritional value. For example, Golden Rice was created by transforming rice with two beta-carotene biosynthesizing genes: psy (from daff odil) and crt1 (from soil bacterium Erwinia uredovora) (Ye et al. 2000). Beta-carotene, a pr ecursor of Vitamin A, is produ ced and made available by the plant in its endosperm (the edib le part of rice). Vitamin A deficiency (VAD) is an important nutritional problem in the world (World Bank 1993). VAD is frequent among the poor in Asia whose diet is based mainly on rice whic h does not contain Vitamin A precursors. Supplementation of carotenoids in deficient populations has shown to reduce morbidity and mortality in children (Sommer 1997). Dawe, Robe rtson, and Unnevehr (2002) compare the costs of Vitamin A supplementation via Golden Rice with supplementation by other methods such as health programs. The evidence shows that modifying the nutritional content of staple foods of the poor, like rice in this case, holds huge potential benefits in terms of reducing the costs of nutritional supplementation. Other examples of traits that improve qua lity and nutritional contents of food are: Foods that exhibit impr oved processing traits Foods that have improved ripening, texture, or flavor


32 Foods that show improved nutritional contents besides carotenoids Foods that hold lower concentrations of allergens, toxins, or antinutrients In addition to the benefits presented above, so me important environmental benefits may be realized by rapid and widespread diffusion of modified crops. For example, higher yields on current agricultural fields may prevent further deforestation. Better absorption of N-fertilizer by crops may reduce losses to envi ronment (by leaching, ammonia vol atilization, denitrification) that have negative effects such as eutrophication and climate ch ange. Drought tolerance traits could reduce irrigation needs in ar eas with limited water supplies. Potential Risks of Biotech Crops Opponents of biotechnology do not discard the cl aim ed benefits of GM crops, but argue that several potential risks asso ciated with these crops might outweigh the potential benefits. Opponents argue that not enough is known about the environmental safety and health risks associated with these foods and that they should be more rigorously controlled (U.S. General Accounting Office 2002). Other opponents have ethical concerns a bout manipulating the genetic material of living beings. A last variety of opponents, believes in the benefits of GM crops, but holds some doubts about how these should be produ ced, by whom (private or public), and what the net effects will be for poor developing nations. Some potential risks to the environment include: Risk of GM crops becoming weeds or invasive species Impact on non-target organisms (i.e., Bt effect on monarch butterflies) Birth of a super weed by accidental cross-polli nation with wild relatives (gene flow) Exposure of insect pests to insect resistan t crops may induce insect resistance to the pesticides. In addition, many argue that while increases in yield due to biotech traits may reduce the pressure on land, traits that expa nd the range of environments in which crops can be grown may affect previously unthreatened ecosystems. As we can see, bo th the potential benefits and


33 potential risks to the environmen t are plausible, which is why the debate goes on. An important counterargument made by supporters of GM crops, is that these environmental risks, to some extent, are not exclusive to biotech engineering. Potential risks to human health from consumi ng GM crops also enter the debate. GM crops have been consumed widely for many years with no conclusive confirmation of serious harmful effects on human health (Pew Initiative on Food and Biotechnology 2004). Some of the less serious potential risks de bated, such as risk of toxicity and allergenicity, are co ntinuously being controlled by government regulation of the seed development process. These risks, again, are not exclusive of GM crops, and may happen in crops developed by conventional breeding (i.e., glycoalkaloids in potatoes). On e important setback, however, in the early evolution of the industry due to a mistake in the regulation pr ocess was the negative ev ent associated with StarLinkTM corn. In May 1998, the U.S. Environmental Protection Agency (EPA) granted a limited license to Aventis Crop Scie nce for the production of StarLinkTM corn. StarLinkTM corn was engineered by isolation and incorporation of the gene responsible for synthesizing the Cry9c protein, occurring naturally in Bacillus thuringiensis subspecies tolworthi bacteria, and responsible for expressed crop resi stance to several insect pests (i ncluding European corn borer (ECB), cornstalk borer, and corn earworm). Th e granted limited license allowed for production of animal feed, industrial nonfood uses, and seed increase; but proscribed its use in food intended for human consumption because the Cry9c protei n shared several molecular properties with proteins that are known food allergens (i.e., stab ility to heat, acid, and enzyme degradation). Nevertheless, in September 2000, Cry9c-DNA was detect ed in taco shells, proof that the variety had either intentionally or ina dvertently been introduced to the food supply chain. The ensuing media coverage resulted in larg e recalls of implicated processed products. This was followed by


34 reports to the Food and Drug Admi nistration (FDA) of adverse health effects from consumers who had eaten potentially contaminated corn pr oducts. In addition, the U.S. Department of Agriculture (USDA) reported that mixing of all corn after harvest, including StarLinkTM corn, was a common practice in the industry suggesting th e leak had happened unintentionally. In June 2001, the Center for Disease Cont rol (CDC) presented a report (Center for Disease Control 2001) to the FDA which had analyzed serum collected from 28 individuals reporting valid allergic reactions apparently associated with th e Cry9c protein. Additional serums were collected from individuals identified as being highly sensit ive to a variety of a llergens. The individual serums where tested for presence of antibodie s to the Cry9c protein using an FDA-developed Enzyme Linked Immunosorbent Assay (ELISA) t ype method. Based on negative results from the serums to the ELISA tests, the study concluded evidence was insufficient to link the allergic reactions to the Cry9c protein. Regardless, the whole event had large and lasting repercussions on the consumers perceptions about GM foods and on GM markets worldwide. Opponents argue that, even though no harmful eff ect has been confirmed as of yet, there exists the risk of unintended h ealth effects in the long run. Th is claim may be impossible to prove a priori (U.S. Gene ral Accounting Office 2002). A third point of controversy involves ethical beliefs of di fferent social and religious groups. The term ethically sensitiv e genes has been used to re fer to genetic transformations that may raise ethical concerns from specific human groups (Aldridge 1 994). For example, the transfer of human genes to animals or crops used for feed or food, the transfer of genes from animals whose consumption is forbidden to certai n religious groups, and the transfer of animal genes into crops which may raise concerns am ong vegetarians. The debate here focuses on


35 whether genes, when taken out of the cell, rema in part of the organism or may be simply considered as chemical molecu les made of nucleic acids. So far we have discussed potential risks to the environment and to the consumer. The last group of opponents focuses on market and economic issues of GM crop introduction. A main concern is whether GM crops will become an effective tool to fight world hunger and benefit farmers in developing nations. This depends, partially, on whether GM crops are tailored to meet the needs of small farmers (i.e., drought toleranc e) in developing countries. It also depends on how these technologies are diffused, distributed, and marketed. The debate focuses on whether public or private (or public-private partnerships) should take on the task. As far as the private sector goes, some are doubtful that the market demand from poor-small farmers in developing nations is insufficiently large to create the ne cessary incentives to invest in research and development. In addition, deficient intellectual property right laws and inadequate regulatory capacity in these nations increase market (and en vironmental) risks and ma y deter market supply. A final point of concern among governments in developing countries, is whether granting of property rights to multinational seed compan ies and guaranteeing enforcement will increase incentives but also result in GM crops been sold at prices unaffordable to small farmers. In the light of these issues, it appears the development of GM crops ta ilored to meet the needs of small farmers in developing nations, and their effec tive supply, will probably require participation from universities, governments and international research centers. Public-private partnerships may have a better chance of success in developing and supplying such GM crops at prices that are affordable to farmers and profitable to private enterprises. Finally, some other market related issues have been raised on the introduction of GM crops that are not exclusive to small farmers. For example, the StarLinkTM corn event resulted in


36 segregation costs that reduced th e revenue that U.S. corn producers would have received in 2000/2001 in the absence of the event (Schmitz, Schmitz, and Moss 2005). Segregation costs were the costs incurred in sepa rating GM from nonGM corn thr oughout the supply chain in order to meet Japans stringent import tolerance levels following the StarLinkTM event. The potential risks and potential benefits pr esented here are not an exhaustive list, but hopefully they have given the reader a broad view of the major points around which the debate is centered. We have not attempted to give a conclu ding assessment of the net effect GM crops will have in the long run, however, it is evident that the technology is relevant for many reasons. Brief History of Early Commercialization of GM Crops Successful comm ercial producti on is the final phase in the biotech crop development process. The process begins with a first phase in which scientists in government or private laboratories and greenhouses investigate potential genetic tra its to be transferred. Once a promising trait has been identified and successfully transf erred into the target crop, the next step is testing it under real life conditio ns in field trials in a second phase. The third phase involves securing regulatory approval for commercializati on for feed or food. The fourth and final phase is widespread commercialization and market acceptance. Most steps in the process involve some degree of regulation, for example, the second phase requires regulatory approval for environmental release.15 Why regulate? The justification is largely based on unfamiliarity (Dale 1995; OECD 1993; James and Krattiger 1996). Different regulatory schemes predominate in different regions (See Appendix A for a more complete description of GM cr op regulatory schemes). 15 For a more detailed description of the development process see http://www.monsanto.com/pdf/pipeline/pipeline_2009_phase.pdfu.


37 The first GM food to be commercially produc ed and marketed in the U.S. was FLAVR SAVRTM Tomato from Calgene Inc., approved by the FDA in May 1994.16 The tomato had been genetically engineered to remain firm for a longe r period of time after harvest compared to other tomatoes. Regular tomatoes are normally harves ted while still green in order to avoid crushing during transportation through the ma rketing chain. They are later ri pened artificially in ethylene gas chambers. The formation of Polygalacturona se (PG), an enzyme naturally occurring in ripening tomatoes, breaks down pectin in cell walls and causes ripe tomatoes to soften. Synthesis of PG is suppressed in FLAVR SAVRTM tomatoes allowing the farmer to leave the fruit on the vine longer than its conventiona l (nonGM) counterpart. This part icular attribute allows the tomato to ripen and reach full flavor before harv est, eases transportation, and results in tomatoes that remain firm for longer periods afte r reaching the final market (Food and Drug Administration 1994a). FLAVR SAVRTM was approved by FDA to be marketed without any special labeling (Food and Drug Administration 1994b). Consumer acceptance was positive (James and Krattiger 1996); however, competition from conventionally bred longer shelf-life tomatoes created profitability pr oblems and prevented FLAVR SAVRTM from successfully penetrating the market (Martineau 2001). The production, approval, and marketing of one single GM food (FLAVR SAVRTM) in 1994 was just the tip of the iceberg when it came to investments and R&D on GM food technologies that had begun in 1971 when the fi rst GM organism was developed (James and Krattiger 1996).17 During the period 1986-1996, more than 3,500 field trials were conducted on 16 The first country to commercialize GM crops worldwide was China in the early 1990s (v irus resistant tobacco and virus resistant tomato). 17 According to Monsanto (2009), the first plant cell modification was achieved in 1982 by scientists working at Monsanto. However, a website published by the Department of Soil Sciences at Colorado State University (2006) identifies four groups that, working independently, simultane ously achieved the first modification of a plant cell: (i) a Washington University Group (antibiotic kanamycin resistance into Nicotiana plumbaginifolia); a Belgium group


38 more than 15,000 sites, in 34 countries with at least 56 crops (James and Krattiger 1996). Commercialization of more GM crops would soon follow. Several more GM foods that had been waiting in the R&D pipeli ne followed FLAVR SAVRTM into the market shortly after. By yearend 1995, 20 petitions had been granted to commerc ially grow 9 transgenic crops only in the U.S. (James and Krattiger 1996). By 1996, the major crops approved in the U.S. included: tomato with delayed ripening qualities, cotton (herbicide tolera nt), cotton (insect resistant), soybean (herbicide tolerance), maize (herbicide tolerant), maize (insect resist ant), canola (modified oil quality), potato (insect resistant), and squash (virus resistant) (James and Krattiger 1996). Table 2-1 shows a list of GM crops approved for sale in the U.S. in the initi al years of the GM crop industry. In all countries (except China and Australia) all approvals up until 1996 were granted to the private sector (James and Krattiger 1996). Current Status of GM Crops The International Service fo r the Acquisition of Agri-Biotech Applications (ISAAA) publishes year to year briefs concerning update s in the global status of commercialized GM crops. The last report corres ponds to the year 2008 (Jam es 2006, 2008). This section draws heavily from statistics and key issues presented in these reports. Many of the brand names introduced in the firs t years of GM commercialization remain as star products today (e.g., RoundupReadyTM, YieldGardTM, LibertyLinkTM), although they have been improved in several ways. After an initial period in which seve ral new traits were introduced, the advances focused not in more ne w traits, but in produci ng stacked-trait hybrids by cross-breeding those initially marketed star products. Monsanto, for example, produced a (antibiotic kanamycin resistance into tobacco plants); a Monsanto group (antibiotic kanamycin resistance into petunia plants); and a Wisconsin group (bean gene into sunflower plant).


39 stacked hybrid variety of corn that was resi stant to the ECB and tolerant to glyphosate by hybridizing two GM moth er lines: NK603 (RoundupReadyTM) and MON810 (YieldGardTM). Since the GM mother lines in these stacked trait hybrids had alre ady been approved for commercialization, no furthe r approval was needed. More recently, seed companies have started producing stacked varieties that are not obtained via hybridization but by simultaneous m ultiple genetic modification. Monsanto has developed its trademark Agrobacterium -mediated process for multiple trait insertion called VecTranTM. The advantages of using direct multiple-trait transformation are: (i) the process is sped up since no cross breeding is necessary, and (ii) better control ov er the transformation process, for example, guaranteed better promoter genes. These two advant ages result in stacked varieties that can be produced in a more time-efficient manner and that produce more consistent results. As far as companies go, acquisitions and divest ures have been more of a rule than an exception in the industry. For example, Syngent a was created by the merger between Novartis and AstaZeneca in 2001. Novartis itself was the re sult of a merger between Ciba and Sandoz in 1996. Monsanto started buying interests in Ca lgene in 1996 and finalized acquisition by 1997. Global adoption of GM crops has occurred at a rapid pace. As of 2008, the number of countries planting biotech crops has soared to 25. Cumulatively, the total number of acres planted with biotech varieties in all years sin ce the first biotech crop was commercialized in 1996, reached the billion-mark in 2005. In 2008, only three years later, the 2 billion-mark was reached. In hectare terms, 2008 s howed a total area of 125 million hectares dedicated to GM crops; this meant a strong growth of 10.7 millio n hectares over the 114.3 million hectares observed in 2007. Figure 2-3 shows the areas plan ted (in millions of hectares) with GM crop


40 varieties by the top eight GM pr oducers in the globe in 2006. Six out of the major GM players (Argentina, Brazil, India, China, Paraguay, and South Africa) are listed by the IMF as emerging or developing economies. Four major crops have reached the market to date: soybean, cotton, canola, and maize. Figure 2-4 presents the global planted area (GM+ nonGM) and the total GM planted area in 2006 for each of these four GM crops. For soybean, a bout 91 million hectares were planted globally in 2006, of which 58 million (or 64%) were GM varieties18. Cotton was planted in an area approximating 35 million hectares, of which 13 million (or 38%) were GM varieties. Canola, observed a global planted area of 27 million hectares, of which 5 million (or 18%) where GM. Finally, corn was planted at a global scale in some 148 million hectares, of which 25 million (17%) came from GM seeds. These numbers ar e considerably large considering 2006 marked only the first decade of GM commercialization. The frequency with which they are mentioned in GM forums may lead some to believe that these four crops are practically th e only GM crops out there. However, this is not the case. Other GM crops, such as papaya, tobacco, and squash have also been approved for commercialization in the U.S. Many more GM crops have been co mmercialized with varying market acceptance: tomatoes, rice, potatoes, melon, and peppers. If we consider crops that have not yet reached commercialization phase, but that have been subject to field and labor atory trials, the list expands. In a report prepared for the Council of Biotechnology Information, Runge and Ryan (2004) surveyed all biotech crop research and tr ials being conducted wo rldwide identifying fiftyseven plants under investigation: 16 field crops, 14 vegetables, 16 fruits, and 11 miscellaneous. 18 In Argentina, 99% of the soybean produced comes from GM seeds.


41 Since the beginning of GM crop commercializat ion in 2996, the most widely adopted trait has been herbicide tolerance. James (2008) estimate s that herbicide tolerant varieties account for 63% of the global total of 125 million hectares planted with GM crops in 2008. In 2008, a new biotech crop, RRsugar beet was commercialized in U.S. and Canada. The success of this crops launch poses a good omen for sugar cane biotech traits which are at advanced stages of development in several countries. Fertilizer Saving and Drought Tolerance Traits Two new prom ising traits are in the R&D pipeline of major seed companies, these are: (i) a DT and (ii) FS trait. Monsant o, for example, has a first generation DT corn seed in Phase IV.19 They also have a N-efficiency corn trait (i.e., N-fertilizer saving trait) in Phase II.20 The relevance and potential benef its of the DT trait are obvious. As they say Water is life. This simple biological fact by itself explains the importa nce of water. However, in a complex global society, water means many other thi ngs. Deficient quantities and quality of water supplies may be linked to livelihood insecurity, health risks, hunger a nd poverty, and social conflict. Natures water supply is unpredictable and is becoming more so due to climate change. In many places, climate-related water events (e.g., floods, droughts) are becoming more frequent and more severe (United Nations 2009). Supplementation of locally defici ent natural supplies via water projects is, in many cases, lim ited by financial and social factors. Water is an important input in many industries, but more so in agriculture. Every acre of corn, even on irrigated fields, suffers some degree of water stress at some point during the 19 Phase IV is the last stage prior to market launch. The first three stages are Phase I: Discovery and Proof of Concept, Phase II: Early Development and Compilation of Pre-Regulatory Data, Phase: 3 Advanced Development, Field Testing, and Data Generation. As mentioned before these two traits are the fo cus of this dissertation. 20 See http://www.genuity.com/Innovation/Expl ore-The-Pipeline/Print-The-Pipeline.aspx.


42 growing season. Agriculture accounts for the larg est share of human water use about 70% of freshwater withdrawals are des tined to irrigated ag riculture (United Nations 2009). The demand for water has increased substantially mainly because of population growth, but an important factor has also been growing incomes and the ensuing changes in dietary habits. Grains and cereals are usually seen by consumers as inferior goods compared to beef and other meats. As global incomes grow, the demand for beef follo ws closely. The UNs World Water Assessment Programme (2009) estimates that meat productio n requires 8-10 times more water than cereal production. The case for the N-fertilizer saving trait can be made too. Nitrogen is one of the most intensively-used inputs in crop production es pecially in U.S. corn production. Among the major field crops produced in the U.S., corn us es the most fertilizer (Huang, McBride and Vasavada 2009). While natural provision of nutrien ts is less variable and much more predictable than the natural provision of water, the price of N-fertilizer is not so. Recent volatility of fertilizer prices has shown the potentially larg e negative impacts on crop profitability these fluctuations may have (Huang, McBride and Vasa vada 2009). Price volatility in fertilizer is closely tied to fuel price volatility. This is be cause natural gas is used in producing ammonia the main ingredient in many N-fertilizers. That futu re fuel prices will stab ilize is a scenario hard to picture. In summary, the need to feed growing populat ions in a sustainable manner; the upcoming increase in demand for water coupl ed with the increased variability in natural s upplies; and the intensive use of fertilizer in crop production couple d with an increased variability in fertilizer prices; suggest that widespread adoption of the DT and the N-fertilizer saving traits holds potentially large benefits for farmers, consumers, governments, and society in general. These two


43 traits, and the factors a ffecting (increasing or decreasing) th eir adoption potential at the farm level, are the focus of this dissertation.


44 Table 2-1. Summary of genetical ly modified (GM) crops approve d for commercial growing in the U.S. in 1996 Product Company Altered trait Approved for sale Commercial name Tomato Calgene Delayed ripening 1994 Flavr SavrTM Cotton Monsanto Resistance to bollworms & budworm (Bt toxin) 1995 BollgardTM Soybean Monsanto Resistance to herbicide glyphosate 1995 Roundup ReadyTM Maize Ciba-Geigy Resistance to corn borer (Bt toxin) 1995 MaximizerTM Cotton Monsanto Resistance to herbicide glyphosate 1996 Roundup ReadyTM Canola Calgene Altered oil composition (lauric acid) 1995 LauricalTM Cotton Calgene Resistance to herbicide bromoxynil 1995 BXN CottonTM Potato Monsanto Resistance to Colorado potato beetle 1995 New LeafTM Squash Asgrow Resistance to viruses 1995 Freedom IITM Tomato DNA Plant Technology Delayed ripening 1995 Endeless SummerTM Tomato Monsanto Delayed ripening 1995 Tomato Zeneca/Peto Seed Thicker skin, altered pectin 1995 Maize DeKalb Resistance to glufosinate 1996 Maize AgrEvo Resistance to glufosinate 1996 Liberty LinkTM Maize Plant Genetic Systems Male sterility 1996 Maize Monsanto Resistance to corn borer (Bt toxin) 1996 YieldGardTM Maize Northup King Resistance to corn borer (Bt toxin) 1996 Cotton Dupont Resistance to herbicide sulfonylurea 1996 Tomato Agritope Altered ripening 1996 Potato Monsanto Insect Resistance 1996 Source: James and Krattiger (1996).


45 Figure 2-1. Estimated and proj ected population of the world by projection variant, 1950-2050. (Source: United Nations 2007). Figure 2-2. Total fertility rates (TFR) in 2008 (estimates). (Dat a Source: CIA World Fact Book 2008).


46 Figure 2-3. Participation in total genetically modified (GM) crop planted area, by country, 2006. (Data Source: Argenibio 2006; James 2006). Figure 2-4. Global GM crop plante d areas, for four major GM crops, as a percentage of their global planted area. (Data Sour ce: Argenbio 2006; James 2006).


47 CHAPTER 3 THEORY AND KEY CONCEPTS The m ain focus of this dissertation is to obt ain estimates of farmers willingness-to-pay for, and to study the adoption potential of, two ge netically modified (GM) corn seed traits: a fertilizer saving (FS) trait and a drought toleranc e (DT) trait. A nonmarket valuation approach is used due to the fact that these traits have not yet reached the market. Th is chapter presents an overview of the economic theories relevant to th e nonmarket valuation pu rsued in this study. The chapter is divided in the following two sections. The first section begins by presenting th e concepts of Equivalent Variation and Compensating Variation which form part of the theory of welfare measures of price change pioneered by Dupuit (1844) in the nineteenth century and developed by contributions of Marshall (1930), Hicks (1941, 1943, 1956), Willig (1976), and others1. The adaptation of these exact welfare measures to changes in quantity sp ace, owed mainly to seminal contributions by Mler (1974), Randall and Stoll (1 980), Hanemann (1991), and others; is also presented in this section. This adaptation to quantity space forms the theoretical basis for methods used in nonmarket valuation theory. The section finishes by presenting the formal definition of the key economic concepts of willingnessto-pay and willingness-to-accept. The second section in this chap ter presents an overview of the theory and key issues underlying the Contingent Valuation method. The Contingent Valuation method is recognized in the U.S. Federal Register as one of three methods recomm ended for measuring value in nonmarket situations. The core of the section begins by descri bing the Contingent Valuation method and goes on to describe (i) the predominant el icitation methods used in the literature (i.e., open ended, iterative bidding, single bounded, and double bounded) (ii) its microeconomic 1 See Just, Hueth, and Schmitz (1982) and Just, Hueth, and Schmitz (2004).


48 foundations, and the (iii) appropria te statistical models to be used with the two predominant elicitation methods (i.e., single bounded and double bounded). The section finishes with (iv) a brief presentation of the Stated Preference vs. Re vealed Preference debate in the Contingent Valuation literature, and (v) a discussion of Krutillas(1967) concept of existence value (or nonuse value). The two sections in this chapter are intende d to present the reader with some basic concepts that are used in the type of nonmarke t valuation pursued in th e remaining chapters of this study. The concepts in this chapter are pres ented at a very abstract level to facilitate exposure. The reader familiar with these basic concepts may want to skip to the next chapters where the concepts presented here are adapted to the specific situati on of technology adoption. Willingness-to-Pay (WTP) and Willingness -to-Accept (WTA) Monetary measures of welfare value for different goods are necessary to conduct appropriate benefit-cost analysis in welfare eco nomics. Traditionally, benefit-cost analysis has played a key role in informing policymakers deci sions with respect to the potential net benefits associated with the price changes (e.g., taxes, subsidies, etc) implied by different policies.2 The two exact measures of welfare impacts due to price changes were proposed by Hicks (1941, 1943, 1956) based on the areas under the Hicksian demand curve3; one which holds indirect utility at the initial level (0u ) as the reference point called the Compensating Variation (C), and another one that holds indire ct utility at the posterior level ( 1u) as the reference point called the Equivalent Variation (E). Form ally, the C and E measures are defined by: 2 Even though measures of welfare change due to price changes are not directly relevant to this study, we present them first because they may result more familiar to the reader, and to emphasize the difference with measures of welfare change in quantity space which are of our direct interest and which are presented later in Equations 3-3 and 3-4. 3 The Hicksian measures are defined as the area under the Hicksian demand curve and above the price line.


49 ),,(),,(0 1ypuCypu q q (3-1) ),,(),,(0 1Eypuypu q q (3-2) where u represents indirect utility, q is the utility maximizing vector of goods, 0p is the initial price, 1p is the price after the change, and y is the income level. In many cases one may be more interested in the effects on welfare of quantity changes as opposed to price changes. This usually is the case of policy makers aiming to evaluate the benefits and costs of proposed projects or programs or the welf are effects of provision vs. nonprovision of a good. These have been popular obj ectives in the nonmarke t valuation theory where no market behavior data on pr ices or quantity is available. Karl-Gran Mler (1974) was maybe the first to show that the concepts of C and E could be adapted from price change effects to measures of welfare impacts due to quantity changes. At this point, it is useful to define C and E for a change in quantity in a formal manner. We restrict our attention to a change in a single commodity4,q which could represent the supply of a public good (or bad), could be an index of quality, or could be a new technology product which is not yet in the market th is final one being our case. Let p and y unchanged, and consider valuing the e ffect on welfare from a change in a situation without to one with the single commodity q. Let us represent this change as a change from 0q to 1q. The individuals utility thus changes from ),,(0 0yqvu p to ),,(1 1yqvu p ; with 01uu if she considers the change an improvement, 01uu if she considers it for the worse, and 01uu if she is indifferent a bout the change. Following 4 Having quantity change in more than one good does not change the results but makes the calculations and discussion much more complicated.


50 Hanemanns (1991) notation, the welfare measur es C and E in quantity space are formally defined as: ),,(),,(0 1yquCyqu p p (3-3) ),,(),,(0 1Eyquyqu p p (3-4) That is, C and E are the monetary values th at make the individual indifferent between a situation where they have the good (1q ) and a situation where they do not have it (0q ). The C measure takes indirect utility at the initial situation where the individual is without the good as the reference point, while the E measure uses the indirect utility in the situation with the good as the reference point. The usual measures of individual value are the willingness-to-pay (WTP) and willingnessto-accept (WTA). In a market exchange situation these corre spond, respectively, to the buyers best offer and the sellers reservation price. The notational convention in Equation 3-3 and Equation 3-4 is used so that )()()(01EsignCsignuusign Therefore, if the change is an improvement: 001 uu, 0 C and 0 E. In this case, C is the individuals maximum CWTP to secure the change, and E is her minimum EWTAto forego it. On the other hand, if the change is for the worse: 001 uu, 0 C and 0 E. In that case, C measures the individuals minimum CWTA in compensation to endure the change, and E is her maximum EWTP to avoid such change. Notice that WTA is not always the C measure a nd WTP is not always the E measure. In the case of quantity changes, methods have been developed to directly estimate WTP and WTA. One of the most popular and most extensively used of these methods is the Contingent Valuation method which will be discussed in the following section.


51 Contingent Valuation (CV) The Contingent Valuation (CV) m ethod is a st ated preference appr oach that has been extensively used to obtain estim ates of WTP in nonmarket situ ations such as the case of environmental goods. Cameron and James (1987b) s uggested that the CV method can be equally useful in pre-testing new market goods. The CV method is recognized in the U.S. Federal Register as one of three methods recommended for measuring economic value in non market situations. The CV method was first recognized in the Federal Register as a valid method for evaluation of project benefits in the Principles and Standards for Water and Relate d Land Resources Planning guidelines published by the U.S. Water Resources Council (1979). It is also named as an approved method in the Comprehensive Environmental Response, Comp ensation, and Liability Act of 1980 (Superfund) (U.S. Department of Interior 1986). The CV method uses especially designed survey s to directly elicit from respondents their individuals preferences for a given commodity by querying them for their WTP (or WTA) to secure (forego) a positive change in the level of provision of that commodity.5 The ultimate aim of a CV survey is typically to obtain an accurate estimate of the benefits (or costs) of a change in the level of provision of some good (Mitchell and Carson 1989). In a CV survey the individual is presented wi th a hypothetical constructed market and is asked to give responses stating her preferences (in terms of do llar amounts he is WTP or WTA) regarding different scenarios in which the quantities of the good are changed.6 The respondents 5 Mitchell and Carson (1989) provide a comprehensive overview of the issues involved in the design and analysis of CV surveys, and Bateman et al. (2002) provide a useful manual for the practitioner. 6 If the change in case is an improvement, CV measures th e WTP to secure the change and/or WTA to forego it. If the change is for the worse, the method measures the WTP to avoid the change and/or WTA to endure it.


52 answers are said to be contingent on the detail s of such hypothetical market as put forth by the survey. In essence, the CV method is seen as havi ng four distinctive a dvantages: (1) the CV method represents a tool to estimate the effects on welfare of changes in the quantity of some commodity when market data on prices and quan tities exchanged is not available (e.g., public goods, new unmarketed products), (2) its hypothetic al nature gives it more flexibility than observed behavior methods allowing for evaluati on of scenarios that havent actually happened but may be very interesting, (3) the CV method obtains the actual Hicksian measures of welfare (WTP and WTA); and (4) the WTP measures obtained by use of the CV method include Krutillas (1967) existence value and Weisbrods (1964) option value. The remainder of this section on CV is or ganized as follows. First we present the predominant elicitation formats used in the lit erature and discuss thei r relative merits and shortcomings. Second, we present the micro economic foundations of CV and develop the appropriate statistical models to be used with the two most relevant e licitation formats (single bounded and double bounded). After this we briefly illustrate the Revealed Preference vs. Stated Preference debate that surrounded the early years of CV research and how such debate has today shifted towards a more holistic view where revealed and stated methods are seen to complement rather than compete against each other. We finish this CV section by discussing Krutillas (1967) concept of existence value framed in a descrip tion of alternative (other than CV) indirect revealed preference methods that ar e used in the literatur e to perform nonmarket valuations. Elicitation Formats, Statistical Ef ficiency, and Starting Point Bias There are several ways in which one can query the respondent about her WTP. Two distinct elicitation formats predominate in the CV literature: (i) the open ended (OE) format and


53 (ii) the dichotomous choice (DC) format (or closed -ended format). In the OE format one asks the respondent:How much are you willing to pay (accept) for a change from 0q to 1q? Suppose the answer given is $Bid, then $Bid is taken as that individuals measure of C (of E). The DC format asks: Would you be willing to pay (accept) $Bid for a change from 0q to 1q? In this case, the respondents answer is dichotomous yes or no. A major stimulus to the development and refine ment of CV methods was the enactment of U.S. laws that allowed for recovery of moneta ry damages for injuries to natural resources (Superfund). The key event was the Exxon Valdez oil spill where the state of Alaska claimed for damages based on estimates obtained using CV methods7. Following the spill, an aggressive campaign was launched by oil companies that ha d several studies published which left CV methods in bad light (See Hausman 1993). At the same time, on the other side of the debate, the U.S. National Oceanic and Atmospheric Admi nistration (NOAA) convene d a Blue Ribbon panel co-chaired by two Nobel Prize laureates (A rrow et al. 1993) to make a comprehensive assessment of CV methods. The now famous NOAA report endorsed CV and in particular endorsed the DC elicitation format. Critics of the OE formulation of CV have referred to it as the Silly Question Method or Pick a number and have characterized it as: Suppose one approach ed people in a shopping mall, made them put their bags down for a mome nt, and asked them what was the most they would be willing to pay for conservation of a sea o tter in Alaska or an expanse of wilderness in Montana. The DC format is generally agreed to give better results because it provides respondents with a more market-like situation. 7 CV method was used by the state of Alaska because it is the only method that measures existence value. Existence value refers to the value people assign to things just from knowing they exists, even if they do not plan to consume them. In contrast, use value is what economists are used to and refers to the value derived from consumption of things.


54 There exist several variations along the lines of DC elicitation: (i) single bounded (SB), (ii) iterative bidding (IB), and (ii) double bounded (DB). The SB format was first implemented by Bi shop and Heberlein (1979) who estimated the WTP for duck hunting permits. In the SB-DC approach, respondents are presented with randomly assigned bids and are asked in a sing le yes/no question whethe r they would pay or refuse to pay the offered bid to secure being provided with the good.8 Several early studies were produced using the SB-DC method, popularized by Hanemann (1984) who demonstrated the economic theory underpinning such method using a random utility framework and an indirect utility approach. The SB approach, since it pr ovides less information about the magnitude of WTP,9 requires a larger number of observations co mpared to the OE format to achieve similar levels of statistical efficiency. The IB method (Davis 1963a; Randa ll, Ives, and Eastman 1974) li es at the other end of the spectrum. The IB method is simply a series of sequenced DC questions. In the bidding method a $Bid offer is made; if the respondent answers is no the offer is se quentially and gradually decreased until a yes is obtained; if the initial response is yes the offer is sequentially and gradually increased until a no is obtained. The value of $Bid at which the respondent switches her answer is used as the dependent variable in regression models that estimate WTP. The IB method, while achieving greater stat istical efficiency compared to SB method, results tiring to the respondent and is known to be greatly affected by what is te rmed starting point bias. Starting point bias is concerne d with the initial offer being used by the respondent as an anchor or focal point. Confronted with a dollar fi gure in a situation where he is uncertain about 8 If instead of WTP we want to elicit WTA, then this changes to whether they would forego the good for the offered bid. From here on our discussion will be about WTP. 9 Open-ended format elicits the actual magnitude of WTP while dichotomous choice obtains only a yes or no bound.


55 an amenitys value, a respondent may regard the proposed amount as conveying an approximate value of the amenitys true value and an chor her WTP amount on the proposed amount (Mitchell and Carson 1989). Therefore, her firs t answer will derive from her prior WTP distribution while subsequent responses in the bi dding process will only be updates to this prior based on information obtained by each offer (H erriges and Shogren 1996). Ideally, we would want each answer to be based on the same distribution, that is, each answer should be independent because in each answer the respond ent should query her preferences and not her prior answers to provide her personal response. Hanemann (1985) and Carson (1985) proposed the DB format which can be thought of as a compromise between the SB and IB methods: (i) improving upon the low statistical efficiency of SB and (ii) ameliorating the effect s of starting point bias of IB. We can think of the SB format being at one end of the spectrum, and the IB method at the other, with DB lying in the middle. In the DB method respondents are as ked one initial question and one single follow-up question. In the initial question an in itial $Bid offer is made, if the re spondents answer is no a second lower offer is made in the follow up question, if the initial answer is yes a second higher offer is made in the follow up question. The bidding process consists only of these two offers. Hanemann (1985) and Carson (1985) first proposed this method to improve efficiency of discrete choice questionnaires. Hanemann, Loomis, and Kanninen (1991) demonstrated the large efficiency gains from using DB instead of SB methods. Cooper and Hane mann (1995) showed that only small gains in efficiency, which do not justify the extra math ematical complications could be obtained by adding a third or further follow ups.


56 The DB method, as part of a larger category of iterative elicitation formats, is not completely free of starting point bias problems.10 Several different models to control for these problems have been proposed (See Cameron and Quiggin 1994; Herriges and Shogren 1996; Alberini, Kanninen, and Carson 1997; Whitehead 2002; DeShazo 2002) but results from these models show the gain in efficiency of the DB method is lost when controlling for the starting point bias.11 Statistical Models for Single Bounded and Double Bounded CV Dat a One must avoid being careless in modeling CV responses to different elicitation formats. Each elicitation format warrants a specific statistic al model. In this subsection we first present the appropriate statistical model to be used with the SB elicita tion format while emphasizing its microeconomic foundations. Hanemann (1984) made the link between SB-CV survey responses and the utility-maximizing agent of economic theory by specifying a Random Utility Model (RUM). We follow his discussion. We then presen t the two available approaches to model the WTP distribution. Hanemann (1984) proposed an Indirect approach to modeling the WTP distribution which remained the only way until Cameron and James (1987a) proposed the more tractable Direct Approach. At the end of this subsection we extend the SB statistical model to be coherent with a DB elicitation format. Hanemann (1984) starts by setting up the indi viduals indirect ut ility. The individual derives utility from consumption of a good q through );,,( xp yquj; 1,0 j; y denotes income, p represents a vector of prices, and x are other observable individual attributes that might affect 10 Some of the alternative explanations for starting point bias are the: (i) anchoring effect (ii) shift effect, and (iii) framing effect. 11 Flachaire and Hollard (2005) have recently proposed an alternative method that seems to correct starting point bias without losing the efficiency gains of DB.


57 her preferences (e.g., sex, age, etc)12. The individuals indirect utility can be considered as being composed by a deterministic por tion and a stochastic portion: 1,0,);,();,( jeyqvyquj j jx x (3-5) where ),,( yqpv is the portion of indirect utility wh ich is explicitly defined in the model, 0 j if q is not provided, 1 jif qis provided, and 1e and 0e are i.i.d. random variables that represent everything else that is not observable to the researcher in the jth state of the world. The addition of a random term is usually justified by arguing that, while the decision making process of rational maximizing agents is not random to the individual herself, there are elements involved in the decision-making process that are not observed by the researcher. The RUM approach enables economists to es timate a distribution for WTP. Single bounded model: two approaches reconciled As described earlier, in the SB-CV m ethod the respondent is asked the single question: Would you be willing to pay $Bid for a change from 0q to 1q?. An individual will be willing to pay for a given good if and only if after pa ying $Bid and acquiri ng the good her utility remains the same or is increased compared to the initial situation of not paying and not having the good. Because of the assumed stochastic nature of the indirect utility, we can define the probability of an acceptance to pay to secure provision of a good as: ]);,();,(Pr[} Pr{Pr0 0 1 1 1eyqveBidyqvpaytowilling individual x x (3-6) 1 0Pr1) Pr(Pr paytowillingnot individual (3-7) 12 For notational simplicity we do not always write utility as a function of prices in what follows, this has no bearing on the theoretical results since all changes in quantity of good q assume that prices are held fixed. We include x to emphasize that the parameters of the estimated distribution are a function of x.


58 Let 10ee and let F be the cumulative distribution function (cdf) of then we can solve the brackets in Equation 3-6 to obtain: )(Pr1vF (3-8) );,();,(0 1xx yqvBidyqvv (3-9) Equation 3-8 defines the probability that the in dividual is willing to pay the offered price $Bid and purchase the good; and depends on quantities 0q and 1q, on y and on the offered price, Bid. Typically, nF is assumed to be either the normal or logistic distributions obtaining the probit and logit binary ch oice models, respectively. The link between the individuals WTP, given by the theoretical measure C, and the empirical survey responses to an offer ($Bid) is made by motivating the bi nary choice model in a different way. When asked to chose whether or no t he would pay $Bid to secure a given change in q, the individual will accept to pay (respond yes) only if C$Bid and refuse otherwise (respond no). Thus the acceptance probability de fined in Equation 3-6 and Equation 3-8 can be expressed in an alternative manner as: )(1)Pr(Pr1BidGBidCC (3-10) By comparing Equation 3-10 w ith Equation 3-8 we can see: ))(()(1 BidvFBidGC (3-11) so that fitting the binary choice model ))(( BidvF is tantamount to estimating the parameters of the distribution of )(1 BidGC, the distribution of WTP. Two different approaches exist in the literature to model the distribution of WTP. We refer to these as the Direct Approach and the Indirect Approach. The Indirect Approach was initially proposed by Hanemann (1984), the Direct Appr oach (Cameron and James 1987a) came later.


59 However, we discuss them in inverted chr onological order to make the point that both approaches can be shown to be equivalent (dual). The Direct Approach, proposed by Cameron and James (1987a), is to specify the mean of CC (i.e., the population mean WTP) direc tly and add a white noise stochastic term: CC. The mean can then be assumed to depend on some covariates. Typically a linear specification is assumed, x C, thus WTP is modeled as: x C. (3-12) This model may be estimated as a linear regres sion in the case of an OE elicitation format, or by some discrete choice model if the DC format is used. The Indirect Approach, proposed by Hanemann (1984), uses the formal definition of )( Bidv derived in Equation 3-9. In th is approach the distribution of C is not directly specified but instead an assumption is made about the functional form of 1,0,);,( jyqvjx. For example, the linear specificati on is given by the following form: y yqvj j );,( x (3-13) j j jyqvyqu );,();,( x x (3-13b) To obtain )( Bidv we use the definition in Equation 3-9, Bid Bidv Bid Bid y Bidy yqvBidyqvv )( );,();,(01 0 1 0 1x x (3-14) where only 01 and can be identified. The statistical choice model for the SB format presented in Equation 3-11 then becomes,


60 )()(11BidFBidGPC (3-15) From this approach we could solve our model to get our expression for C. This is done first by solving Equation 3-13b for )),(,( yquqmyjj to obtain: jj j jjyqv yquqm ),( )),(,( (3-16) We can then plug Equation 3-16 into the Co mpensating Variation Function as follows,13 01 110 0 11 0 01),( )),(,( y y yqv y yquqmyC (3-17) This shows that the ratio of parameters estimates obtained via estimation of Equation 3-15 gives us an estimate of WTP. Moreover, if we let depend linearly on x through a regression such as x and write we obtain back Equation 3-12 which is the Direct Approach. This reconciles the two a pproaches for specifying a functional form for )( Bidv In essence, the Direct A pproach models WTP directly while the Indirect Approach models the indirect utility function to obtain esti mates that can be algebraically manipulated to obtain WTP estimates. McConnell (1990) showed the two approaches are dual or equivalent. In 13 See Hanemann (1991) for a formal derivation of the Compensating Variation function.


61 general, for any given regression formulation of a WTP distribution, one can always find a RUM formulation which generates this distribution. In this sense, an y given WTP distribution can be derived using either approach (Carson and Hanemann, 2005). However, the Direct Approach has become the most popular due to its tractabi lity and the easiness with which one can obtain the marginal effects of indivi dual characteristics on WTP. We finish deriving the appropria te statistical model for the SB elicitation method under the Direct Approach framework. Going back to our linear specification under the Direct Approach we rewrite Equation 3-12 for individual i: ii C i iWTPC x (3-12) where x denotes a set of individual characteristics thought to ha ve an effect on WTP. In practice, each respondent receives a randomly chosen threshold bid offer, iBid. Her response depends on the magnitude of her WTP which is not observed, we observe only a dichotomous yes or no answer, iy; i C i i C i iBid WTPifno Bid WTPifyes y 0 1 (3-18) using Equation 3-12, her probability of accepting (i.e., answering yes) the offer is given by )/)'(Pr( )' Pr( ) 'Pr( ) Pr()|1Pr( ii i ii i i ii i C i i i y iBidz Bid Bid Bid WTP y x x x x (3-19) here we assume ),(~ N so that iz is distributed standard normal, this results in a probit model given by, )/)'((1 ii y iBid x (3-20)


62 )/)'(( ii n iBid x (3-21) While a regular probit yields parameter estim ates only up to a factor of proportionality (i.e., /), Cameron and James (1987a) show that the variability in iBid(due to its random assignment across the sample) allows us to identify and estimate and separately. The estimation technique resembles an ordered pr obit with given cutoff points (also known as interval data estimation). Double bounded model We now develop the appropriate statistical m odel for the DB elicitation method. The model presented here forms the basis for the mo dels used throughout this study. The model is an extension of the SB model. In the SB case we had a dichotomous (1=yes, 0=no) dependent variable, iy in response to a single bid offer, iBid. In the DB methodology a second question is asked with a higher offer, i H iBidBid if the response to the first question is yes and a lower offer, i L iBidBid otherwise. Combining the answers to both questions we obtain four different possible scenarios, yesyes noyes yesno nono di,4 ,3 ,2 ,1 If we assume the utility maximizing respondent queries his preferences to answer each of the questions, we can derive the formulas fo r the respective likelihoods (See Hanemann, Loomis, and Kanninen 1991) of each outcome. The likelihood for individual i giving a 4 yesyesdi response is,


63 } Pr{ } Pr{} | Pr{ } Pr{ ),(i H i i H i i H ii i i H i i i H ii yy iWTP Bid WTP Bid WTPBidWTPBid WTP BidandWTPBid BidBid );(1H iBidG (3-22) since by definition i H iBidBid so that 1} | Pr{ i H i iWTP BidWTPBid; where G is a cumulative distribution function with estimable parameters Similar reasoning is used to derive the remaining likelihoods, );();(),( i H i H ii yn iBidGBidGBidBid (3-23) );();(),( L i i L ii ny iBidGBidGBidBid (3-24) );(),( L i L ii nn iBidGBidBid (3-25) We can then write the log lik elihood for the DB model as; ),(ln ),(ln ),(ln{)(ln1 L ii ny i ny i H ii yn i yn i N i H ii yy i yy iBidBid d BidBid d BidBid dL )},(lnL ii nn i nn iBidBid d (3-26) Where ny i nn i yy iddd ,, and yn id are binary-valued indicator va riables for each outcome. Using the linear specification for the Direct Appro ach (Equation 3-12) to model WTP and assuming a normal distribution for i we obtain the log-likelihood for the DB elicitation format, )]/)'(()/)'((ln[ )]/)'(()/)'((ln[ )]/)'((1ln[{)(ln1 i L i ii ny i ii i H i yn i N i i H i yy iBid Bid d Bid Bid d Bid dL x x x x x )]}/)'((ln[i L i nn iBid d x (3-27)


64 Which is maximized for ],[ The estimated are directly interpreted as the marginal effects of x on WTP. Stated vs. Revealed Preferences Econom ists distinguish between revealed (or observ ational) preference (RP) methods and stated preference (SP) methods to study individual behavior. RP met hods (i.e., revealed preference theory) are heavily relied on by economists for estimating parameters to explain demand for market goods. In RP methods consumer s reveal information about their tastes and preferences through their behavior in real markets. Consumers pr eferences are revealed in the market by the quantities they choose to consume at different prices. Because environmental goods seldom reach any market, environmental economists were pioneers in pointing out the data availability limitations of the RP approach and the possibilities inherent in the SP approach to tackle this. For economists to be able to estimate the value of nonmarket goods or new unmarketed products, for example, it usually becomes necessary to reach into the realm of SP methods su ch as CV where the value the consumer associates to the good (WTP and WTA) is directly elicited from her through survey quest ionnaires about some hypothetical market situation. Sc helling (1968) makes the point qui te clear by saying that while the price system is one way to find out what thin gs are worth to people, another way is to ask them directly. Proposed early by the famous environmental economist Ciriacy-Wantrup (1947), who wanted to measure the dollar benefits of soil cons ervation, SP methods have been of interest in the academic world since the late 1940s. In his book Resource Conservation: Economics and Policies, Ciriacy-Wantrup (1952) advocates for the use of the direct interview method to measure the values related to natural resour ces. In a contemporaneous contribution Bowen


65 (1943) reached the same conclusion when studying the welfare benefits of beautification of landscapes, and suggested the use of polls. It was Robert K. Davis (1963b) who was first to use CV methods in his doctoral dissertation at Harvard. Regardless of the potential advantages asso ciated with embracing SP methods, their use has borne some antagonism from traditional RP economists for several reasons. Economists usually shy away from SP approaches in fa vor of RP economics. One reason is that many economists believe that SP methods may spur strategic behavior from respondents and as a result what people say they would do in a market during an artificial ma rket session is not necessarily what they actually do in the real world situa tion. Samuelson (1954) makes this point in his seminal paper: It is in the selfish interest of each person to give false signals, to pretend to have less interest in a given collective activity than he really has. Milton Friedman also contributed to the debate with his famous analogy of the pr ofessional pool player not knowing the underlying physics behind the shot. In his view, maximizing agents might behave as such even without consciously knowing they do so, as a result their survey responses are meaningless (not linked to their behavior). The debate between RP supporters and SP supporters has lasted more than three decades. At a beginning, attempts were made to validate CV results by comparing them to WTP estimates obtained via RP methods (e.g ., Hanemann 1978; Cameron 1992; Adamowicz, Louviere, and Williams 1994). In Resource Economics, the most commonly RP methods used for valuation and comparison are the hedonic pricing and the travel cost method. Results from such comparisons have been inconclusive, showing convergence in some cases and divergence between RP and SP methods in others.


66 More recently, a different approach has b ecome popular which views both SP data and RP data as flawed in differing aspects. In this view SP and RP methods are seen as complimentary to each other rather than as opposing and mutua lly exclusive. Both methods are thought as containing important but incomplete information a bout the preference structure of consumers. As such, each method has its relevanc e and applicability. Under this paradigm, the convergence tests applied in studies mentioned above make no se nse since both methods are imperfect and one cannot arbitrarily pick one or the other as a benchmark. A result of this late conceptualization is the production of studies that combine SP and RP data to formulate a more complete picture of consumers preference structures (Azevedo, Herriges, and Kling 1993; Cooper 1997; Hubbell, Marra, and Carlson 2000; Qaim and DeJanvry 2003). Existence Value and Alternative Methods to O btain WTP Measures of Non-Market Goods This subsection presents some of the alternative methods that are commonly us ed to obtain estimates of WTP for non-market goods. In some cases, regardless of observable market data unavailability, economists are sti ll able to estimate part of the value of nonmarket goods by using specialized indirect-observed be havior methods such as hedoni c pricing (Rosen 1974) or the travel cost method (Clawson a nd Knetsch 1966). For example, a hedonic pricing study would estimate the value of a nonmarket good such as air quality by modeling the price of real estate as a function of: characteristics of the property, air qu ality, and other attributes of the housing zone which may or may not be pure public goods (i.e., non-market valuated goods ). The idea is that the value people assign to real state is composed by the value of the pr operty per se, plus the value of nonmarket attributes of the housing zo ne surrounding the propert y (e.g., air quality, crime rate, parks and recreation area, closeness to main roads, etc); as such, the parameter estimate associated with the air quality variable in our example is interpreted in hedonic pricing studies as the component value of real state assigned to ai r quality by the purchasing


67 individual. As opposed to direct market data on air quality prices which is inexistent, market data on real estate prices is readily available. A second example w ould be the travel cost method which is typically used to estimate economic use value associated with recreational parks or natural ecosystems. The method estimates indivi duals WTP (or WTA) using data on the number of trips made to visit the park at different trip di stances and different travel costs. The idea is that the cost of the trip provides ev idence of the value the individual assigns to the nonmarket good, in this case the natural park. Other examples of a lternative indirect RP me thods used to estimate the value of nonmarket goods include the household produc tion model and the averting expenditure method (Freeman 1993). All of these methods suffer from the same lim itation, as we have tried to emphasize in the preceding discussion by using quotati on marks, these methods are able only to measure what is known since Krutilla (1967) as th e use value of nonmarket g oods. Krutillas contribution was influential in that it argued, quite convincingl y, for the importance of non-use values (or existence value) and defined them clearly enough to distinguish them from conventional use values. Non-use value covers situations in which individuals who do not use, nor plan to use, a commodity would nevertheless feel a loss if the commodity would cease to exist. Many individuals, for example, would assign a dollar value to things such as the existence of the Amazon Rainforest even if they do not have plans of ever visiting or using it. From an economist perspective, the total economic value (TEV) of a commodity is the sum of its use value (UV) and its non-use value (NV):14 14 Kerry Turner (2002) argues that a broader categorization of values which encompasses TEV=UV+NV as part of what he calls Anthropocentric Instrumental Value can be done by including four categories based on whether the valuer is a human being or not; and whether the value assigned is of instrumental or intrinsic type. The categories are: (1) Anthropocentric Instrumental Value, (2) Anth ropocentric Intrinsic Value, (3) Non-Anthropocentric


68 NVUVTEV (3-28) Since market price data is inevitably tied to a decision to consume, market prices hold only information about the use value consumers give to things; lacking info rmation about existence values. Thus, in general, indirect-RP methods ba sed on market data are only able to obtain estimates of the relevant use values while unable to obtain existence values. Therefore, it may be argued that if the non-use portion of value is quite large market-data driven methods obtain defective TEV estimates. There are some indirect RP me thods that can, under certain circumstances, obtain estimates of non-use value. Analysts can switch attention from the market system towards the political system and, for example, estimate demands fo r local public goods using the collective choice method (Oates 1994). Collective choice models make use of the theory of the median voter proposed initially by Duncan Bl ack (1948). In this approach, by assuming simple majority voting, the results of an election are shown to be equivalent with the preferences of the median voter. Candidates adjust their polic y stances to match median voter s preferences as closely as possible in hopes of being the winning ticket. Th erefore, the quantities offered of a public good in a given municipality can be thought of as poi nts in the demand function of a single voter with median preferences. In other words, we can obtain price and income elasticities of public goods by simply treating quantities offered of the public good in different municipalities as the dependent variable, and using median income and tax share of median voter as the independent variables (we can also include other socio-demographic variables as long as they are Instrumental Value, and (4) Non-Anthro pocentric Intrinsic Value. The UV is bounded by the NV which in his view is also anthropocentric since it is composed of (i) intrag enerational altruism (i.e., vicarious consumption) (ii) intergenerational altruism (i.e., bequest value), and (iii) stewardship motivation; which are all human motivations that require a human valuer (i.e., are Anthropocentric).


69 representative of the median voter).15 What the method essentially is suggesting then is taking an observed outcome in a community and associat ing it with a point on the demand curve of a decisive voter; then, each jurisd iction serves as a unit of obser vation (Oates 1994). The potential limitations in gathering the data needed for this type of study become obvious when we think of obtaining only one observation per ju risdiction; in some cases the logistics and costs associated might prove impossible. All of the methods mentioned so far are popul arly used in the environmental valuation theory. There is another group of non-market va luation techniques worth mentioning, these are known as Conjoint Analysis (or choice modeli ng) methods (Green and Rao 1971). They are SP techniques that are more commonly used in the marketing community. Even though these methods were not pursued in this study, it is worth mentioning them so as to have a more complete picture of the methods available to perform economic valuations in non-market situations. Conjoint Analysis (CA) techni ques are very similar to CV; th ey are in essence a multiple attribute valuation exercise.16 CA and CV, however, differ in th eir origins. CA techniques hold their origins in the psychometric literature and their development in the marketing literature. CV grew out of the need to valu ate non-market environmental go ods in the resource economics literature. Both CV and CA, however, share thei r statistical foundations in the development of discrete choice models.17 15 The method also allows for estimation of the grade of publicness of the good via a single parameter (See Borcherding and Deacon 1972; Bergstrom and Goodman 1973). 16 For an introductory text on Conjoint Analysis see Orme (2005). 17 See for example Luce (1959), Marschak (196 0), and McFadden (1974) for the logit model.


70 In a CA exercise the respondent is presen ted with an array of hypothetical potential products showing different combinations of previously selected salient attributes. Depending on the exact type of CA method bei ng used, the respondent is aske d to rank or to express their preferred choice among the presen ted products. In Rank CA, for ex ample, the individual is presented with profiles of products with varying pr ices and attributes and is asked to rank them or rate them. In this way the researcher is able to obtain estimates of latent utilities (part-worths) that emulate choice behavior. In Choice Based Conjoint (CBC), on the other hand, the individual is asked to choose their preferred product, as op posed to ranking every single product presented. CBC is in essence a multinomial choice CV (i.e., where the choice set has more than one good). One problem with Full Profile CA is that even a small number of goods, prices, and attributes gives rise to a large number of alternatives which increases costs of carrying this type of study and may overwhelm the respondent. Adap tive CA (ACA) improves on this by using classic Experimental Design methods to elimin ate unnecessary alternatives (Johnson 1974). An improvement on CBC and ACA is possible by usi ng Hierarchical Bayes estimation methods (Allenby, Arora, and Ginter, 1995). We have discussed the methods used in non-ma rket valuation studies and some of their advantages and disadvantages. Many of the met hods presented here are not applicable to our case of pre-testing market technol ogies and studying their potentia l for adoption. The discussion, however, illustrates the concept of existence value (or non-use va lue) and some of the reasons why CV is preferred in many situations. Two distinct groups are identified: those that can obtain existence values and those that cannot (See Figure 3-1). In gene ral, SP methods obtain existence values, while RP methods do not.18 18 In Figure 3-1, Rank CA, CBC, and ACA are listed und er choice modeling as contingent ranking, choice experiments and paired co mparisons, respectively.


71 Among SP methods (which are in fact applicable to our case ), two groups or methods, one derived from the marketing literature (i.e., CA ) and another derived from the resource economics literature (i.e., CV) were discu ssed. CA methods are a powerful tool for obtaining simultaneous valuations of multiple attributes, but ar e prone to large implementation costs. CV methods are a better fit for the scale and for the attribute by attribute valuation targeted by this study, plus they allow for captu ring any potential existen ce values. Non-user (or non-adopter) valuation of GM crops is important because of the controversial facets and uncertainty associated to the te chnology (See Chapter 2). The tota l value that farmers associate with a GM crop may be in part given by its use va lue (i.e., profitability) but also may be affected by its existence value (i.e., a negative penalizatio n due to uncertainties an d controversial aspects of the technology or a positive additional value associated with non-market benefits such as ease of use).


72 Figure 3-1. Total economic value and valuati on techniques. (Source: Pearce and zdemiroglu 2002).


73 CHAPTER 4 ESTIMATING PRODUCERS WT P FOR CORN SEED TR AITS NOT YET IN THE MARKET In Chapter 2 we presented an overview of ge netically modified (GM) crops and the GM crop industry. From our di scussion there, the relevance of GM crops to international policy is evident. GM crops have the potential to incr ease our chances of success in utmost important goals we have set regarding re duction of world hunger and protect ion of the environment. The use of biotechnology presents an opportunity to develop highly sustainable (high-S)1 seed technologies that, if prope rly diffused, have the potential of improving food security for future growing populations while at the same time minimizing impact to the environment. On the other hand, the controversial facets and uncertain long -term outcomes that many associate with using the technology also call for the attention of policy makers. Several GM crops have alrea dy been developed and market ed. Where made available, these technologies have obs erved widespread adoption.2 The four most widely adopted GM crops have been: soybeans, corn, cotton, and canola. In terms of GM traits, the two most widely adopted have been the herbicide to lerance trait and the insect resistance trait. Many other are at different stages of the development process. Two important GM traits currently in advanced or medium stages of development are: (i) a trait that increases the crops tolerance to drought, and (ii) a trait that reduces inefficiency in nitrogen fertilizer (N-f ertilizer) applications by increasing the crops N-absorption efficiency. These two traits and their application to corn seeds are the focus of this dissertation. 1 High-S technologies are defined in Chapter 2. The I-PAT equation is commonly used to measure human impact on the environment. I represent such im pact, while P represents population, A represent output per capita, and T represents the technological state-of-t he-art measured by the environmental impact per dollar of income. A high T technology is deemed not environmentally friendly. If we define S as 1/T, we can define a sustainable technology as a high-S technology. 2 See section Current state of the GM crop industry in Chapter 2.


74 In specific, we focus on studying the adoption po tential and valuation of these traits by corn farmers in Minnesota and Wisconsin. In order to make a true assessmen t of the benefits and global adoption potential of these traits, it would be ideal to ob tain farmers valuation on a region-by-region basis (i.e., developed vs. developing; by continen t; or by country). Such task, however, results prohibitively expensive for a sing le study. Our intention here is not to make a complete assessment on a global scale; rather, the main contribution and goal of this study is to obtain results at the local level that provide valuable insights which may guide further studies evaluating adoption of these traits on other region s. Also, the results we obtain for U.S. corn farmers may serve as a benchmark for comp arisons with future valuation studies. On a more practical and more immediate level, a better understanding of the adoption potential and of farmers valuation for thes e forthcoming GM seed products should prove valuable both to the private and pub lic sectors for several reasons. The effective and timely development of i nnovative GM crop technologies necessitates participation from both the private and public sect ors. Given appropriate incentives, the market will respond to the demand for agricultural innovations. However, without the existence of a patent system and the appropriate enforcement of intellectual property (IP) laws that ensure appropriate incentives are in place, the market will fail to respond and will invest too little in research and development (R&D). This fact is well known in economics, being a result of the public good characteristics of pure scientific knowledge. In esse nce, a granting of a patent confers monopoly power to the r ecipient; the prospects of being able to charge monopoly prices to recover R&D costs and observe a gain are what motivate the inventor to engage in the long and costly process of innovation. Thus patent systems play an important role in securing adequate investments in R&D and promoting th e creation of innovations in the market. The


75 implementation of a patent system for the ca se of GM crops, while observing some initial resistance from some groups, has been successf ul in promoting R&D investments and producing seed innovations. In fact, the exis tent GM crops, including those al ready in the market and those in process of gaining market approval, have almost exclusively been developed by private companies as a result of large investme nts made by these companies in R&D. In practice, the patent system, although effec tive in setting appropr iate incentives, may restrict access to scientific information and thus prevent the creation of further innovations that develop on existing patented innovations. This obse rvation has led to public investment in R&D at the basic science level to complement pate nt systems. The basic knowledge produced by public investment in R&D is then freely available to firms and efficiently used by the market to produce a myriad of practical innovations.3, 4 Beyond the issues of innovation development, th ere is the issue of adoption and diffusion. Developing a new high-S t echnology is one thing, to have it adopted on a wide spread basis is another completely different thing. For true success, a technological development must be followed by its adoption and widespread diffusion. Final adoption is not determined in the science laboratory, but in the market. From the supply side, a key f actor in determining adoption rates is sale price. The functioning of patent systems implies that in or der to recover their R&D investments and observe gains, it is reasonable to believe that private innovators in the GM cr op industry will pursue 3 Of course, this last point depends on disclosure policies an d patent structures at the public sector level. However, information sharing and disclosure should in general be higher in public research because it is financed by public funding. 4As discussed in Chapter 2, even when appropriate patent systems are in place, public-private research partnerships may still play a role in developing innovations that may improve the lives of many but whose effective demand is hindered by low levels of income. Such is the case of developing GM crops tailored to the needs of small poor farmers in developing countries. In these places, public-private partnerships may have a better chance of success in developing and delivering such GM crops at prices that are affordable to farmer s and profitable to private enterprises.


76 monopolistic pricing strategies. In welfare economics, if a product is desirable, aggregate welfare will be maximized when the product is sold at competitive prices. Monopolistic pricing above the competitive price level will increase seller surplus and reduce buyer surplus. In terms of adoption, prices above the competitive price will resu lt in lower relative adoption rates. This is of concern to policy makers who may want to overs ee appropriate diffusion which may be hindered in the market by overly aggre ssive monopolistic pricing. From the demand side, customer preferences play a big role in adoption decisions. The maximum amount an individual is willing to pay for a given quantity of a product is termed his willingness-to-pay (WTP). An individuals perc eived value for a product is embodied in his WTP. It is the interaction between WTP and sale price that ultimately determines adoption in the market place. An individual will adopt (purchase) a GM crop technology if his WTP exceeds the sale price; otherwise, he will refuse to adopt. Setting the right sale price for a product is maybe the most im portant and also the toughest task for a company.5 Under the non-competitive setting conferred by patent systems, private firms are able to exert some control over pricing. Avoiding excessive prici ng is of interest not only to policy makers overseeing adoption, but also to the seed companies that sell the innovation. In the absence of an adequate inst rument to provide them with the necessary knowledge, firms may overestimate customer percei ved value and set excessi ve prices at levels incompatible with customers WTP. In this case adoption may be so low that profits obtained are less than optimal. This danger is particularly true for seed companies, which have little experience in pricing of GM crops (Hubbell, Ma rra, and Carlson 2000). This is evident, for example, in the early years of Bt cotton marketin g. Bt cotton was initially marketed in 1996 at a 5 Price is the only element in the marketing mix that generates income; all the other elements (advertising, placement, packaging, etc) generate costs. More so, price is the elem ent that is most easily adjusted.


77 small premium over conventional seeds plus a te chnology licensing fee of $38/acre. In 1997, the second year of marketing, a di scount pricing offer of $10/acre fo r the first 50 acres planted by new adopters was introduced to improve on prior adoption rates. Still, the adoption rate that would occur at the discounted price level was overestimated by seed companies who had prepared seed for planting 7.5 million acres bu t sold only seeds for 5.5 million acres (Hubbell, Marra, and Carlson 2000). In any case, logic tells us that innovations in general, which by definition are new in the market and new to firms and customers, are initially difficult to price. Besides low adoption and ineffici ent profit levels, failure to understand the consumer from the part of seed companies may result in fa ilure to recognize opportuni ties for segmenting the market. Missing such market-segmenting opportuniti es and failing to identify different types of customers and what they value will also result in flawed pr icing strategies. In marketing theory, firms are considered to hold two distinct approaches to pricing: (i) the value-based approach, and (ii) the cost-based approach. In the value-based approach companies base their pricing decisions on information a bout customers perceive d values. The company first estimates the customers perceived value for the product. After th is, the decisions about product design, costs incurred in its development, and pricing, are simultaneously based on the firms improved knowledge about the customers perc eived value. In contrast, in the cost-based approach, the product is first deve loped and price is set by decidi ng some desired profit margin above of incurred costs. A worki ng value-based pricing strategy, being based on more and better information, holds the potential for higher effici ency and larger profits compared to the costbased approach (Monroe, 2003). Besides informing pricing strate gies, knowledge of farmers va luation for different traits may help companies decide which seed products are worth developing. Given the large costs


78 associated with the development of each seed pro duct (i.e., trait), mainly R&D costs incurred in identifying and testing product concept, it woul d be valuable for seed companies to have information about the value that farmers associate with a given trait before incurring such costs. In other words, it would be valuable to have an idea of the adoption potential of each seed product on a trait-by-trait basis at the early stages of product deve lopment, before investing large sums in it. Valuation on a trait-by-trait basi s is possible due to recent tr ends in market differentiation strategies of GM seed products. Crop seeds and primary agricultural products have been traditionally considered commod ities with little space for di fferentiation. Commodities are products for which there is a demand but which ar e marketed with no qualitative differentiation. With the advent of biotechnology, the line separating such crop seed s into undifferentiated commodities rather than as differentiable goods seems to be fading. Seeds today are conferred traits that are advertised in marketing programs designed to diffe rentiate them as a product. The market for corn seeds today is filled with numer ous differentiated seed brands (e.g., Monsantos RoundupReady and YieldGard product lines, Dow AgroSciences Herculex seeds). Differentiation seems to have occurred at two le vels: by trait and by type Differentiation by trait has been facilitated by the in creasing popularity of asexual methods in seed production. Recombinant ribonucleic acid (RNA) and tissue culture techniques allow seed companies to create a seed product that is more consistent in its attri butes (traits) than what conventional breeding allows. Differentiation by type, into GM and conventional (nonGM), has also been apparent. This chapter hopes to inform the decisions of both policy makers and seed companies by developing estimates of farmers mean WTP (in $ pe r acre) for two corn seed traits that are not


79 yet in the market but appear as potentially profita ble to private companies and are highly relevant to the global search for future food security a nd environmental sustainability. We compare two different versions of the two traits, one versi on in which the trait is obtained by conventional selective breeding (i.e., nonGM) and another version in which RNA recombinant techniques are used (i.e., GM). The two specific traits considered are: (i) a trait that maint ains same yield but reduces Nfertilizer requirements by one third, and (ii) a tr ait that increases the crops DT so that under severe drought conditions it would still yield 75% of the normal yi eld. In all of the following we will refer to these two as fertilizer saving (FS) trait and drought tolerance (DT) trait. To the best of our knowledge, no other fo rmal study, as of yet, has cons idered valuation or adoption of the traits studied here. In addition to estimating mean WTP, the me thods used in this chapter allow us to understand what farm or farmer characteristics ha ve an effect on farmers WTP for these traits, what is the direction of that effect, and what the size of that effect is. Estimating WTP for both a nonGM and a GM versions provides us with the opportunity to gain some valuable insights (or confirm those found in the litera ture) with respect to these effects. We also show how the estimates obtained on farmers WTP for a given technology can give us some idea of its adoption potential. Finally, we compare the estimated adoption potential of the di fferent versions (nonGM and GM) of each trait. Theoretical Model Since no m arket choice data is available for thes e, as of yet, unmark eted technologies, we use a Contingent Valuation (CV) method to estimate farmers WTP fo r each trait. CV is a stated preference (SP) approach that has been extensively used to obtain estimates of WTP of nonmarket goods in the environmental econom ics literature. Cameron and James (1987b)


80 suggested that the CV method can be equally useful in pretesting new market goods. The CV method obtains WTP estimates of goods in nonmarket situations.6 A similar approach to the one taken here wa s also used by Hubbell, Marra, and Carlson (2000) in estimating adopters and non-adopters WTP for Bt Cotton among farmers in the United States, and similarly by Qaim and DeJanvr y (2003) in estimating farmers WTP for Bt Cotton in Argentina. In those studies the use of CV was warranted because market data is not available for the non-adopters that those st udies aimed to include in their models. In our study, the double bounded (DB) dichotom ous choice (DC) elicitation format was selected over other possible formats for several reasons. In the DC format the individual is presented with a price scenario and asked about her purchasing decision yes or no. This approach has been shown to better resemble a real market situa tion compared to the alternative OE format which asks directly to the respondent How much would you pay? As a result, DC formats are considered to give more valid es timates than open ended (OE) formats. Federal guidelines (U.S. Department of Commerce, 1993) recommend using the DC approach. Several types of DC formats are ava ilable: single bounded (SB), double bounded (DB), and iterative bidding (IB).7 These elicitation formats differ in the number of price scenarios and therefore in the number of DC questions asked. The SB-DC approach considers one single price scenario and asks one single yes or no que stion. The IB-DC gradually increases (decreases) the price asked if the respondent answers yes (no), and stops when the respondents answer switches to no (yes). The DB-DC asks one in itial and only one follo w-up question. The price scenario presented in the follow up question is higher (lower) if the individual answers yes 6 See Chapter 3 for a more complete presentation of the CV method. 7 See Chapter 3 or Chapter 5 for a detailed description of each of these DC elicitation formats.


81 (no) to the initial question. The DB-DC approach is usually preferred over the SB-DC method since it provides much larger statistical effici ency for same sample sizes (Hanemann, Loomis, and Kaninnen 1991). The gains obtained from addi ng a third (or further) question to the DB-DC method are insufficiently large to justify the added mathemati cal complications (Cooper and Hanemann 1995). Willingness-to-Adopt In this sec tion, the abstract theories (WTP, random utility model (RUM), and CV) presented in Chapter 3 are applied and adapted to the case of technology adoption, which is the situation concerned by this study. In the case of technology adoption, th e individual faces two possible actions: adopting the technology or refusing to adop t. In this case, a WTP measure makes more sense than a willingness-to-accept (WTA) measure.8 An individual will be willing-to-pay for the technology a price per unit of P and adopt such technology if her utility w ith the profit (net of the tec hnology cost) provided from adopting the technology minus the cost of the technology is at least as high as her utility without the technology. Formally, the indivi dual will adopt a new technology if9 );,();,(0 0 1 1x x yquPyqu (4-1) where 1q indicates adoption, 0q indicates non-adoption, 1y and 0y are profits (net of technology costs) with and without the technology, respectively, and x are individual characteristics and characteri stics of the production unit. Given that utility is only pa rtially observable to the analyst we use the RUM approach 8 It would be awkward to as k the individual how much he would be w illing to accept (in monetary terms) to disadopt a given technology. 9 We make a slight switch in notation from Bid to P to be tter reflect the technology adoption situation (See previous chapters).


82 }1,0{ ,);,();,( j yqvyquj i j i jx x (4-2) where );,( xi jyqv represents the observable portion of utility associated with technology situation j and j is a stochastic zero mean term representing the respective unobserved portion of utility. In this RUM framework, the decision to adopt (Equation 4-1) may be reexpressed as; 0 0 0 1 1 1);,();,( x x yqv Pyqv (4-3) Following the Indirect Approach (See Chapter 3) a linear specification, common in CV studies, may be assumed for the individuals indirect utility )(' Py vj j j x (4-4) where represents the marginal utility of inco me. Applying Equation 44 to Equation 4-3 and solving for 10 gives )(' Pyj x (4-5) where 01 and 01yyy The change in profits (y ) is unfortunately unobserved, not even for the individual himself, since it be longs to the counterfactual world. An assumption is necessary in our application, th at the expected change in profits y can be explained by some of the individual and farm characteristics contained in x, so that y is implicitly included in x. Measuring v in monetary terms means that1 Thus we may express the probability of adoption as )'Pr()Pr( P adopt x (4-5b) In the DB-DC format, individuals in the sa mple are randomly assigned an initial price scenarioiP. Based on their answer to the initial pric e scenario, they are offered a higher price


83 H iP or a lower price L iP in the follow-up question. In an y given price scenario, a given individual will answer yes and adopt the techno logy only if his WTP is larger than some asked price iP (i.e., iWTP* iP ). Four possible answer sequences to the initial and follow-up questions are possible yesyes noyes yesno nono di,4 ,3 ,2 ,1 (4-6) The likelihood of an individual giving a response sequence 4, yesyesdi is given by ) Pr( ) Pr() | Pr( ) Pr(),(i H i i H i i H ii i i H i i i H ii YY iWTPP WTPP WTPPWTPP WTPPandWTPP PP (4-7) where we have used 1 ) | Pr( i H ii iWTPPWTPP The term YY i represents the likelihood of adoption by individual i at some offered price H iiPP *. Assuming is distributed ),0( N this likelihood may be expressed (using Equation 4-5b) as 1 )'Pr() Pr(),(i H i H ii H ii i H i H ii YY iP P P WTPP PP x x x (4-8) Let NY i YN i,and NN irepresent the likelihoods of the re maining possible answer sequences shown in Equation 4-6. For example, the likelihood of adoption by individual i at a price *iP :H iiiPPP *, is given by YN i. Applying a similar reasoning to that shown in Equation 4-7


84 and Equation 4-8 to these remaining likelihoods the log-likelihood func tion of our willingnessto-adopt model for a sample of Nindividuals is given by )]}/)'((ln[ )]/)'(()/)'((ln[ )]/)'(()/)'((ln[ )]/)'((1ln[{)(ln1 i L i NN i i L i ii NY i ii i H i YN i N i i H i YY iP d P Pd P Pd P dL x x x x x x (4-9) where NY i YN i YY iddd and NN id are binary-valued indicator variables for each answer sequence. Estimation of Equation 4-9 is made using maxi mum-likelihood methods to obtain estimates of },{ The estimated may be directly interpre ted as marginal effects of x (in dollars per additional unit of x) on WTP. Empirical Application W e apply CV methods to the case of technol ogical adoption of new GM corn seeds (DT and FS traits) by farmers in Minnesota and Wisconsin. Using a DB-DC elicitation format, corn farm ers in Minnesota and Wisconsin were asked about their WTP (in $ per acre) fo r having a FS trait added to thei r current most-used corn seeds. WTP was elicited for two different versions of the same trait. In the first version of the trait the farmer was told that the trait was to be a dded to the seed by conven tional selective breeding methods (i.e., nonGM), in the second version the farmer was told that the trait was to be conferred to the plant via gene tic modification (i.e., GM). A second trait was also considered in a sep arate DB-DC question. In this second CV question farmers were asked about their WTP for having a DT trait added to their current mostused corn seeds. Two different versions of the trait (nonGM and GM) were also presented for this second trait. Table 4-1 shows the four di fferent traits that were presented for farmer


85 evaluation. The CV questionnaire po rtion of the survey used to elicit WTP for both traits and both versions may be seen in Figure 4-1. Data and Surveys The data for this study was m ainly provided by two separate intervie w-based surveys: (i) the 2006 Corn Poll (CP06) and (ii) the 2007 Corn Poll (CP07). Both surveys were administered to corn farmers in the states of Minnesota and Wisconsin by the University of MadisonWisconsin Program on Agricultural Technol ogy Studies (PATS). The CP06 (conducted in 2006) contained sections of questions asking 945 randomly selected corn farmers about their individual demographics, farm ch aracteristics, purpose of corn production (i.e., grain, silage, sweet corn, or other), and previous experience with GM-corn varieties. The CP07 (conducted in 2007) was administer ed to 451 randomly selected farmers and was in essence a short CV survey designed to complement the CP06 and find out which corn seed technologies farmers are usi ng and how they value certain traits. The CP07 also yielded data on production practices, insurance practices and, most importantly, it provided the CV responses per se. The CP07 listed 9 different seed codes that identi fy each of the corn va rieties planted in the two states. Seed codes and varietie s are listed in Table 4-2. Farm ers participating in the survey were asked to identify the seed code with highest acreage in their farm. Data on production practices (i.e., yield and costs) and CV responses was gathered w ith reference to this highest acreage seed code. The CV questionnaire had four different ve rsions (A, B, C, and D) each presenting a different initial price scenario and also varying in the subseq uently presented follow-up price scenarios this is typical of CV surveys. Table 4-3 shows the different versions and their


86 respective price offers. For example, version A first asked would you pay iP$10 dollars for the seed; if the respondent answer ed yes to this init ial question the survey then asked would you pay H iP$15?; if the respondents answer to the initial question was n o he was instead asked would you pay L iP$5? In contrast, version B asked a different initial price of iP$15 instead of the iP$10 asked in version A; and it also asked different follow-up prices of L iP$15 and H iP$25. The 451 randomly selected farmers who partic ipated in the CP07 were randomly assigned one of each of the different versi ons of the survey. This randomi zation is made to avoid possible bias in the estimation. Also, the produced variation in iP is what allows us to identify instead of only being able to identify / (Cameron and James 1987a). Farmers in the CP06 were assigned a unique identification number. The same identification number was maintained for each fa rmer in the CP07. This allowed us to merge both data sets and obtain a data set containing only those farmers that were randomly selected to participate in both surveys. Out of the 451 total fa rmers participating in th e CP07, only a total of 345 farmers also participated in the CP06. These 345 farmers form the base sample for this study.10 Factors Affecting Adoption and WTP W e should clarify that the focu s in this study is adoption as opposed to diffusion. In the technological adoption literatu re, adoption studies focus on i f and to what extent the technology is adopted by a given farmer at a given point in ti me (Does the farmer adopt the 10 Since both surveys were admi nistered randomly, we have no reason to suspect any bias is created from reducing the sample size to 345 farmers when merging both polls.


87 technology? If so, on how many acr es does he use it? ). Diffusion studies, on the other hand, focus on the dynamic evolution of aggregate adop tion through time in a given social unit. Static (Griliches 1957) and Dynamic (Knudson 1991; Fernandez-Cornejo, Alexander, and Goodhue 2002) diffusion models are the most popularly used when studying the diffusion phenomenon. Choice models based on farmer profit comparisons (Q aim et al. 2006) or expected utility comparisons (Payne, Fernandez-Cornejo, and Daberkow 2002) and willingness-to-adopt models (Hubbel, Marra, and Carlson 2000; Qaim and DeJanvry 2003) are some of the approaches that have been u sed to study adoption decisions. Among the factors influencing adoption, an innov ations profitability ov er the profitability of traditional alterna tives has long been considered a ke y factor (Griliches 1957; Sunding and Zilberman 2001). Besides profitability most studies in the literature also acknowledg e the role of farm and farmer heterogeneity in explaining adoption decisions (Feder, Just, and Zilberman 1985; Khanna and Zilberman 1997; Fern andez-Cornejo and McBride 2002). Before continuing, let us clar ify that our model holds two in terpretations of how farm and farmer attributes influence the adoption decision. Taken literally, the linear specification in Equation 4-4 implies that profitability (y) is one factor influencing adoption decisions, but final adoption decisions are made based on comparisons of utilities that are also shaped by variables affecting consumption decisions such as farm and farmer attributes x. This non-separability of production and consumption decisions is usually couched on assumptions of missing markets. When markets are missing, variables that aff ect consumption decisi ons may also affect production decisions (Vakis, Sadoul et, DeJanvry, and Ca fiero 2004). In lieu of missing markets for the externalities associated w ith GM technologies, the farmer in our model may be thought of as internalizing these into his adoption decision. That is, the farmer may see the hypothetical


88 market as an opportunity to expres s his value for the trait as a pr oducer and as a consumer of GM crops. Alternatively, a different interpretation of E quation 4-4 takes farm and farmer attributes as proxies for expected changes in ut ility from adoption. At any give n point in time, the farmer has adopted the technology (in which case we observe profits under adoption, 1y) or has not adopted the technology (in which case we obs erve profits under no adoption, 0y). The two parallel realities (adoption and no a doption) cannot possibly be observed simultaneously for a given farmer; therefore, the change in profits (01yyy ) is unobservable. The farmer, however, most certainly holds expectations about the change in profits. In the abse nce of a known change in profits the farmer should base his valuation of the technology (i.e., WTP) and his adoption decision on his expected change in profits, y Arguably, the farmer develops this expectation based on what he is currently able to do w ithout the new technology (i .e., current production practices) and what he thinks he could do (given his individual and farm characteristics) if given the chance to try the new technology.11 Accordingly, we may suppose that the farmers expected change in profits from adoption is implicitly ex plained by his individual a nd farm characteristics, and his current production practices. To the best of our knowledge, no other study, as of yet, has considered valuation or adoption of the traits studied he re. There is an extensive literat ure, however, focused on adoption and valuation of GM traits already in the market; namely, herbicide tolerance and insect resistance. This literature offers some guidance as to what farm and farmer attributes may have significant effects on adoption and the possi ble explanations of those effects. 11 Alexander, Fernandez-Cornejo and Goodhue (2003a) write:The decision to adopt a new technology depends on its expected profitability. The expected profitability of an innovation depends on the suitability of the innovation, given its characteristics, for a specific farm er and farm, given their characteristics.


89 A list of the explanatory variabl es used in our models with th eir descriptions is presented in Table 4-4. We work only with complete obser vations. For the FS trait we have a subsample of 175 complete observations to estimate producers WTP for the nonGM version of the trait and a subsample of 155 observations for the GM version of the trait. For the DT trait we have a subsample of 149 complete observations for the n onGM version of the tra it and a subsample of 137 observations for th e GM version. We ran a series of difference in proportion te sts to see if the reductions in sample size affected the randomization of the different CP07 versions. Table 4-5 presents the results of these tests for both traits and both versions of the traits. Calculation of the statistic z is based on differences in sample proportions, where )1,0(~ Nz. No evidence of changes in sample proportions due to the reduction in sample size is observed for any of the subsamples. We also ran a series of t-tests for differen ces in means between each subsample and the base sample of 345 farmers for each of the explanat ory variables. These test s check if individuals in the subsamples differ significantly in observabl e characteristics from individuals in the base sample. The results of these tests and summary sta tistics for each variable are presented in Table 4-6a and Table 4-6b. In all of the four sub samples we find no important differences in means.12 Farm and farmer characteristics The literature identifies farm size, education, age and off-farm em ployment as among those farm and farmer characteristics th at are likely to aff ect adoption decisions. Almost every study of GM crop adoption includes farmers age and edu cation as possible explan atory variables. These factors are considered to repres ent the farmers physical and ma nagerial abilities (i.e., human 12 The statistically significant difference in mean education indicates that if education influences WTP values, then WTP estimates will be biased (Whitehead, Groothuis, and Blomquist 1993). As we shall see, the effect of education is not significant and relatively small in all of our estimated models. In any case, one possible correction for the bias would be to use the population mean instead of the sample mean to calculate the average WTP (Whitehead, Groothuis, and Blomquist 1993).


90 capital). Average age among farm ers in the CP06 is 54 years. As far as education, the average farmer in the CP06 has at least some college instruction. Farm size has been a central focus of many studies of technological adoption (Feder, Just, and Zilberman 1985). The literature offers mixed results. Farm size was the main focus of a USDA study of GM crop adoption (FernandezCornejo and McBride 2002) in which the direction of the estimated eff ects suggest a positive but decre asing influence of farm size on adoption (not all estimates were found to be signif icant at the 0.1 level). Positive effects are seen as evidence of scale dependency of the technolo gy. Hubbell, Marra, and Carlson (2000) find a negative and not statistically signifi cant effect of farm size (i.e., cotton acres) for a subsample of Bt cotton non-adopters and a positive significant effect when the full sample is considered. Fernandez-Cornejo, Hendricks and Mishra (2005) find a negative a nd significant effect of farm size on adoption of herbicide tolerant soybean among U.S. farmers. Other studies find no significant effect of farm size on adoption of GM crops (e.g., Alexander and Von Mellor 2005). Daberkow and McBride (2003) use income from farm sales as a measure of farm size. However, farm size and farm income need not be correl ated. Alexander, Fernandez-Cornejo, and Goodhue (2003b) include both farm size and farm income as possible explanatory variables of farmer acreage allocation between GM a nd nonGM varieties. They find farm income has a significant and positive effect while the eff ect of farm size is negative and not statistically significant. We follow this latter approach and include both m easures. The correlation between farm size and farm income is only 0.11 in our base sample of 345 famers. Farms in the CP06 vary largely in size with an average size of 338 acres and a standard deviation of 565 acres. The average farmer in the CP06 has income somewhere between $60,000 and $79,999.


91 Facilitated in part by the adop tion of labor saving technologie s, off-farm employment and off-farm income have grown steadily in importance over the past decades in U.S. farming (Mishra et al. 2002). Fernandez-Co rnejo, Hendricks, and Mishra (2005) find significant positive correlation between off-farm household income and adoption of herbicide tolerant (Ht) varieties for U.S. soybean farmers. The observed positive corre lation is believed to be consistent with the notion that ease-of-use and management-time save d by Ht varieties allows farmers to increase their off-farm activities and income. However, such positive effect is not guaranteed. The argument may be tuned on its head where, de pending on its degree of importance, off-farm employment and income may take the farmers focus away from the farm (i.e., hobby farming). Off-farm employment may constrai n adoption if it competes with on-farm managerial time. In a different study, Payne, Fernandez-Cornejo and Daberkow (2002) find a negative correlation between off-farm job hours and a doption decisions and attribute this finding to the possibility that off-farm occupied farmers may be less informed about new GM technologies. In our estimations we include a binary indicator =1 if th e farmer reported having an off-farm job and =0 otherwise. A large proportion of farmers in the CP06 reported having some kind of off-farm employment (42%). Production data Production data is also included in our m odel. One important factor determining WTP for a FS trait should be current fertilizer costs. The larg er a farmers fertilizatio n costs are (per acre), the more he should value a trait that reduces such costs. Qaim and DeJanvry (2003) apply a similar reasoning to the case of Bt cotton in Arge ntina and find that higher current insecticide costs increase the farmers WTP for the Bt seed.13 13 Fertilizer costs ($52/acre) represent the largest expense in the CP07 sample followed by seed costs ($46/acre).


92 As noted by Qaim and DeJanvry (2003), the dire ction of the effect of each input cost on WTP will depend on whether the farmer considers th e input as a substitute or a complement for the trait in case. Estimate values smaller than one in absolute value are interpreted as the farmer seeing the trait and the input as imperfect substitu tes (or complements). We also consider yield per acre; average yield in the CP07 was 149 bushels per acre. Purpose of production An indicator for purpose of production is in cluded in the FS trait m odels. Different purposes of production may entail different fertiliz er requirements. For example, a good field of corn silage can yield 20-25 tons of wet fo rage per acre. A 20-ton yield will remove approximately 150 pounds of nitrogen per acre. In comparison, a 100-bushel corn crop will only remove 100 pounds nitrogen (Bates 2009). This difference derives from two agronomic facts: (i) Nitrogen is used by the plant for vegetative grow th, that is, for production of biomass, and (ii) when producing for silage the purpose is to bu ild up biomass whereas producing for grain entails maximizing grain production (i .e., reproductive growth). The CP07 included a section where farmers were asked about the acreag e they destined to different purposes: (i) grain, (ii) silage, (iii) sweetcorn, and (iv) other. We used this data on acreage to construct binary choice indicators that classify farmers into the following categories: grain, silage, sweet corn, other, or diversified (i.e., in purpose). Fa rmers were first assigned into a given category, for example grain, if they dedicated the largest acreage (>50%) to that purpose compared to other purposes. Some of these farmer s were found to plant the same acreage for two purposes (i.e., 50% grain, 50% silage), these farmer s were reclassified as diversified. A similar criteria was used to also r eclassify the following farmers as di versified: (i) produced for two purposes and had at least 40% acreage on each pur pose; (ii) produced for three purposes and had at least 30% dedicated to each purpose; and (iii) produced for four purposes and had at least 20%


93 dedicated to each purpose. A high proportion of farmers (76%) in the original CP06 survey were specialized in grain with the next largest category being silage (10%)14. For the sake of parsimony, in our estimations we group farmers in to only two groups creati ng an indicator =1 if corn is grown mainly for grain purposes, or =0 if corn is grown for other purposes (other=silage+sweetcorn+other). Early adopters and familiarity Several studies of adoption have found perceive d risk and lack of in for mation about a new technology to be important barriers to adop tion (e.g., Dufflo, Kremer, and Robinson 2006; Conley and Udry 2007). Understandably, farmers who are familiar with a similar technology will be more likely to adopt the new technology. Pa yne, Fernandez-Cornejo, and Daberkow (2002) find that likelihood of adoption of corn rootworm resistant GM corn varieties (Bt-CRW) among U.S. corn farmers is significantly and largel y increased by prior experience with GM corn varieties resistant to Eu ropean corn borer (Bt-ECB) measured in a binary dummy variable. We adopt a similar dummy that tells us whether th e farmer was an early adopter of a recently introduced corn variety: Bt-CRW corn. Our measure of familiarity differs from the one used by Payne, Fernandez-Cornejo, and Daberkow (2002) in that it identifies familiarity and also early adoption of a recent GM seed technology. Early adopters, in marketing theory, are thought as having attributes different from other types of a dopters (e.g., more education, less risk averse, more exposure and better access to informati on ) (Rogers 2005). They are usually a small subgroup of the population (Sunding and Zilberman 2001). Our binary indi cator takes a value =1 if the farmer adopted Bt-CRW corn varieties in their first year in the market; and takes a 14 Not reported in Table 4-6a.


94 value =0 otherwise.15 According to this measure, 4.7% of the farmers in the CP07 sample are considered early adopters. This indicator ma y also be interpreted as a measure of habit formation. NonGM farmers One im portant farm characteristic that has been overlooked by the lite rature is the current seed type being planted. Almost every single study of adoption recognizes that the farmer bases his adoption decision on comparisons between th e new technology and his current technology. A farmers expected yield (profits) should be expected to depend on the characteristics of his current seed technology. Most ce rtainly, in any given sample of farmers, there will be heterogeneity in terms of the seed type currently being used. In orde r to capture the effects of this heterogeneity, we include an indicator of whet her the farmer is currently producing only nonGM varieties. Under the non-separab ility scenario, this measure could help capture differences in attitudes towards GM crops betw een nonGM specialists and other t ypes of farmers. A total of 24% of the farmers in the CP07 sample produced exclusively nonGM varieties. Drought measures Probability of severe drought is hypothesized to have an effect on farm ers WTP for a DT trait. We used historic weekly data from th e Drought Monitor Index ar chives (National Drought Mitigation Center 2006) to calculate a measure of the probability of severe drought in each of the counties in the two states (Minnesota and Wisconsin) for 2006. The Drought Monitor Index is produced by a partnership consisting of the U.S. Department of Agriculture (Joint Agricultural Weather Facility and National Water and Clim ate Center), the National Weather Service's Climate Prediction Center, National Climatic Da ta Center, and the National Drought Mitigation 15 An alternative measure was also used which included not only the first but also farmers that adopted in the second year. No significant changes in the results were observed.


95 Center at the University of Nebraska Lincoln. Advice and info rmation from many other sources is incorporated in the index, including virtually every governme nt agency dealing with drought. The Drought Monitor Index identifies four (D1, D2, D3 and D4) different types of drought areas by intensity. D1, D2, D3 and D4 indicate moderate, severe, extreme and exceptional drought, respectively. D0 are drought watch areas that are either heading for drought or are recovering from drought but not yet back to norm al. The four dr ought categories are based on six key indicators and numerous supplementary i ndicators. Table 4-7 shows the definitions for each of the drought categories and the different drought indicators used to construct them. Since th e ranges of the various indicators of ten don't coincide, the final drought category tends to be based on what the majority of the indicators s how. The Drought Monitor reports the countys percentage area under each drought category. The CV questionnaire asked the producer to consider a trai t that produced 75% of its normal yield under severe drought; therefore, we base our measure on D2. The data reported by the Drought Monitor for D2 may be summarized by ctD 2 where c indicates county and t indicates time in weeks. Fame rs in the CP06 and CP07 were ge ographically identified at the county level. The exact measure we use is the county yearly average .2cD ; this measure is included in the model as a proxy for the farme rs subjective probability of drought. Figure 4-2 shows a graphical summar y of our drought measure .2cD Insurance costs For the DT m odels we include insurance costs as a factor hypothesized to affect WTP for the trait. Data on insurance costs in dollars per acre were collected as part of the CP07. The average cost paid by farmers in our base sam ple of 345 farmers is $10.86 per acre.


96 Results Fertilizer saving trait The estim ation results for the two versions (nonGM and GM) of the FS trait are presented in Table 4-8. The likelihood ratio test16 of the global null hypothesis that all coefficients are equal to zero is strongly rejected at the 1% level in both models. A Likelihood Ratio Index17 of 0.62 and 0.55 for the nonGM and GM versions, r espectively; seems to indicate good explanatory power.18 A higher farm income ( f_income) results in higher WTP fo r both the nonGM and GM FS traits. In contrast, farm size ( farmsize) as measured in acres has no effect. We take this as evidence that the technology has no scale depend ency per se. This seems to make sense because a seed input is perfectly divisibl e. The positive effect of farm income suggests that scale dependency may be more related to cash flow constraints. Farmers specializing in corn for grain were found to have a statisti cally significant lower WTP (-$5.12 per acre) than farmers producing fo r other purposes in the nonGM case. This goes in agreement with what was hypot hesized earlier, that is, that producing for grain demands less 16 The likelihood ratio test is a test of significance that compares a restricted versus an unrestricted model. The null hypothesis is 0...21k oH The likelihood ratio test statistic is given by| |2 LLrLLur LR, and is distributed chi-square with degrees of freedom equal to the number of restrictions. 17 The Likelihood Ratio Index (also known as pseudo-2 R or McFaddens 2 R ) is defined as: )0( ) ()0( )0( ) ( 1 ll llll ll ll LRI A perfect probabilistic model would have a log-likelihood equa l to zero. A pure chance (intercept) model is obtained by evaluating the models likelihood function at 0 The likelihood ratio test is essentially a measure of how far we have moved from a pure chance model towards the perf ect model. The likelihood ratio index takes values from 0 to 1. In our case, all parameters in except (which is set to one to avoid division by zero) are set to zero for the pure chance model. 18 Hubbell, Marra, and Carlson (2000) and Qaim and DeJanvry (2003) report similar values.


97 fertilizer than other purposes which in turn tra duces into a lower WTP for a FS trait. As for the GM trait, we still find a negative effect as expected, but the effect is not stat istically significant. Alternatively, one could hypothesize that this negative effects fo llow because, as opposed to corn for silage or other purposes, gr ain production will most likely end in the consumer table and so farmers may be willing to pay less due to risk concerns associated with consumer preferences. However, if this were true, we would have found exactly the opposite a significant effect for the GM trait and no effect for the nonGM. Early adopters ( earlyad ) are in average will ing to pay more than non-early adopters for both the nonGM and GM traits; $9.61 per acre and $10.94 per acre, respectively. The measure used to identify early adopters, as constructed, was also intended to measure the effect on WTP and adoption of farmers familiarity with GM crops. If the effect would have been significant only for the GM trait we could have attributed the effect to familiarity only. That an effect is observed for both traits seems to indicate that our measure and its influence are also related to farmers risk attitudes toward s new technologies in general. As hypothesized, higher fertilizer costs increase farmers WTP. A positive effect is found for both types of traits. In average, farmers are willing to pay $0.06 pe r acre (nonGM) and $0.05 per acre (GM) more for the trait for each extra do llar they spend in fertili zation costs per acre. As suggested by Qaim and DeJanvry (2 003), that this effect is less th an one indicates the trait is not a perfect substitute for the conventional input, in this case fertilizer. Qaim and DeJanvry (2003) explain this for Bt traits by stat ing that one possible reason for th e imperfect substitution is that the trait is an ex ante input. That is, the farmer purchases the trait before the pest pressure is known; in contrast, pesticide provides more flexib ility in that it may be purchased at a later stage


98 in the production process according to need.19 In our case, a similar reasoning may not apply so directly because fertilizer requirem ents and nutrients in soil are much more predictable than pest pressure. However, if fuel and therefore N-fertili zer prices fluctuate larg ely during the year, the ex ante argument may be justifiable. An alternative explanation for the imperfect substitution is that farmers may need to see to believe. If the traits were a proven technology, the farmer might consider them as closer substitutes. A stat istically significant higher WTP for both traits was also found for farmers with higher seed costs. This suggests that fa rmers consider the seed with the new trait to be a subs titute of thei r current seed. Farmers who currently plant only nonGM varieties are, in average, willing to pay $7.67 less per acre for the GM trait compared to farmers who do not specialize in nonGM varieties. This negative effect on the WTP of nonGM farmers is strongly si gnificant and large for the GM trait but small and not significan t in the nonGM trait case. This contrast may be interpreted in several different ways. One interpretation is that, if the non-separability assumption holds, farmers may have seen the hypothetical market pr esented in CV exercise as an opportunity to express their own personal attitudes towards GM crops. Another interpretation could be that farmers could be penalizing the GM trait becau se they consider it a riskier investment. 20 The mean WTP for each trait is calculated by a pplying the estimated model to the mean of the sample data. The standard error for this lin ear function of the estim ated parameters was calculated using the Delta Method. Both of the estimated mean WTPs are strongly significant 19 We discuss this in detail in Chapter 6. 20 Farmers in our base sample were asked a series of a ttitudinal questions regarding possible barriers to Bt seed adoption. Only 8% of the 130 farmers who completed this part of the questionnaire expressed concerns about possible trouble in selling while 2% expressed concerns about getting a lower price compared to nonGM varieties. On the other hand, 16% of the farmers identified concerns about possible environmental and safety issues as an important barrier to adoption.


99 (at 1% level). We find that in average, farmers in the sample were willing to pay around $19.72 per acre for the nonGM trait and $17. 25 per acre for the GM trait. The empirical cumulative distribution for the predicted individual WTP values (iPTW ) is given by ) Pr()(i nonGM i i nonGMPPTWPF and ) Pr()(i GM i iGMPPTWPF for the nonGM and GM traits, respectively. Anothe r way to present these distributions is in the form of adoption potential curves. The adoption curve is simply a graphic representa tion of the predicted proportion of adopters in the sam ple at each price level (i.e.,)(1) Pr(i iiPFPPTW ). The adoption potential curve for both of the FS traits (GM and nonG M) is presented in Figure 4-3. Apparently, the adoption potential for the nonGM trait stochastically do minates (first order dominance) that of the GM trait (i.e., ii nonGM iGMPPFPF )()(). 21 This suggests that the nonGM trait has a better adoption potential. A non-parametric Kolmogorov-Smirnov (KS) test was used to test for differen ces between the two distributions. The KS test uses the maximum vertical distance D between two empirical cumulative distribut ions as the test statistic. The test gave a calculated D value of 0.2161 which implies strong evidence of difference between the distributions. One interesting point is that at prices above $20 pe r acre the adopti on potential for both traits is similar (Figure 4-3). Moreover, there is a shar p increase in the difference of adoption potentials at the $20 per acre mark. Fi nally, as prices drop belo w this mark, the two adoption curves move in parallel fashion. The shapes of these two adoption curves may be indication that different market segments would be captured at prices below or above the $20 per acre price. 21 Stochastic dominance is a concept used to rank distributions. First order stochastic dominance implies that one distribution unambiguously dominates the other one because it deposits the bulk of it probability mass at higher values of the random variable. It is assumed that higher values are better than lower ones.

PAGE 100

100 Drought tolerance trait Table 4-7 sh ows the results for the DT traits. The choice of variables is a little different compared to the FS trait models. The main diffe rence is the exclusion of production costs and yield and the inclusion of insurance costs a nd a measure of probability of severe drought. Including both insurance and production costs may result in collinearity problems. Our choice between these emphasizes the poin t that soil nutrients and fer tilizer requirements are less variable, or at least more predic table, than rainfall and irrigation costs. An important motive for farmers adopting the DT trait could be to insure against unpredictable rainfall; thus the trait could act as a substitute for insurance costs. The likelihood ratio test of the global null hypothesis that all coefficien ts are equal to zero is strongly rejected at the 1% level in both models. A Likelihood Ratio Index of 0.46 and 0.49 for the nonGM and GM versions, respectively; seems to indicate good explanatory power. The results for the earlyadopt and the nonGM100p variables are similar to what was found in the FS trait models so the following discussion focuses on the other significant effects. Farmers seem to take the DT trait as a substi tute for insurance costs. For every current extra dollar spent on insurance pe r acre, the average farmer in our sample is willing to pay 24 cents and 19 cents more for the nonGM and GM DT trait, respectively. Premium rates and indemnity payments of several crop insurance progr ams are tied to individua l-specific yield (e.g. Actual Production History (APH ), Crop Revenue Coverage (CRC)). The improved consistency in yields that could follow from adopting the tr ait should signify lower premium rates. Federal policy today recognizes this for cu rrently marketed GM seeds. In 2008, the Federal Crop Insurance Corporat ion (FCIC) started a pilot program under the name of Biotech Yield Endorsement (BYE). Under the BYE, farmers in the states of Illinois, Indiana, Iowa, and Minnesota pr oducing non-irrigated corn for grai n with at least 75 percent of

PAGE 101

101 their corn acres planted to a specifi c triple stacked biotech variety are eligible for a premium rate reduction in their yield or revenue individual insurance policies (U.S. De partment of Agriculture 2008a). In August 2008 the FC IC approved an expanded Biotechnology Endorsement (BE) that replace s the BYE for the crop years 2009-2011 (U.S. Department of Agriculture 2008b). The expanded BE includes additional seed technologies and additional states. The program is based on proved increases in consistency of yields provided by qualifying GM varieties compared to conventional ones. By planting these va rieties farmers reduce expected losses which in turn reduce the number of insurance claims. The key point underpinning the rate reduction, as the significance of the insurance substitution effect in both types of traits suggests, is the difference in cons istency of higher yields and not whether the trait is conventional or GM. To the best of our knowledge, no public study evaluating the effects of the BE on adoption of qualifying GM seeds has been conducted yet. However, it seems reasonable to expect that it has had a positive e ffect on adoption. If this is the case, seed companies selling GM varieties should find it highly valuable to cont inue investing in good research that provides proven differences in yield consistency. We find that a higher probability of severe drought increases the average farmers WTP only in the GM case. The effect is not significant for the nonG M case. This contrast seems a bit puzzling. We find that if the models are estimat ed for a subsample of insured farmers only, the effect results significant in both traits (Appendix B). The Drought Monitor seems to be a good proxy for farmers subjective probabilities in this case. One possible explanation could be that pooling both groups in a single est imation may be obscuring the e ffect of drought probability on WTP making it not significant.

PAGE 102

102 The mean WTP for the nonGM trai t is estimated at $20.87 per acre. For the GM trait the mean WTP is estimated at $18.73 per acre. Both estimates are significant at the 1% level. The adoption potential curve for the two DT traits is presented in Figure 4-4. No first order stochastic dominance is observed in this case, thus it ca nnot be said which trait version unambiguously has a better adoption potential. However, the KS D-statistic was calculate d at 0.2085, showing a significant difference (p-value=0.003) between th e adoption potential curves. According to the shapes of the adoption curves, the GM trait ho lds better adoption potential at high prices compared to the nonGM trait. A final point is that similar to what we find in the FS traits, the two curves separate from each other at the $20 per acre mark. This seems to reinforce the different market segment argument.

PAGE 103

103 Table 4-1. Traits consid ered for farmer valuation Trait Version 1 Fertilizer saving nonGM 2 Fertilizer saving GM 3 Drought tolerance nonGM 4 Drought tolerance GM Table 4-2. Seed codes and variety descriptions Seed code Description 1 Genetically-modified (GM) herbicide resistant corn (e.g., Roundup Ready, LibertyLink) 2 Non-GM herbicide resistant seed variety (e.g., IMI-corn) 3 GM Bt variety for insect resistance to cont rol European Corn Borer (Bt-ECB) (e.g., YieldGard, NatureGard, Knockout, Herculex) 4 GM Bt variety for insect resistance to control corn rootworm (Bt-CRW) (e.g. YieldGard Rootworm, Herculex RW) 5 Stacked gene variety with both GM Bt-ECB and Bt-CRW(e.g. YieldGard Plus, Herculex Xtra) 6 Stacked gene variety with both GM Bt-ECB and herbicide resistant (e.g. YieldGard or Herculex +Roundup Ready) 7 Stacked gene variety with both GM Bt-CRW and herbicide resistant (e.g. YieldGard Rootworm or Herculex RW + Roundup Ready) 8 Triple stacked gene variety with GM Bt -ECB and Bt-CRW plus herbicide resistant (e.g. YieldGard Plus or Herculex Xtra+ Roundup Ready) 9 None of the above Table 4-3. CV survey versions with first and second offers Initial offer Follow-up offer Version iP L iP H iP A $10 $5 $15 B $15 $10 $20 C $20 $15 $25 D $25 $20 $30

PAGE 104

104 Table 4-4. Description of variables selected from 2006 corn poll and 2007 corn poll Farmer characteristics f_age Age (in years) f_educ Education level (less than high school=1, high school diploma=2, some college=3, completed 2-year degree college=4, completed 4-year degree college=5, and some graduate school or graduate degree=6) f_nfjob Off-farm job (=1 if farmer has a full or part time off-farm job) f_income Gross income received from all farming activities in 2005 (Under $20,000 =1, from $20,000 to $39,999=2, from $40,000 to $59,999=3, from $60,000 to $79,999=4, from $80,000 to $99,999=5, from $100,000 to $119-999=6, from $120,000 to $139,999=7, from $140,000 to $159,999=8, $160,000 or more=9) earlyadopt Past adoption practices with respect to GM corn seeds (habit formation) (=1 if farmer adopted Bt-crw varieties in their first year in the market) nonGM100p Current use of GM crops (=1 if farm production is 100 percen t nonGM varieties, =0 otherwise) Farm characteristics farmsize Total farm size reported in 2005 (in acres) yield Corn yield reported for most used seed code in 2006 (bushels per acre) seedcost Seed costs reported for most used seed code in 2006 ($ per acre) herbcost Herbicide costs reported for most used seed code in 2006 ($ per acre) insectcost Insecticide costs reported for most used seed code in 2006 ($ per acre) fertcost Fertilizer costs reported for most used seed code in 2006 ($ per acre) insurcost Corn insurance costs reported by farm in 2006 ($ per acre) grain Purpose of corn production (=1 if corn produced for grain, =0 if produced for other purposes (i.e., silage, sweet corn, other) mD2 Probability of drought (average percentage area under sever e drought in farms county in 2006)

PAGE 105

105 Table 4-5. Proportion test s and number of farmers by version of Contingent Valua tion (CV) survey 2007 Corn Poll Subsample nonGM Subsample Version (n=451) GM n Proportion n Proportion Diff zi n Proportion Diff z (1)ii (2) (1)-(2) (3) (1)-(3) Fertilizer Saving A 85 0.1885 30 0.1714 0.0171 0.51 26 0.1677 0.0208 0.59 B 121 0.2683 41 0.2349 0.0340 0.90 35 0.2258 0.0425 1.09 C 128 0.2838 52 0.2971 -0.0133 -0.33 45 0.2903 -0.0065 -0.15 D 117 0.2594 52 0.2971 -0.0377 -0.93 49 0.3161 -0.0567 -1.30 Total 451 1.0000 175 1.0000 155 1.0000 Drought Tolerance A 85 0.1885 28 0.1879 0.0006 0.02 26 0.1898 -0.0013 -0.03 B 121 0.2683 36 0.2416 0.0267 0.67 31 0.2263 0.0420 1.03 C 128 0.2838 45 0.3020 -0.0182 -0.42 41 0.2993 -0.0155 -0.34 D 117 0.2594 40 0.2686 -0.0091 -0.22 39 0.2847 -0.0253 -0.57 Total 451 1.0000 149 1.0000 137 1.0000 iThe calculated statistic is )/) 1( ()/) 1( () (21 2 12 1 21nppnppppz where )1,0(~ Nz. iiThe numbers in parenthesis (e.g., (1)) indicate the columns being compared. iiiA value of 65.1 z indicates a significance level of 0.1, a value of 96.1 z indicates a significance level of 0.05.

PAGE 106

106 Table 4-6a. Summary statistics and compar ison of means (fertilizer saving trait) 2006 Corn Poll n=945 2007 Corn Poll n=451 Subsample nonGM n=175 Subsample GM n=155 n Mean s.d. n Mean s.d. Mean s.d. Diff i t ii Mean s.d. Diff i t ii f_age 911 54.20 12.51 53.01 11.11 1.19 1.27 52.83 10.90 1.37 1.41 f_sex 919 1.02 0.13 f_educ 918 2.87 1.36 3.27 1.47 -0.40** -3.34 3.33 1.48 -0.46** -3.62 f_nfjob 910 0.42 0.49 0.39 0.49 0.03 0.74 0.39 0.49 0.03 0.77 f_income 832 3.75 2.17 3.93 2.22 -0.18 -0.98 3.87 2.21 -0.12 -0.63 grain 833 0.76 0.42 0.79 0.41 -0.03 -0.88 0.80 0.40 -0.04 -1.13 earlyadopt 833 0.05 0.21 0.05 0.22 0.00 0.00 0.05 0.21 0.00 0.27 yield 300 149.28 46.62 143.90 43.91 5.38 1.26 143.12 44.26 6.16 1.38 seedcost 310 46.00 22.85 46.07 22.35 -0.07 -0.03 46.51 21.32 -0.51 -0.24 herbcost 310 21.72 12.62 20.08 11.97 1.64 1.42 19.86 12.11 1.86 1.54 insectcost 310 3.32 7.06 3.14 6.90 0.18 0.27 2.88 6.52 0.44 0.67 fertcost 310 52.92 35.08 50.25 33.29 2.67 0.83 51.36 32.72 1.56 0.47 farmsize 898 338.40 565.22 388.46 446.49 -50.06 -1.29 400.54 439.29 -62.14 -1.55 nonGM100p 429 0.24 0.43 0.23 0.42 0.01 0.26 0.20 0.40 0.04 1.04 i The column diff presents calculated differences between subsample means and their respective original sample means. ii The t statistics are calculated for two tailed tests assuming different variances. iii Degrees of freedom are sufficiently large in all mean comparisons to allow for use of critical values from the standard normal distribution (** indicates significance level of 0.05).

PAGE 107

107 Table 4-6b. Summary statis tics and comparison of means (drought tolerance trait) 2006 Corn Poll n=945 2007 Corn Poll n=451 Subsample nonGM n=149 Subsample GM n=137 n Mean s.d n Mean s.d Mean s.d Diff i t ii Mean s.d Diff i t ii f_age 911 54.20 12.51 53.07 10.92 1.13 1.15 52.82 10.62 1.37 1.37 f_sex 919 1.02 0.13 f_educ 918 2.87 1.36 3.19 1.41 -0.32** -2.59 3.25 1.44 -0.38** -2.92 f_nfjob 910 0.42 0.49 0.39 0.49 0.03 0.63 0.39 0.49 0.03 0.66 f_income 832 3.75 2.17 3.93 2.24 -0.18 -0.91 4.02 2.32 -0.28 -1.30 grain 833 0.76 0.42 0.79 0.41 -0.02 -0.56 0.79 0.41 -0.02 -0.62 earlyadopt 833 0.05 0.21 0.06 0.24 -0.01 -0.65 0.06 0.24 -0.01 -0.54 yield 300 149.28 42.62 142.94 43.28 6.34 1.47 143.32 43.65 5.96 1.33 seedcost 310 46.00 22.85 43.96 21.80 2.03 0.92 44.20 20.14 1.80 0.83 herbcost 310 21.72 12.62 20.07 12.02 1.65 1.35 19.88 12.09 1.84 1.46 insectcost 310 3.32 7.06 2.92 6.61 0.40 0.59 2.82 6.50 0.50 0.72 fertcost 310 52.92 35.08 48.58 28.08 4.35 1.43 49.31 26.52 3.62 1.20 farmsize 898 338.40 565.22 375.14 427.56 -36.73 -0.92 392.79 444.90 -54.39 -1.28 nonGM100p 429 0.24 0.43 0.25 0.43 -0.01 -0.14 0.23 0.42 0.02 0.39 insurcost 394 10.86 18.23 9.35 15.09 1.51 0.98 9.71 15.53 1.15 0.71 i The column diff presents calculated differences between subsample means and their respective original sample means. ii The t statistics are calculated for two tailed tests assuming different variances. iii Degrees of freedom are sufficiently large in all mean comparisons to allow for use of critical values from the standard normal distribution (**indicates significance level of 0.05).

PAGE 108

108 Table 4-7. Definiti ons of drought measures Ranges Category Description Possible impacts Palmer Drought Index CPC Soil Moisture Model (Percentiles) USGS Weekly Streamflow (Percentiles) Standardized Precipitation Index (SPI) Satellite Vegetation Health Index D0 Abnormally Dry Going into drought: short-term dryness slowing planting, growth of crops or pastures; fire risk above average. Coming out of drought: some lingering water deficits; pastures or crops not fully recovered. -1.0 to -1.9 21-30 21-30 -0.5 to -0.7 36-45 D1 Moderate Drought Some damage to crops, pastures; fire risk high; streams, reservoirs, or wells low, some water shortages developing or imminent, voluntary water use restrictions requested -2.0 to -2.9 11-20 11-20 -0.8 to -1.2 26-35 D2 Severe Drought Crop or pasture losses likely; fire risk very high; water shortages common; water restrictions imposed -3.0 to -3.9 6-10 6-10 -1.3 to -1.5 16-25 D3 Extreme Drought Major crop/pasture losses; extreme fire danger; widespread water shortages or restrictions -4.0 to -4.9 3-5 3-5 -1.6 to -1.9 6-15 D4 Exceptional Drought Exceptional and widespread crop/pasture losses; exceptional fire risk; shortages of water in reservoirs, streams, and wells, creating water emergencies -5.0 or less 0-2 0-2 -2.0 or less 1-5 Source: Drought Monitor website available at http://drought.unl.edu/dm/classify.htm

PAGE 109

109 Table 4-8. Estimation results for fertilizer saving trait Fertilizer nonGM Fertilizer GM Variable Estimate p-value Estimate p-value constant 12.035** (0.031) 2.733 (0.684) f_age 0.029 (0.698) 0.096 (0.260) f_educ 0.569 (0.324) 0.740 (0.267) f_nfjob -0.267 (0.879) -1.734 (0.394) f_income 0.630* (0.079) 0.868** (0.030) grain -5.129** (0.013) -2.534 (0.300) earlyadopt 9.617** (0.017) 10.994** (0.013) yield -0.016 (0.419) 0.003 (0.901) seedcost 0.086** (0.031) 0.075* (0.096) herbcost 0.074 (0.266) 0.090 (0.245) insectcost -0.212* (0.078) -0.027 (0.853) fertcost 0.067*** (0.007) 0.051* (0.078) farmsize -0.001 (0.429) -0.002 (0.315) nonGM100p 0.572 (0.776) -7.678*** (0.002) sigma 8.768*** (0.000) 9.360*** (0.000) N 175 155 Log-likelihood -223.931 -193.012 Mean WTP i 19.723*** (25.999) 17.248*** (19.814) C.I. 95 L 18.236 15.542 C.I. 95 U 21.210 18.954 LR Index 0.62 0.55 LR statistic 198.501*** (0.000) 131.608*** (0.000) i For Mean WTP the t-statistic is reported in parenthesis. Standard error for mean WTP was calculated using the Delta Method. *** indicates significance at 0.01 level, ** indicates significance level of 0.05, and indicates significance level of 0.1. LR statistic is for likelihood ratio test with 0... :21 oH, sigma is left unconstrained.

PAGE 110

110 Table 4-9. Estimation results for drought tolerance trait Drought nonGM Drought GM Variable Estimate p-value Estimate p-value constant 25.291*** (0.002) 7.543 (0.321) f_age -0.149 (0.201) 0.147 (0.191) f_educ -0.736 (0.462) -0.937 (0.331) f_nfjob 0.951 (0.748) 3.524 (0.222) f_income 0.847 (0.146) 1.075** (0.045) earlyadopt 12.842** (0.021) 14.967*** (0.007) farmsize -0.006 (0.125) -0.003 (0.359) nonGM100p 3.085 (0.321) -6.638** (0.031) insurcost 0.243** (0.039) 0.189* (0.072) mD2 0.127 (0.348) 0.210* (0.092) sigma 12.868*** (0.000) 11.527*** (0.000) N 149 137 Log-likelihood -186.997 -163.210 Mean WTP i 20.873*** (16.852) 18.738*** (16.154) C.I. 95 L 18.445 16.465 C.I .95 U 23.301 21.012 LR Index 0.46 0.49 LR statistic 107.609*** (0.000) 99.072*** (0.000) i For mean WTP the t-statistic is reported in parent hesis. Standard error for mean WTP was calculated using the Delta Method. *** indicates significance at 0.01 level, ** indicates significance level of 0.05, and indicates significance level of 0.1. LR statistic is for likelihood ratio test with 0... :21 oH, sigma is left unconstrained.

PAGE 111

111 7) Consider a new corn seed techno logy that comes along that mainta ins your same yield but reduces your nitrogen fertilizer requirements by one-third (33%). Would you pay $10 extra per acre for this seed if it did not involve genetic modification? Yes (check one) No If yes, would you buy it if it cost $15 extra per acre? YES NO If no, would you buy it if it cost $5 extra per acre? YES NO if it did involve genetic modification? Yes (check one) No If yes, would you buy it if it cost $15 extra per acre? YES NO If no, would you buy it if it cost $5 extra per acre? YES NO 8) Now consider a new corn seed technology that is drought tolerant so that under severe drought conditions it would still yield 75% of your normal yield. Would you pay $10 extra per acre for this seed if it did not involve genetic modification? Yes (check one) No If yes, would you buy it if it cost $15 extra per acre? YES NO If no, would you buy it if it cost $5 extra per acre? YES NO if it did involve genetic modification? Yes (check one) No If yes, would you buy it if it cost $15 extra per acre? YES NO If no, would you buy it if it cost $5 extra per acre? YES NO Figure 4-1. Continge nt Valuation (CV) questionnaire portion of the 2007 Corn Poll. Figure 4-2. Map of average percentage area affected by severe drought (D2) for counties in Minnesota and Wisconsin, 2006. (Data S ource: Drought Monitor Archives 2006).

PAGE 112

112 Figure 4-3. Potential adoption curve fo r nonGM and GM fertilizer saving trait. Figure 4-4. Potential ad option curve for nonGM and GM drought tolerance trait.

PAGE 113

113 CHAPTER 5 TESTING FOR SAMPLE SELECTION BI AS IN OUR WTP ESTIMATES This chapter explores the possibility that sam ple selection bias ma y be affecting the estimates of farmers willingness-to-pay (WTP) for the genetically modified (GM) traits that we obtained in Chapter 4. The application of a sample selection model was motivated by the observation, in our sample, of a considerab ly larger number of non-respondents to the Contingent Valuation (CV) ques tions concerning the GM versi on of a trait compared to the number of non-respondents to CV questions concerning the nonG M version of the same trait. While high rates of non-response seem to be ubiquit ous in CV research, the contrast in response rates between the GM and nonGM questions appeared unusual. This observation led us to consider the possibility that failure by the farmer to res pond to GM questions could be systematically related to his WTP. If that were the case, it would imply underlying sample selection problems. Testing for sample selection in our results from Chapter 4 is important because if detected it would suggest we have inconsistent estimates for the parameters of the population initially targeted. On the other hand, if no sample selection is detected, it would provide more reliability to our findings. At first instance, it could appear as a simple matter to assume de facto that non-responses to GM questions are correlated to respondents WT P. However, as we shall see, it could be that individuals who fail to respond do so because they are not familiar enough and fail to construct their valuation. Which is culpr it, unfamiliarity or sample selection, is empirically testable and empirically determined. The possibility of sample selection bias has not been considered by prev ious studies of GM seed adoption similar in spirit to this one. The reported non-response rat es in those studies are

PAGE 114

114 varied. For example, Hubbell, Marra, and Carl son (2000), using CV me thods to study farmer adoption of Bt-cotton varieties, report a non-response rate of 62% which is low even for CV studies. On the other hand, Qaim and DeJanvr y (2003) report only a 10% non-response rate in their study of Bt-cotton adoption among farmers in Argentina. Regardless of non-response rates, it would have not been possible fo r those studies to observe a cont rast in behavior since they only considered GM versions of the trait in their CV questionnai re. Observing the contrast in response rate between the two versions (nonGM and GM) of each trait was possible in our study due to the specific design of the survey instrument with nonGM and GM questions being asked side by side. Besides being unable to observe the contrast just mentione d, another possible reason for the unpaid attention to potential sample selection problems in those studies may have been the lack of an appropriate model to test for its pr esence. Application of He ckmans sample selection framework to the case of a double bounded (DB) di chotomous choice (DC) elicitation format has escaped the CV literature for many years. The model used in this chapte r to test for sample selection was developed by Yoo a nd Yang (2001) and has only recently seen some applications in the CV literature (Yoo 2007; Yoo, Lim, and Kwak 2009).1 When we discuss this model later, we will refer to it as the DB-DC Sample Selection model. Non-Response and Sample Selection in CV Non-response is comm on and often large2 in CV surveys (Elkf and Karlsson 1999). This makes CV surveys vulnerable to serious nonresponse bias problems (Mitchell and Carson 1989). Non-response in CV studies can bias results in two ways (Elkf and Karlsson 1999). 1 To the best of our knowledge, no other sample selec tion model has been developed for DB-DC elicitation formats. 2 Response rates for the mail shot were good by CVM standards, being greater than 50% in almost all subsamples. (lvarez-Farizo, Hanley, and Wright 1996), 40-60% [response rate] seems average for general population CVM surveys (Loomis 1987), minimizing both sample non-response and item non-response are

PAGE 115

115 The first case is known as non-response bias3. Obtaining mean WTP estimates is usually one important objective in CV studies. The m ean WTP estimate is usually calculated by estimating a model and then eval uating it at the mean values of the observable characteristics (i.e., the explanatory variab les). If the population mean for each observable characteristic is unknown it is usually estimated using the averages calculated from the sampl e of respondents. If non-respondents differ significantly from respondents in observable ch aracteristics that influence WTP, the mean WTP will be affected by non-res ponse bias. The common way to test for this type of bias is to perform a ttest comparing the means of the characteristics of respondents versus those of non-respondents. If non-response bias is identified, a possible correction is to use the population means instead of the sample means when calculating mean WTP (i.e., when the population means are available). That this type of bias may affect our estimates was explored in Chapter 4 (See Table 4.6a and Table 4.6b). In this chapter we focu s on the second type of potential bias which we explain in the following. The second type of bias is sample selection bi as. Bias may result even if non-respondents and respondents are similar in observed characteri stics but differ systematically in their WTP due to unobservable characteristics (i.e., if non-response is systematically related to WTP). As we explain later in detail, this type of bias may be explained as a correlation between the error term in a selection equation determining participation in the sample and the error term in an outcome equation of interest. If the error terms are correlated, the estimates obtained using only the sample of respondents will be inconsistent estimates for the parameters of the population initially targeted. important. The former is unlikely to be below 20% even in very high quality surveys (Arrow et al. 1993). Response rates on CV mail surveys typically range between 20% and 60%. (Whitehead, Groothuis, and Blomquist 1993). 3 We follow Elkf and Karlssons (1999) terminology.

PAGE 116

116 Sample selection bias in CV studies may ari se from two sources (Edwards and Anderson 1987). First, researchers may unint entionally create an artificial sample selection bias by culling outliers. If the culling criter ion is systematically related to the response variable (WTP) the result is sample selection bias. Second, non-resp ondents may naturally self-select themselves out of the sample (i.e., decide not to respond). In this case, sample selection bias will be present if non-respondents WTP for the good in case is substantially and systematically different from that of respondents. This latter case is the focus of this chapter.4 The remainder of this section is organized as follows: (i) First we present Heckmans (1979) framework and sample selection model in the context of CV methods and discuss the statistical theory underlying it. (ii) Second, we define the scope of Heckmans two-stage procedure as those cases where the outcome equa tion is linearly specif ied and the dependent outcome variable is continuous; and survey th e literature for exte nsions of Heckmans framework to the case of SB and DB elicitation fo rmats. (iii) Finally, we outline the difficulties in gathering the necessary data to apply self-se lection tests and models in CV research, and describe some of the surveying strategies found in th e literature that are used to deal with these difficulties. Heckman Selection Model a nd Two-Stage Procedure It seem s Randall, Hoehn, and Tolley (1981) were the first to suggest the use of Heckmans model in the CV literature. Heckman (1979) deve loped the basic sample selection model in the context of specification error a nd proposed a two stage estimator to correct for the presence of sample selection. In the following we devel op the Heckman model in the context of CV methods. 4 We focus in this type of sample selection because no culling rule was used when estimating the models in Chapter 4.

PAGE 117

117 Consider a sample of size N and the two following behavioral equations: Nni u WTPiii ,...,1, x (5-1) Ni v Siii,...,1, z (5-2) Equation 5-1 is usually called th e output equation, however, in the CV literature it is directly referred to as the WT P equation. Equation 5-2 is ca lled the selection equation (or selection mechanism equation) and determines which individuals give responses to the CV survey (i.e., enter the hypothetical mark et) and which individuals dont. Also, ix is a vector of factors influencing the individuals WTP and the iz is a vector of factors influencing his decision to participate in the CV questionnaire; iS is a binary choice variable (=1) if the individual gives a response and (=0) if he selects out of the sample; and are parameter vectors to be estimated; and iu and ivare zero mean stochastic error terms. Notice that Equation 5-2 is de fined over the whole sample of N individuals. On the other hand, Equation 5-1 is defined only for a subsample of Nn individuals for which1 iS. The subsample of n individuals is usually referre d to as the selected sample. One important point to consider is that sample selection can only be an issue after a population of interest is clea rly specified (Wooldridge 2002). For example, if our target population were only individuals who usually accept contingent markets,5 then the n observations satisfying 1 iS form a representative random sample from the population, and regression fit of Equation 5-1 gives consistent estimates. The sample selection problem possibility arises only when we define the population of interest as the whole population, including those who may usually reject contingent markets. 5 And therefore answer CV surveys.

PAGE 118

118 Heckman showed that sample selection bias occurs when behavioral parameter estimates ( ) in the WTP equation (Equation 5-1) ar e confounded with parameter estimates () appearing in the selection equation (Equati on 5-2). Notice that E quation 5-1 is defined only for the selected sample, thus it is correctly modeled as ]1|[']1,|[ ii i iiiSuE SWTPE x x (5-3) Using Equation 5-2, we can re-write Equation 5-3 as: ]'1|[']1,|[ i ii i iiivuE SWTPEz x x (5-4) From Equation 5-4 it becomes obvious that th e existence of sample selection bias depends on whether iu and iv are independent. If this condition is not satisfie d, using regression fit on Equation 5-3 will omit the final term in Equation 5-4 and the obtained parameter estimates will be inconsistent estimates for the s of the whole population. Thus, the sample selection bias problem can be thought of as an omitted variable problem. In order to obtain selection-corrected estimat es of the parameters, the model presented in Equations 5-1 and 5-2 may be estimated us ing maximum likelihood methods by specifying a joint distribution for (iu,i ) and normalizing the variance of the error in the selection equation to one (i.e., 1)(2iVar). 6 Heckman (1979) devised a simple two step proc edure that can be used to estimate the same model of Equations 5-1 and 5-2 but avoids the complications of using ML estimation. Under the assumption of a joint normal distribution for (iu,i ), the following can be shown for the last term (omitted variable) in Equation 5-4: 6 The respective log-likelihood functio n is presented as Equation [25] in Dubin and Rivers, (1989, pp. 370).

PAGE 119

119 )'()'(]'1|[2 12 ii ii i iivuE z z z (5-5) where (.)i is the inverse Mills ratio and is a parameter to be es timated. Heckman showed a consistent estimator for i is ) '( iz, where the estimate is available from a conventional probit estimation of Equation 5-2 alone. Heckman procedure th en amounts to: (i) using probit estimation of Equation 5-2 alone to obtain then (ii) using to construct ) '( iz, and finally (iii) using ) '( iz as an explanatory variable in regre ssion fit of Equation 5-3 alone to obtain Modeling Sample Selection in Si ngle Bounded and Double Bounded CV Heckman developed his model to correct for se lectivity bias when the outcome equation (i) presents a continuous dependent variable, and (ii) is modeled using a linear specification. In the CV literature obtaining a continuous dependent variable requires using an open ended (OE) elicitation format. In the context of CV methods this means that the Heckman model is correct only for the OE elicitation format. Most of the ea rly CV studies that dealt with sample selection issues used only OE formats (e.g., Edward s and Anderson 1987; Loomis 1987). However, as mentioned in previous chapters, DC (or closed ended) formats are considered superior and recommended over OE formats. Closed ended formats require discrete-choice st atistical models that differ from the linear regression model assumed by H eckman. For example, correct modeling (without sample selection) when assuming normal errors in th e single bounded (SB) case requires a probit model for the outcome equation, while assuming logist ic errors implies a logit. Even though the conceptual framework used by Heckman carries over naturally to the case of probit and logit

PAGE 120

120 outcome equations, the convenience of th e Heckman two-step procedure does not.7 Nevertheless, some theoretical studies have succeeded in ad apting Heckmans framework to deal with probit and logit outcome equations (Dubin and Rivers 1989; van de Ven and van Praag 1981; Meng and Schmidt 1985). CV authors have followed to appl y these models to SB CV data. For example, Whitehead, Groothuis, and Blom quist (1993) use the bivariat e probit (or biprobit) model developed by Dubin and Rivers, while Whitehead et al. (1994) and Ek lf and Karlsson (1999) use the censored probit presented in Meng and Schmidt. Applying a sample selection framework to th e case of DB CV data has proven more complicated. The bivariate probit of Dubin and Rivers, used to deal with sample selection in SB CV research, is correct for cases where the de pendent variables in bot h the selection and the outcome equations are binary and the joint distribution is assumed normal. The dependent variable in the outcome equation for the DB case is not binary as it is in the SB case, so the correct model is not a biprobit. An appropriate model for studyi ng sample selection in the DB case has only recently appeared, introduced by Yoo and Yang (2001). To the best of our knowledge, only two other applicatio ns of this model currently exist (Yoo 2007; Yoo, Lim, and Kwak 2009). This chapter explores the possibility of sample selection in the estimations presented in Chapter 4 using Yoo and Yangs mo del. The DB-DC Sample Selection of Yoo and Yang is developed in detail in a later section in this chapter. The derivation follows closely that used by Dubin and Rivers (1989) to develop their biprobit model. 7 A two-step procedure, analog to Heckmans procedure, exists for the case of an outcome equation with binary choice dependent variable (Dubin and Rivers 1989). However, the computational advantages of the two-step estimator over ML estimation are less in this case compared to Heckmans case with a continuous dependent variable and linear outcome equation.

PAGE 121

121 Issues with Data Availability and Surveying Strategies As m entioned, issues of self-selection are fr equent in CV studies and are well-known to CV researchers. However, tests and correction fo r self-selection bias have been scarce in the literature mainly because data on non-respondents necessary to conduct such tests is usually not available (Yoo and Yang 2001). Earlier we identifie d culling and self-sel ection as two possible causes of sample selection bias in CV. In practice, from a logistics perspective, te sting for sample selection bias in the selfselection case is considerably more difficult th an doing so in the cull ing case (Edwards and Anderson 1987). In the culling case, the researcher initially holds N observations for the variables iWTP, ix and iz. He then uses iS to censor some of these da ta points leaving only a subsample of n valid observations NniWTPii };,{x to estimate Equation 5-1 and has at hand the necessary N observations NiSii };,{z to estimate Equation 5-2. Data availability is considerably more complicated in the ca se of self-selection. Suppose a random sample from a populat ion is designed to have N observations but af ter data collection we observe n responses and nN non-responses. CV studies have usually collected data for ix and iz in a same single survey. Respondents that pa rticipate in the survey will provide data on both ix and iz; similarly, those that self-select out will answer neither. This means the researcher always ends up with n observations on all variables (i.e., Nni WTPiii };,,{ zx). However, in order to test and correct for sample selection, the researcher needs n data points for {iiWTP x ,} and nN data points for {iiS z ,}. In this case, the researcher must push the issue and interview the nN non-respondents for their data on iz.

PAGE 122

122 Those studies in the CV literature that addre ss self-selection issues have adopted different surveying strategies in attempts to obtain the necessary data on non-respondents. Edwards and Anderson (1987) experimented with switching from mail to phone interviews in order to collect information on non-respondents. However, they desi sted due to concerns about the length of the survey and the complications associated with de scribing changes in the good (water quality) via telephone. They were, in the end, unable to test for self-selec tion. Recognizing that concerns about explaining changes in th e good via telephone are vali d, Whitehead, Groothuis, and Blomquist (1993) used a two-stage phone sampling and mail survey strategy. They first used a phone interview to collect socio economic data on 926 people. Phone interviewees were asked if they would complete a mail questionnaire whic h contained the detailed CV scenarios and questions. Out of the 926 phone calls, 641 agreed to complete the questionn aire and 487 actually returned a completed questionnai re. Yoo and Yang (2001) also us ed a two-stage approach but opted for an initial face-to-face inte rview with a follow-up phone interview.8 Trained interviewers visited a sample of randomly selected individuals and asked th em first if they would participate. If they agreed, the interviewer proceed ed with the main interview. If they declined, the interviewer asked only their names and telephone. After a few days, trained enumerators called to collect socioeconomic data on those non-respondents. We used a surveying strategy involving tw o separate face to face interviews. The advantages and disadvantages of our strategy are discussed later in the empirical section of this chapter. 8 They did not use mail surveys due to the unfamiliarity and extremely low-response rates these have in Korea where the study was conducted.

PAGE 123

123 Theoretical Model In this sec tion we present the details of the DB-DC Sample Selection model (Yoo and Yang 2001) that we use to test for the possibility of sample selection bias in the estimates obtained in Chapter 4. If sample selection is detected, the same m odel obtains consistent estimates of the parameters explaining individu al WTP. The section is organized as follows. First, we present a brief initial setup of the two equation Sample Selection model in the context of CV research. Second, we re visit the DB-DC mechan ism (without sample selection) and look at it in a less formal but more intuitive manner th an we did in Chapter 3 and 4. This allows us a better understanding of the DB-D C mechanism and provides the tools to develop the DB-DC Sample Selection model that we use in the em pirical application section of this chapter. Initial Setup W e consider the two following behavioral equations. Let the equation representing the selection mechanism be: i iiy111 1' x (5-6) and the outcome or WTP equation be: i iiy222 2' x (5-7) We do not observe th e latent variable 1 iy, only whether 1 iy is greater than zero or not. Therefore, we observe: )0(1* 1 1i iyy (5-8) where 1(.) is an indicator function equal to 1 if its argument is true (i.e., the person participates in the CV survey) and 0 otherwise (i.e., does not participate). The latent dependent variable 1iy may be understood as a threshold which triggers the individuals decision to participate in the

PAGE 124

124 hypothetical market presented by the CV exerci se; possibly the level of interest of the respondent. Information on each individuals WTP for the technology (* 2 iy) is also limited (i.e., categorical) due to the dichotomous nature of the DB elicitation format we use. In the following we revisit a topic already presented in a Chapter 3 the DB-DC mechanism and reformulate it in a manner that will facilitate the derivation of the DB-DC Sample Selection Model. To simplify exposition, we abstract for a moment from the sample selection issues and work only with the WTP equation (Equation 5-7) which represents the traditional DB-DC model. That is, for a moment we ignore any selec tion mechanism (Equation 5-6). We will come back to incorporate the sel ection equation into our model later when we discuss the DB-DC Sample Selection model. Double Bounded Mechanism Revisited Intuitively, the DB-DC e licitation method resu lts in a classification of individual WTPs into four different regions (or bins) across the range of possible WTP values. These regions are presented in Figure 5-1 and are labeled from 1 to 4. From Chapter 4 we know that ) Pr()Pr(iiPWTP Adopt that is, the respondent will agree to pay a given price iP and adopt the technology if his WTP exceeds that price; ot herwise, he will reject the technology. Observe Figure 5-1: In the case of SB elicitation, a No answer to a p rice iP places a given individuals WTP to the left of iP in the region ),(iP Alternatively, a Yes answer would place his WTP on the right of iP, in the region ),( iP. The follow-up offer in DB elicitation allows for further subdivision of the range of WT P values into four regions: ),(),,(),,(H iii L i L iPPPPP and ),( H iP. The mechanism of DB-DC elicitation ma y be better understood with an example. Suppose an individual responds No to the original offered price iP. This response places his

PAGE 125

125 WTP somewhere on the left of iP in Figure 5-1 (i.e., region 1+ region 2). If the DB-DC m ethodology is followed, a No re sponse to the first price offer iP would be followed by a second price offer L iP. If the respondent then answers Yes to the second price offer, his WTP falls in region 2. If his answer to the second price offer is a N o, his WTP falls in region 1. On the other hand, if his response to the first price offer would have been a Yes, the second price offer would have been H iP and the response to this higher bi d would place the individuals WTP in either region 3 (if no) or region 4 (if yes). This further subdivisi on into four distinct regions is responsible for the significant improvement in statistical effi ciency observed in the DB-DC method compared to the SB-DC method (Hanem ann, Loomis, and Kaninnen 1991). We can see that more information is revealed about th e individuals WTP by us ing the DB-DC methodology as opposed to the SB-DC methodology. More informa tion results in higher efficiency in the DBDC method versus the SB-DC meth od at similar sample sizes. The possible answer sequences to the original and follow-up bids and their corres ponding regions are summarized in the Table 5-1. The indicators (nyynyyIII ,,, and nn I ) identify each individual by th eir response sequence (i.e., 1 yy I (respondent answers were Yes, Yes)) where 1(.) is an indi cator function equal to 1 if its argument is true and zero otherwise. The indicators basically tell us which region the individuals WTP falls into. In Chapter 4 we presented a formal mathema tical derivation for the likelihood of a yesyes sequenceYY i; while the likelihoods for the th ree other possible sequences (NY i YN i, and NN i) were briefly discussed. We present here a less rigorous but more intuitive derivation of these four likelihoods. From Figure 5-1 and the preceding discussion, we know that d ifferent response sequences to DB-DC questions result in classification of the individuals WTP into four

PAGE 126

126 different regions. The four likelihoods are simply the areas below each region of the distribution. The upper and lower bounds (U iAand L iA, respectively) of integra tion for each region are given by L iiPP, and H iP; and depend on the response sequence given by the individual (see Figure 5-1 and Table 5-1). The following derivation uses Eq uation 5-7 as the model for WTP and assumes a symmetric distribution for the error term 2i You can refer to Figure 5-1 to visualize each region. Region 1 (L i U i L iPAA ,): 2 22 2 22 2 2222 2' Pr ) 'Pr() Pr( i L i i L i i L ii i L ii NN iP F P P Py x x x Region 2 (i U i L i L iPAPA ,): 2 22 2 22 2 22 2 2 22 2222 2'P'P 'P 'P Pr ) 'Pr() Pr( i L i ii ii i i L i ii i L i ii L i NY iF F P PPyP x x x x x Region 3 (H i U ii L iPAPA ,): 2 22 2 22 2 22 2 2 22 2222 2' Pr ) 'Pr() Pr( ii i H i i H i i ii H ii ii H iii YN iP F P F P P P PPyP x x x x x

PAGE 127

127 Region 4 (U i H i L iAPA ,): 2 22 2 2 22 2 22 2' 1 Pr ) 'Pr() Pr( i H i i i H i i i H i i H i YY iP F P P yP x x x where 2 is a scale parameter. We can then write the log-likelihood function un der the assumption of )1,0(~2Nei by using our indicator variables as: 2 22 2 22 2 22 2 22 2 22 1 2 22 2,2' ln ln ln 1ln )|,(ln i L i nn i i L i ii ny i ii i H i yn i N i i H i yy i iixP I xP xP I xP xP I xP IxyL (5-9) The likelihood function presented in Equation 5-9 is identical to that presented in Equation 4-9 of Chapter 4 but we have deri ved it in a different manner. As we have already mentioned, Maximum Likelihood methods may be used on Equation 5-9 to obtain estimates of the population parameters: 2 and 2 This completes the mathemati cal derivation of the likelihood for the traditional DB-DC model. We now follow to derive the likelihood for the DB-DC Sample Selection model.

PAGE 128

128 Sample Selection Model Our understanding of the DB-DC mechanism should now allow us to introduce the selection mechanism shown in Equation 5-6 into our model. The selectio n mechanism shows that response to the WTP questions is contingent on some latent variable 1 iy being positive; that is, the respondent only participates in the sample if 0* 1iy, otherwise we observe a missing value for that individual. In the context of CV research the latent variable 1 iy may be understood as a threshold which triggers th e decision to participate in th e hypothetical market presented by the valuation exercise; possibly the le vel of interest of the respondent. As before, denote L iA and U iA as the lower and upper bounds for WTP, respectively. The general likelihood function for th e model that takes the selection mechanism into account is: 01 2 1 111) ,0Pr()0Pr(iiyy U ii L i i iAyAy yL (5-10) A Maximum Likelihood approach to estimating this model necessitates full specification of the joint distribution of 2 1,(iiyy). We assume a bivariate normal distribution ),,,','(2 2 2 12211i ixxBVN, where 1 2 and are the standard deviat ions of the marginal distributions of 1 iy and 2 iy, and the correlation coefficient between 1 iy and 2 iy, respectively. Let represent the covariance matrix of the errors from both equations: ),cov(21 The parameters in the covariance matrix must be normalized for identification. Following Amemiya (1984), if there is no constr aint on the parameters, we can use 11 and 12/ for identification without any loss of generality. The covariance matrix is given by: 2 221 12 2 1 21'),cov( Eii (5-11)

PAGE 129

129 where 21 represents an 2 xn matrix containing the distur bance vectors, and the second equality follows from 0021 E. The normalization is ach ieved by (i) dividing each element of by 2 i, (ii) realizing 21 21 12 and (iii) letting 12/: 2 2 1 2 2 2 1 21 2 1 12 2 1 2 11 so that is distributed, 2 2 11 0 0 ~ Ni i Let, iiz11 and /221 iz denote the two correspondi ng standard normal errors distributed jointly as ),1,1,0,0(),(21 BVNzzfii. Our model, as presented in Equation 5-10, divides the entire ),(21 iizz support plane of ),(21 iizzfinto the following five separate regions: Region 1 (L i U i L iPAA ,): ,' ),( ,'Pr ,' Pr ) ',0'Pr() ,0Pr(22 11 '/ )'( 2121 22 2 111 22 2 11 1 2 22 111 2 11 12 2i L i i P iiii i L i i ii i L i i ii L ii i i i L ii iP dzdzzzf P z z P z z Pz z Py yii L ix x x x x x x xxx

PAGE 130

130 Region 2 (i U i L i L iPAPA ,): ,' ,' ),( ,'Pr ,' Pr ) ',0'Pr() ,0Pr(22 11 22 11 '/ )'( /)'( 2121 22 2 22 111 22 2 22 11 1 2 22 111 2 11 12 2 22i L i i ii i P P iiii ii i i L i ii ii i i L i ii ii i L ii i ii L iiP P dzdzzzf P z P z P z P z Pz Pz PyPyii i i L ix x x x x x x x x x x xxx x Region 3 (H i U ii L iPAPA ,): ,' ,' ),( ,'Pr ,' Pr ) ',0'Pr() ,0Pr(22 11 22 11 '/ )'( /)'( 2121 22 2 22 111 22 2 22 11 1 2 22 111 2 11 12 2 22ii i i H i i P P iiii i H i i ii ii i H i i ii ii H ii iii i H iiiiP P dzdzzzf P z P z P z P z Pz Pz PyPyii H i iix x x x x x x x x x x xxx x Region 4 (U i H i L iAPA ,): 1 12 2 11 11 22'/ )'( 2121 2121 /)'( 2121 2 22 111 2 22 111 2 22 111 2 1),( ),( ),( ,'Pr ,' Pr ) ',0'Pr() ,0Pr( ii H i i i i H iP iiii iiii P iiii i i H i ii i i H i ii i i H ii i i H iidzdzzzf dzdzzzf dzdzzzf z P z z P z z Pz yPyxx x x xx x x x x x

PAGE 131

131 ,')'(22 11 1 i H i i iP x x x Region 5: )'(1)' Pr()0'Pr()0Pr(1 1 1 1 1 1 i ii i i iz z yx x x where, to simplify notation, we have used ),,(21 iizz to summarize the double integral over the bivariate standard normal density function with integration limits from up to the arguments inside (.) The third equality in the derivation for each region follows from symmetry of the bivariate normal (BVN) density. Th e first term in the last equality in the derivation of Region 4 follows from the fact that integration of ),(21 iizzf across all values of iz2 returns the margin al distribution of iz1. All five regions are presented in Figure 5-2 where the support plane for the random error terms in our bivariate model is plotted with iz1 on the horizontal-axis and iz2 on the vertical-axis. For example, those farmers who did not enter the CV market and are therefore considered missi ng values are represented in Region 5. The contribution of such individuals to the like lihood function (Equation 5-12) is simply the integration of the marginal distribution of iz1 over the interval )',(11 ix Since iz1 is distributed standard normal, this is simply )'(111 ix Those individuals who entered the market and gave a yes,yes response sequen ce are represented in region 4 and so forth. Using our indices and the math ematical derivations for the likelihoods associated with each region we write the log likelihood function for the DB-DC Sample Selection model as.

PAGE 132

132 ,'ln ,' ,'ln ,' ,'ln ,')'(ln )]'(1ln[)1()ln(22 1 22 1 22 1 22 1 22 1 22 1 11 1 1 1 11 1 i L i i YN i i L i i ii i NY i ii i i H i i YN i i H i i i YY i N i i N i i iP I P P I P P I P Iy y L x x x x x x x x x x x x x x (5-12) Estimation of Equation 5-12 is possible by Maximum Likelihood (ML) methods to obtain estimates of ,,,21. If the true value of is zero the two equations are independent. Estimating the model while forcing 0 will give the same results as estimating each equation separately. Testing for sample selection is pos sible within the model and is done by testing a single parameter restriction () 0: oH. Empirical Application We use the DB-DC Sample Selection model to test for possible sample selection bias in the estimates of farmers WTP for GM seed tech nologies we obtained in Ch apter 4. Due to data limitations, estimation of a correct sample selecti on model is estimated and tests are carried out only for the GM fertilizer saving (FS) trait. Data and Surveying Strategy: Some Highlights Most of the details concerning the data were already presented in Chapter 4. Here we discuss only those characteristics of the surveying strategy that are relevant to this chapter. The two main surveys 2006 corn poll (CP06) and 2007 corn poll (CP07) were carried out in separate face to face interviews with the farmers. Most of the socio-demographic data was

PAGE 133

133 collected with the CP06. The CP07 included some questions on individu al characteristics but was mainly a CV questionnaire designed to elicit farmers WTP for the GM traits. Farmers who were randomly selected to partic ipate in the CP06 were not necessarily also randomly selected to pa rticipate in the CP07.9 Out of the 451 farmers randomly assigned to participate in the CP07, only a tota l of 345 farmers were also select ed to participate in the CP06. Farmers were assigned the same identification numb er in both surveys so the two data sets were merged for those 345 farmers that participated in both polls. These 345 farmers formed the base sample for estimations in Chapter 4 and in this chapter. The strategy just described presented some sp ecific advantages and disadvantages. As we discussed earlier in our section on surveying strategies, estimating a self-selection model requires gathering data for non-respondents. In specific data on variables that are hypothesized to have an effect on the decision to participate in the hypothetical market generally these are socioeconomic variables. The main advantage of our surveying strategy was derived from using one survey instrument to gather CV data and a separa te survey instrument to gather data on socioeconomic variables. This strategy held the potential to provide a data structure that permits estimating the DB-DC Sample Selection mode l and testing for self-selection bias. Unfortunately, the surveying strategy ha d the disadvantage that some of the nonrespondents in the CP07 sample, wh ich are the primary suspects for sample selection, were not selected to participate in the CP06. This acciden tal culling represented th e main disadvantage of the surveying strategy. In addition, some of those suspect non-respondents from the CP07 that were indeed selected to participate in the CP06 did not respond all of the socio-economic questions, thus the observation had missing values and could not be considered for estimation. 9 This was mainly because both surveys formed part of a larger project studying GM adoption and because of funding availability.

PAGE 134

134 Fortunately, in the case of the FS GM trait, the data left after merging the two data sets and eliminating incomplete observations, had enough va riability to allow us to estimate the DB-DC Sample Selection model. Results The application of a sample selection model was motivated by the observation, in our sample, of a considerably larg er number of non-respondents to the CV questions concerning the GM version of a trait compared to the number of non-respondents to CV questions concerning the nonGM version of the same trait. To observe this contrast in survey response be havior we started by id entifying the different types of respondents that were allowed by the CV survey design Table 5-2 presents the three main types of respondents that we identified: (i) respondents, (ii) non-bargainers, and (iii) nonrespondents. Respondents are define d as individuals who provided answers to both the initial and follow-up questions. Non-respondents are exactl y the opposite, those who answered neither initial nor follow-up. The last type are individuals who gave a response to the initial price offer, but refused to engage in the bargaining situati on presented by a follow-up offer we call these non-bargainers. All three types of respondent s are possible for each version of the trait (nonGM and GM). This is shown in Figure 5-3 where the arrows are drawn to illustrate the possibility of farmers changing their response behavior wh en asked about different versions of the trait. For example, an individual who answered both initial and follow up questions (i.e., respondent) when queried about the nonGM trait could decide to answer only th e initial question when asked about the GM trait (i.e., non-bargainer). The most repeatedly observed change in behavior was switching from being a respondent in the nonGM case to a non-respondent in the GM case.

PAGE 135

135 Table 5-3 presents a tabulation of the re sponses to the nonGM que stions against the possible responses to the GM questions. Respondents are represented in the columns and rows labeled No No, No Yes, Yes No and Yes Yes. Non-bargainers are represented in columns and rows labeled No Miss and Yes Miss. Non-respondents are represented in the column and row labeled Miss Miss. Each element in any row/column represents the number of individuals in the sample who gave the response sequences presented in the labels of that row and column. For example, 77 individuals gave a No No response se quence in both the nonGM case and the GM case. Or for example the elemen t in the Miss Miss column and the No Yes row indicates that 2 people gave a No Yes response sequence when asked about the nonGM trait, but answered neither initial nor followup question when asked about the GM trait. Notice that out of a total of 54 missing values (see column Miss Miss) 41 are individuals who changed from being respondents in the non GM questions to non-respondents in the GM questions (i.e., 16+5+6+14). These individuals were the suspects that motivated our exploration of potential sample selection problems. The estimation results are presented in Tabl e 5-4. The results shown under the column labeled Univariate correspond to those obtaine d by estimating each equation alone. In other words, the Univariate-Selection column presents the results of estimating Equation 5-6 using a univariate probit; while the Univariate-WTP column presents the results of estimating the traditional DB-DC model (i.e., Equation 5-7 alon e). The results presented under the column labeled Sample Selection show the estimate s obtained via Maximum Likelihood estimation of the DB-DC Sample Selection model. The likelihood function in sample-selection mode ls may behave in an irregular manner, but is usually well behaved for fixed values of (Strazzera et al. 2003). Convergence of our sample

PAGE 136

136 selection model was achieved in three steps. In th e first step, the model was allowed to iterate but the estimate of was fixed at a value of zero. This fi rst step obtained the estimates for the univariate models (Table 5-4). Th e second step used the univariate estimates as init ial values in the sample selection model but restricted them so that only could vary in each iteration. The last step allowed all parameter estimates to vary freely.10 The choice of variables in the selection equati on was based on three criterions: (i) intuition as to which variables may possibl e affect interest and therefore decision to participate in the hypothetical market, (ii) prelimin ary univariate probit estimations testing for significance of the variables selected using the first criteri on, and (iii) a close respect for parsimony. Comparing the estimates obtained by the univariate models to those of the sample selection model shows no major differences in magnitude or in significance of the parameter estimates. This is similar to what was found by previous studies applying this model. As explained earlier in this chapter, the ex istence of sample selection bias depends on whether the errors terms in the selection equati on and the outcome equation are correlated. In the DB-DC Model, the error terms are jointly modele d as a bivariate standa rd normal distribution. Thus, sample selection is dependent on the valu e of the correlation parameter in the bivariate normal distribution (i.e., ). A value of statistically significant di fferent from zero would be evidence of sample selection bias affecting the univariate WTP model estimates. Our estimated value for is 0.307 with a p-value of 0.887 which is far from significant. Alternatively, one could test for sample sel ection by using the Likeli hood Ratio (LR) test statistic11 10 All of the estimations were carried out in GAUSS 7.0. using the maxlik add-in. 11 Not the same as the one presented in Table 5-4. The LR st atistic presented in Table 5-4 is a test for the global null hypothesis that all the betas are equal to 0.

PAGE 137

137 2 1~| lnln ln|2ction SampleSele WTP SelectionLL L LR where SelectionLln and WTPLln are the log-likelihoods for the univariate models and ction SampleSeleLln is the log-likelihood for the sample selection mode l. The calculated LR st atistic in our case was equal to 0.03=-2*|-54.29-193.01+247.28| which has a non-significant p-value of 0.84. Based on these two tests, it seems that no sample selection bias affected out estimates from Chapter 4. It appears that farmers who d ecline participation in the hypot hetical market presented by the CV exercise are making such decision inde pendent of their WTP. Being a nonGM specialist was the only factor found to have a significant eff ect on the probability to participate the effect is negative. Given no sample selection wa s detected, nonGM specialists who declined responding the GM CV question may have simply abstained from giving any comments regardless of their WTP for the GM trait. One possible explanation for this is that farmers who chose not to participate in the hypothetical market may have done so simply becau se of the complexity associated with GM crop production. For example, the non-respondin g nonGM farmers may have been unaware (or improperly informed) of all the possible tran saction costs and produc tion requirements (e.g., fees, refuge practices) associated with GM crops and may have found themselves unable to construct their valuations. Anot her possible explanation for nonparticipation could be that nonGM farmers, regardless of their WTP for the GM trait, simply felt insufficiently motivated to give comments about a competing technology. As a final note, one should be careful when in terpreting the results and remember that, like any other ML estimate, the obtained estimates are validated by asymptotic theory. In our case, the fact that the model was empirically estima ble is a good indication, but optimally, we would have liked a larger sample size. In our interpretation, we have assumed that sample size and

PAGE 138

138 variation have been sufficient, but we cannot provid e definitive evidence th at this is true given the complex nature of the model. In fact, an inte resting research question for future studies could be determining appropriate sample sizes for the DB-DC Sample Selection model by use of Monte Carlo methods, for example.

PAGE 139

139 Table 5-1. Regions, indicator s, and possible answer seque nces to DB-DC questions Region Indicator Pi Pi L Pi H 4 Iyy Yes Yes 3 Iyn Yes No 2 Iny No Yes 1 Inn No No Table 5-2. Types of respondents Initial bid Follow-up Respondents Non-Bargainers X Non-Respondents X X Table 5-3. Tabulation of responses to the fe rtilizer saving trait CV questions (nonGM and GM) GM question NonGM question No No No Yes No Miss Yes No Yes Yes Yes Miss Miss Miss Total No No 77 4 1 3 3 0 16 104 No Yes 16 60 0 4 2 1 5 88 No Miss 1 0 7 0 1 0 2 11 Yes No 21 8 0 45 9 0 6 89 Yes Yes 13 5 1 9 100 1 14 143 Yes Miss 1 0 2 0 0 1 0 4 Miss Miss 0 1 0 2 0 0 9 12 Total 129 78 11 63 115 3 52 451

PAGE 140

140 Table 5-4. Sample selection model estimation results Univariate Sample Selection Selection WTP Selection WTP Estimate p-value Estimate p-value Estimate p-value Estimate p-value constant 1.765** (0.024) 2.733 (0.683) 1.802** (0.027) 2.628 (0.717) f_age -0.008 (0.504) 0.096 (0.259) -0.009 (0.497) 0.089 (0.355) f_educ 0.134 (0.189) 0.740 (0.267) 0.133 (0.194) 0.824 (0.371) f_nfjob -0.333 (0.266) -1.734 (0.394) -0.340 (0.260) -1.962 (0.458) f_income -0.029 (0.631) 0.868** (0.030) -0.025 (0.705) 0.840** (0.063) grain -2.534 (0.300) -2.477 (0.342) earlyadopt 10.994** (0.013) 11.008** (0.013) yield 0.003 (0.901) 0.003 (0.897) seedcost 0.075** (0.096) 0.075** (0.097) herbcost 0.090 (0.245) 0.090 (0.251) insectcost -0.027 (0.851) -0.028 (0.845) fertccost 0.051** (0.078) 0.051** (0.077) farmsize -0.002 (0.315) -0.002 (0.315) nonGM100p -0.782** (0.005) -7.678** (0.002) -0.786** (0.005) -8.408 (0.120) sigma 9.360** (0.000) 9.465** (0.000) rho 0.307 (0.887) N 174 155 174 Log-likelihood -54.294 -193.012 -247.288 Mean WTP i 17.272** (20.094) 16.287** (4.011) LR Index 0.550 0.548 0.544 LR statistic ii 11.413* (0.076) 131.608** (0.000) 184.706** (0.000) i For Mean WTP the t-statistic is reported in parenthesis. Sta ndard error for mean WTP was cal culated using the Delta Method. ii LR statistic is for likelihood ratio test with 0... :21 oH, sigma is left unconstrained. ** indicates signifi cance level of 0.05 and indicates significance level of 0.1.

PAGE 141

141 Figure 5-1. WTP regions based on the traditional double bou nded (DB) model. Figure 5-2. Regions over the (iizz21,) plane implied by the double bounded dichotomous choice (DB-DC) sample selection model. )' Pr(11 1 iiz x iz2iz1 11' ix 22'iiP x 22'i L iP x 22'i H iP x i i H i iiz P z2 22 111' ,' Pr x x 22 2 22 111' ,' Prii i i H i iiP z P z x x x 22 2 22 111' ,' Pri L i i ii iiP z P z x x x 22 2 111' ,' Pri L i i iiP z z x x 1 2 3 4 5 1 4 2 3 Pi H Pi Pi L

PAGE 142

142 Figure 5-3. Possible changes in survey response behavior between the nonGM and GM questions.

PAGE 143

143 CHAPTER 6 FURTHER RESEARCH AVENUES In this chapter we further explore so me concepts and assertions that were made in previous chapters and present some new concepts and hopefu lly some valuable insights. The discussion is centered upon the uncertainty in crop production The model used in Chapter 4 considered production uncertainty by in cluding expected profits y instead of simply profits y as the relevant variable influencing adoption decisions. Because of the framework of the model used in that chapter and because of the uncertainty inherent in agricultural production, the y approach seems more correct. A similar observation is made by Qaim and DeJanvry (2003) but ignored by Hubbell, Marra, and Carlson (2000). We provide an analysis of ge netically modified (GM) seed technologies framing production uncertainty under the state contingent paradigm put forth initially by Arrow (1953, 1964) and Debreu (19 52) and further developed by Chambers and Quiggin (2000).1 As we shall see, the state continge nt approach provides a more intuitive conceptualization of crop uncertainty. Our purpose in this chapter is not that of developing and implementing a new theory, rather we hope that reframing and discussing GM crops and farming uncertainty under the more intuitive state contingent approach will help us shed some light on the GM crop farmer adoption scenario. From an agronomic perspective, is there any inherent difference between GM seed inputs and any other agri cultural input? Is it correct, as Qaim and DeJanvry (2003) suggest, to consider GM crops such as Bt varieties an ex ante imperfect substitute of more traditional inputs such as pesticide? Can we find any linkages between the state contingent approach presented here and the approach we followed in Chapter 4 that would reinforce our understanding of the GM seed adoption scenario? These are some of the questions we try to answer here. 1 See also Quiggin and Chambers (2006).

PAGE 144

144 Besides the production model developed by Ch ambers and Quiggin (2000), the statepreference approach inaugurated by Arrow (1953,1964) and Debr eu (1952) spanned another body of literature which is relevant to us. Grah am (1981) formally developed the theory of option price under the state-continge nt paradigm. At the end of th is chapter we briefly discuss the concept of option price and again try to fi nd the linkages between such approach and our findings in Chapter 4. Our analysis starts by discussing uncertainty in agricultural production. Uncertainty in Agricultural Production In the absence of knowledge, everything would be uncertain. If we had perfect knowledge about everything (including the future) uncertain ty wouldnt exist. The ongoing objective of scientific human knowledge is to explain and gain control over our natural environment. As scientific knowledge evolves we develop more re fined explanations of the world, we learn how our actions may affect outcomes in given stat es of the world, and we reduce uncertainty.2 This is true in all sciences and especially true in agricultural sciences since agricultural production relies heavily on the use of natural resources. The discovery and transition into agriculture by hunter gatherers ( 10,000 years ago) is in fact a good example of human attainment of understanding followed by appropriation and c ontrol of the natural environment to reduce uncertainty in the provision of food. It must be emphasized, however, th at agriculture did not invent food production, photosynthesi s already existed in nature. Ag riculture simply gained us better control over the food production process that happe ns naturally in the plant. Enormous advances have taken place sin ce the first crop was grown by early protofarmers. The set of possible actions provi ded by todays improved understanding of plant 2 This is evident in the historic path followed by variou s physical sciences which st arted as primitive (sometimes metaphysical) explanations of rare observed events and evolved through time into sophisticated theories (Haavelmo 1944), and finally into practical applications used in industry and science.

PAGE 145

145 biology and used to obtain desired outcomes with more predictability, may be categorized into two distinct strategies : (i) taking actions to control and modify the environment surrounding the plant, and (ii) taking actions to control and m odify the metabolism and genetic material of the plant itself. The latter strate gy is consistent with the action taken by farmers who adopt GM seeds. Although both strategies have different imp lications in terms of wh o is appropriating the benefits and the knowledge acquired through developing and implementing the action, we will argue that from an economic perspective which is less concerned with the distribution of these benefits but, rather, with the analysis of uncerta inty, the two strategies may be analyzed using similar conceptual frameworks. Strategy 1: Modifying the Environment Surrounding the Plant The first strategy is to take actions in or der to modify the environment immediately surrounding the plant. We now have a more or less complete knowledge of the natural factors (water, nitrogen, air, etc) that play a role in photosynthesis and that regulate crop metabolism and growth. Each crop and even each crop variety has its own characteristics. We have developed an extensive knowledge bank of optimal levels of these factors that more consistently produce high output outcomes. An army of agro nomists, entomologists, plant pathologists, soil scientists, and other agriculture professionals are constantly working to expa nd our understanding of crop plants by performing controlled trials and presenting us with results such as optimal nutrient requirements, pest population dynamics and thre sholds, water requirements, and the like. From the perspective of a farmer whose goal is that of output maximization, nature results inefficient in providing optimal levels of production factors wher e and when they are necessary. Nature is unpredictable and inconsiderate to th e farmers goal of output maximization. Timely and efficient modification of the envir onment surrounding the plant has required the understanding and appropriation of th e natural cycles of the impli cated natural factors. Whereas

PAGE 146

146 nature provides us with uncontrollable and unpred ictable factors necessary in crop production, we are persistently finding ways to appropr iate those factors and develop proper (more controllable) substitutes that we use to supplemen t natural supplies when they are deficient. The water cycle has been appropriate d to produce irrigation that suppl ements rainfall, the nitrogen cycle has been appropriated to produce ferti lizers that supplement natural fixation by microorganisms, etc. In microeconomics, these appropriated natural fa ctors are called inputs. Inputs are used in agriculture to supplement uncertain natura l supplies when the situation requires so3 for example, irrigation supplements deficient na tural rainfall, fertilizer supplements low concentrations of nutrients in soil. In contrast with natural supplies, inputs are: costly and controllable. Natural supplies, on the other hand, are uncontrolle d by the farmer, are free, and their provision levels are uncertain. Strategy 2: Modifying the Geneti c Material of the Plant Itself A second strategy for reducing uncertainty is to control the metabolism of the crop by modifying its genetic composition. Rather than gaining control over the plants environment, plant breeders and crop engineers are in the busine ss of appropriating the ge netic material of the crop itself. By altering the ge netic material of crops, eith er by sexual reproduction or by recombinant DNA techniques, scientists are able to endow crops with previously identified desirable traits. As with the case of other natura l factors, here also nature is inconsiderate of the farmers goals. There is no guarantee, nor there should be, that natural selection and mutation will provide the agronomic-desired genetic mate rial. The use of genetic engin eering and selective breeding to 3 Which situations are considered to require supplementation depend on the requirements of the crop and the farmers goal which is usually taken to be output maximization.

PAGE 147

147 confer desired genetic material (and traits) to a crop is simply a way to supplement the uncertain genetic material provided by natu ral mutation and natura l selection. Thus, from an agronomical perspective, appropriating the genetic material in a crop and controlling it is, in essence, no different than appropriating any other resource provided by nature. As such, in technical terms, conferred genes (or tra its) could be seen as inputs in the production function with no fundamental difference to water, fertilizer, etc. State Contingent Model of Production The uncertainty inherent in GM crop adop tion has traditionally been oversimplified and sometimes even neglected. Analogously, the anal ysis of crop production in general, models uncertainty using a stochastic pr oduction function which is simply a production function with a random error term added to it: )(x fy (6-1) This approach assumes that the decision maker cares only about the distribution of outcomes or payoffs he receives, not about the unde rlying events, or states of nature, that cause these outcomes. This approach is more consistent with a lottery apparatus generating these outcomes, rather than with natural factors that are localized and differ across individuals. When the random outcomes are generated by some underly ing causes, a more detailed description of uncertain alternatives is possible. For example, Chambers and Quiggin (2000) propose a slightly different approach to modeling production unde r uncertainty based on the notion of state contingent production. Rasmussen (2003) notes that one of the major problems with the stochastic production approach is that the well -known marginal principle used to prescribe optimal production decisions (i.e., MC=MR) unde r certainty breaks down under uncertainty. Quiggin and Chambers (2004) argue that an even more critical weakness of the stochastic

PAGE 148

148 approach is that it is not amenable to diagramm atic analysis of the kind that remains the main source of intuition for economists studying produ ction under certainty. The state-contingent approach, on the other hand, permits the applicat ion of a whole range of concepts developed for production theory under certainty, including both marginal and diagrammatic analysis and duality theory. In the basic state-contingent m odel with only one output there are N different inputs and S possible states of nature. Inputs N x are committed ex ante and fixed ex post. The state contingent output (or output profile)1 Sx z is chosen ex ante but produced ex post. Uncertainty enters the model through the stochastic states of natu re represented in the probability space } ,..,2,1{ S. The model may be understood as a two period game with nature. In pe riod 0 (i.e., ex ante) the decision maker commits inputs N x. The level of inputs she commits determines the vector of possible outcomes 1 Sx z she may observe in all possibl e states of nature, the exact outcome depending on which state of nature occurs. Nature reveals the state of nature after the decision maker has committed N x; this results in output z sz being produced in period 1 (i.e., ex post). 4 The state-contingent approach seems to be a very good conceptual ization of farming activities. On the field, the farmer commits inputs ex ante towards producti on but he does not immediately reap the products, he must wait for the crop to complete its cycle in order to harvest. 4 A fundamental presumption of the state-space approach is that the decision maker can do nothing to determine which state of nature will occur. The states of nature, re dundantly speaking, are provided by nature and uncontrolled by the decision maker. This does not mean, however, that the decision ma ker is impotent when it comes to the future. Instead of her decisions affecting which state of nature occurs, they a ffect the outcome realized if a given state of nature occurs. For example, a farmer may not be able to influence the chance of rain but he may be able to take actions that prepare him in the case of drought.

PAGE 149

149 Basically, life and growth takes time. In the mean time, as crop growth takes place, there are numerous possible states of natu re that may have different infl uences over out put and make it uncertain. Ex post nature reveals the state of nature and a given output results. However, one could still think of a more caref ul treatment of the time compone nt. In the following subsections we introduce some new concepts (i.e., perfect timi ng and perfect supplementation) that allow for a more detailed treatment of the time component; of uncertainty; and of th eir interactions with information technologies and characteristics i nherent to the different production inputs. Perfect Information, Perfect Ti ming, and Perfect Supplementation For the sake of argument, suppose for a moment that seed is sowed on a field and then no other action is taken by the farmer (i.e., zero inputs). In this case, the uncertainty in natures provision of natural factors will traduce directly in to uncertainty in yields. For example, all other natural factors fixed, if nature provided deficien t (optimal) rainfall this would traduce directly into low (high) yields. Of course, the farmer w ouldnt be much of a farmer if he were to let nature alone dictate his year to year output. To continue this argument, let us now try to picture what may be considered the other extreme. What should happen and what actions should the farmer take to completely eliminate uncertainty in yields so that he obtains the same yield every singl e time regardless of what nature does? In order to answer this question, let us fo r a moment ignore that inputs are costly. First, suppose that some new information-gathering tec hnology provides the farmer with the ability to know exactly how much of each factor (e.g., water, nutrients) is provided by nature at any given moment in time to the plant (i.e., the farmer ha s perfect information). In addition, suppose some other technology allows the farmer to take action on this new information and supplement natural provision on a real -time basis (i.e., perfect timing). Given this perfect information about what enters and leaves the production system and the ability to supplement with perfect timing,

PAGE 150

150 in theory, a farmer should be able to perfectly increase or decrease supplementation as necessary and completely eliminate yield uncertainty. This makes sense agr onomically because, for example, the plant doesnt mind where the water is coming from, whether it is from irrigation or from natural rainfall. Of course, this implicitly assumes that controlled inputs are made to quality standards that make them perf ect substitutes of natural factors. The assumptions made perfect information and perfect timing to achieve perfect supplementation are quite extreme. They are, however, the ultimate (possibly utopian) goals of precision agriculture. Precision agriculture uses information-gathering and input-delivery methods and technologies (e.g., soil mapping, geos tatistics, geographical information systems, weather forecasting, GPS) to mana ge natural resource variability.5 In the going we have ignored costs. Notice, however, that supplementation entails costs. For example, obtaining a normal yield (h z ) in corn requires ar ound 22 inches (2,2503m) of water per acre during the grow ing season (Wright et al. 2008).6 Suppose rainfall during the growing season is 16 inches (1,6353m) per acre. This means that in order to obtain yield h z the farmer will have to incur the costs asso ciated with supplementing the 8-inch (6153m) water deficit via irrigation. If rainfall during th e season were to be only 10 inches (1,0223m), the farmer would have to incur highe r irrigation costs necessary to supplement not 8 but 12 inches of water via irrigation in order to obtain the same yield h z As should be obvious, perfect supplementation eliminates uncertainty in yield but traduces it to uncer tain, potentially high costs. 5 See Precision Agriculture Journal aims and scope available at http://www.springer.com/life+sci/agriculture/j ournal/11119?detailsPage=aimsAndScopes. 6 This depends on timing of supply (e.g., emergence, matu rity) and on other production practices (e.g., fertilization, pest management).

PAGE 151

151 Perfect Supplementation and the State Contingent Approach In our example of perfect supplementation, we argued that given the right tools there is no reason why the farmer shouldnt be able to comp letely eliminate uncertainty in yields that follows from uncertainty in states of nature. We also argued that using perfect supplementation to eliminate uncertainty in yield simply traduces to uncertainty in costs. Quiggin and Chambers (2004) make this same point: in the state-contingent model, output uncertain ty is a result of producer choices. Producers, if they choose, can stabilize out put completely, though they may incur (potentially very large) costs in doing so. In making this note, Quiggin and Chambers seem to make the same assumption of perfect information we used to discuss perfect supplementation. However, the additional assumption of perfect timing that we use to define perfect supplementation implies that no two-period game ta kes place between farmer and nature. That is, if not only the farmer could monitor perfectly each production factor (e.g., water, nutrients) but also supplement found deficiencies on a real-time basis, then he could avoid acting before nature and eliminate uncertainty. No ex ante or ex post is necessary, everything happens on the spot.7 Seed Technologies Before making quick generalizations, we should ask ourselves Is perfect supplementation is possible for every input in agricultural productio n? To answer this, let us first note that agroecological systems are an incredibly dynamic pl ace. Factors constantly enter and leave the system. Nitrogen, for example, may enter the sy stem by fixation or fertilizer application and leave the system via leaching, volatilization, or denitrification. Water may enter via rainfall or 7 Of course, in practice this does not happen for obvious reasons (costs, time), but entert aining the example help us gain some insights.

PAGE 152

152 irrigation and leave via evaporat ion or drainage. Pests and weed s appear and are allowed or forced to disappear. One factor, however, is a constant in the system the seed.8 For the case of seeds, the farmer must nece ssarily engage in a two-period game with nature. Gathering more and better information or improving timing in delivery (i.e., precision agriculture) will not allow the farmer to avoid acti ng before nature in the case of seed inputs. The decision on quality and type of seed is made once and only once at the beginning of the growing exercise. Other factors may ente r and leave the system by natura l means or by actions taken by the farmer. Removing the seed from the fiel d means the end of the farming exercise. Thus, the case for a state contingent approach to uncertainty seems more robust when dealing with seed technologies. Going back to our definitions of strategy 1 and strategy 2, one could say that, given cost and te chnological limitations that pros cribe perfect supplementation, the state-contingent approach makes a good conceptualization of the farm ing situation for both strategies; but if perfect supplem entation were possible, the state-contingent approach would still be a good conceptualization for strategy 2.9 As for GM crops, one should consider that ge netic traits are embedded in the seed. It should be clear now that in contrast to othe r inputs which may allow some flexibility, the decision of whether to adopt or not adopt a GM seed technology is necessarily ex ante. This reasoning is consistent with our find ings in Chapter 4, where we argued10 that GM seed technologies may be seen by the farmer as im perfect substitutes for other production inputs because of their ex ante nature. 8 Soil structure may also be constant. 9 Only if the farmer had perfect information about the future could the farmer avoid playing a two-period game with nature. 10 Like Qaim and DeJanvry (2003).

PAGE 153

153 State-General and State-Specific Inputs At this point, we should define some concepts developed in the stat e contingent literature before continuing our discussion of GM crops Rasmussen (2003) dist inguishes between two important different types of i nputs: state-general and state-sp ecific. For simplicity, following Rasmussen, in the following we will cons ider only two states of nature } 2,1{ A state general input (Figure 6-1) is defined as an input that in fluences output in one or more states of nature. Part A of Figure 6-1 shows the production function )(11xffor the state-general input 1x in state 1.11 Different levels of contingent output 1z are obtained for different levels (a, b, and c) of 1x. In state 2 (Part B) varying 1x also has an effect on con tingent output, in this case 2z, via the production function )(12xf. The possible pairs of contingent out puts for each level (a, b, and c) of 1x (i.e., the transformation curv e) are shown in Part C. An example of a state-general input given by Rasmussen is fe rtilizer in grain production. In this case one could think of wet weather as state 1 and d ry weather as state 2. State contingent output )(111xfz in the case of wet weat her increases with fertilizer application. In the case of dry weather one can still improve output by applying fertil izer, but the effect on output for each extra unit of fertilizer is smaller as shown by a lower)()(1112xfxf. A state-specific input is pres ented by Rasmussen as a special case of state-general inputs. A state-specific input is one that influences output in only one stat e of nature (Figure 6-2). If the state-specific input has an effect on output in state 1 (Part A), it will have a flat production function )(12xf in state 2 (Part B); its transformati on function will also be flat (Part C). 11 All other inputs are assumed fixed.

PAGE 154

154 An example of a state-specific input given by Rasmussen is a pesticide that is only effective under dry weather conditions. The states of nature are the same as for our last example. If the farmer applies the pesticide and state 2 (d ry) occurs, the pesticide is effective and a higher yield result. On the other hand, if state 1 occurs, marginal increases in pesticide applications have no effect on output. Several conclusions may be derived from these definitions. First, all else fixed, statespecific inputs, since they result in larger variab ility between states of nature, are riskier than state-general inputs. Second, whether an input classifies as stategeneral or state-specif ic depends on how we define the set of states of nature. For example, if we considered three states wet, dry, and very dry; then the pesticide example may classi fy as a state general input. If instead we considered the two states water and no water then fertilizer may classify as state-specific instead of state-general. If it is the decision maker who defines the state set then a discrete set would be consistent with a view of decision makers as having bounded rationality. The decision maker uses heuristic methods to find the optimal action by defining a discrete number of plausible states of the world. The decision ma ker holds positive (>0) subjective probabilities },...,2,1{, S about each plausible state. Finally, one could think of a third type of input a perfect ly state-general input. Such input may be defined as an input that has the same ef fect on yield in all possible states of nature (Figure 6-3).12 A perfect state-general i nput would completely elimin ate uncertainty in yields with respect to that input. 12 As should be obvious from our water and no water ex ample, if we consider all possible states of the world (not only those that are plausible) this type of input is impossible since it would imply that no essential factors exist in agricultural production.

PAGE 155

155 Genetically Modified Traits: St ate Specific or State General? In this section we see how different traits in the market (or in development) may be identified as different types of inputs based on Rasmussens categorization. For example, glyphosate herbicid e tolerance (Ht), the most wi dely adopted trait, seems closer in its characteristics to a state-general input. It is generally recognized that producers of Ht crops benefit mainly from lower costs and ease of use. In most farming scenarios, chemical control results less costly than hand-weeding (e.g., Gianessi et al. 2002). The decision to adopt the Ht trait is usually depicted (in terms of gains in profitabilit y) as depending on whether cost savings in weed control are larger than seed cost premiums. Glyphosate is a wide-spectrum control herbicid e that provides effective weed control in most cases. However, if applied on conventional crops it will also kill the crop. For conventional crops, the systemic and unselective mechanism of glyphosate requires careful application to avoid crop poisoning. The possibility of output losses due to careless application by field workers gives place to a principal-agent problem. Gains in ease of use from adoption of Ht varieties, as Fernandez-Cornejo and McBride (2002) put it, occur becaus e herbicide tolerant programs allow growers to use one product instead of several herbicides to control a wide range of both broadleaf and grass weed s without sustaining crop injury Carpenter and Gianessi (2001) state that the prim ary reason growers have adopt ed Roundup Ready weed control programs is the simplicity of [the] weed control program. Because of its wide-spectrum and its systemic mechanism, glyphosate would be a good example a state-general input. Applied on conven tional crops, it has an effect on output (i.e., a negative one) in more than one, possibly all stat es of the world. Suppose the farmer sets out (prior to planting) to use glyphosate herbicide fo r controlling any emerging weed competition. In doing so the farmer has narrowed the set of possible states of nature. All po ssible states of nature

PAGE 156

156 will now include the presence of glyphosate on the fi eld. For simplicity, consider only two states of nature: state 1, glyphosate and state 2, glyphosate. Let 1x (in Figure 6-1) represent the proportion of total acres planted wi th Ht varieties in a given farm unit. Since the Ht trait is simply an antidote to glyphosate, given the sure presence of glyphosate on the field, changes in the level of1x, by definition, should have an effect on output szin both states of nature (i.e., 1x is state-general). The ease of use so often associated with Ht adoption may be explained in part by the farmers newly endowed ability to narrow the state set (i.e., reduce uncertainty).13 The design is simplified by an a priori decision for using glyphosate. The farmer avoids having to decide which herbicide to use on a case by case basis. The farmer also avoids having to monitor the careful application of th e glyphosate herbicide. The a priori decision is possible because of: (i) the availability of a glyphosate tolerant variety, (ii) the wide spectrum of weed control and effectiveness over different states of nature of glyphosat e herbicide (i.e., stat e-general input), and (iii) the stability and predictability of annual weed infestations. The second most adopted trait is insect re sistance (Bt). The main benefit generally associated with Bt seeds is an increase in yields as crop losses due to pest infestation are reduced. The Bt trait seems closer in its characteristics to a state-specif ic input. In contrast to weed infestations, annual pest infesta tions are less stable and less pr edictable (Gray and Steffey 1999). Insects are much more mobile th an weeds. Such mobility allows insects to search for the best food available and concentrate in most favorable areas or niches, leaving surrounding crop areas mostly unaffected. In addition, insect populations, even when unchecked by farmers, remain checked by populations of natural enemies. All of these factors make the frequency of crop 13 Adoption of Ht varieties may ease farming in other aspects, like for example, allowing the implementation of notillage systems.

PAGE 157

157 damage due to pest infestati on to differ across geographic regions and vary between growing seasons. It seems reasonable to consider the follow ing two states of nature: pest and no pest. Let 1x (in Figure 6-2) represent the proportion of total acres planted with Bt varieties in a given farm unit. Changes in the level of 1x (i.e., rate of adoption) will onl y have an effect on output in the state of nature pest. In the case of a Bt trait the farmer purchases the seed with the trait ex ante in case a pest infestation occurs. At the moment of purchase he is not sure that he will actually make use of the trait. If no infestation occurs the farmer still pays the technology fee but observes the same yield he would have observed with a similar crop variety that did not posses trait. This is, in essence, similar to what farmers do when they purchase crop insurance. With insurance, the farmer pays a risk premium and receives compensation only if cer tain states of nature (that are previously defined in an insurance policy) occur. This re sembles closely what we argued in Chapter 4 and what seems to be part of the reasoning be hind the implementation of the Biotech Yield Endorsement. Consider now the drought tolera nce (DT) and fertilizer saving (FS) traits. We can compare these two traits by analogy with the Ht and Bt tr aits. Like insects compared to weeds, natural water supplies are much more unpredictable and variable than natura l nutrient supplies of nutrients which are more easily pred icted and in general less variab le. Like the Bt trait, the DT (as we argued in Chapter 4) may serve as a subs titute for crop insurance. When adopting a DT trait (i.e., against severe drought), the farmer holds a relatively larg er probability of not using it (i.e., no severe drought occurs) compared to the case where he adopts a FS trait. A final point in this subsection can be made by considering our definition of a perfectly state-general input. As we defined it, such input has the same effect on yield in all possible states

PAGE 158

158 of nature. A perfect state-general input would completely eliminate uncertainty in yields with respect to that input. A seed technology w ith this characteristic, even though planted ex ante, would hold no uncertainty and so in some sense it would allow the farmer to avoid playing the two-period game with nature. For example, consider a perfectly state-general version of the DT trait and a set of possible and plausible states of nature that spread across a spectrum ranging from no drought to severe drought. That is, suppose multiple traits with diffe rent methods of action were inserted into the crop seed so that the same yiel d was obtained regardless of whic h of these states of nature occurred. Such a trait would eliminate uncertainty. What is more interesting is that such trait would substantially reduce or even eliminate the need to gather information rela ted to irrigation, rainfall, and other water supplies. Although being a bit of a stretch, this example provides a valuable insight. What it implies is that improvements in seed technologies may have the effect of deep ly altering the information data sets gathered by farmers. In fact, to some exte nt, seed traits may reduce the need for information-intensive technologies such as precision agriculture. It is not hard to see how stacked or multiple trait varieties, as they become more state-general (e.g., Bt varieties that control a larger number of insect species or glyphosate which controls most weed species), will provide farmers with ease of use such as the general application of glyphosate over glyphosate-to lerant varieties as opposed to the more precise application needed for non-tolerant varieties. Option Price Besides the production model developed by Ch ambers and Quiggin (2000), the statepreference approach in augurated by Arrow (1953, 1964) and Debreu (1952) spanned another body of literature which is relevant to us. Grah am (1981) formally developed the theory of option value under the state-contingent paradi gm. The related concept of option price,

PAGE 159

159 presented initially by Weisbrod (196 4), is defined in the CV literat ure as the price a current nonuser would be willing-to-pay for a good in order to secure the possi bility of future use of that good. The definition of option price, as presen ted by Graham (1981), considers two possible states of the world. In state 1 the individual is a user of the good while in state 2 the individual is a non-user. Graham defines option price as the sure payment [the individual] would be willing to make in both states. The concept may be readily adapted to develop a model of GM crop/trait adoption as follows. We prefer the terminology sta tes of nature as opposed to states of the world to emphasize the farmers inability to influence which state occurs. We take state 1 to be pest and state 2 to be no pest. Let } 2,1{,Ss represent the respective subjective probabilities associated with each state of nature. Adoption of the GM trait q is represented by 1q while nonadoption is represented by 0q. The farmers utility is assumed contingent on the state of nature and on whether the farmer adopted the trait:14 }2,1{,}1,0{ ;),( sj yqujs j s (6-2) where jsy represents the profits (net of the traits technology fee) observed if adoption decision j is made and state s occurs. Expected utility for a non-adopter is given by ),()(),()(00 0 00 01 0 11yquyquU (6-3) The option price OP is defined as the ex ante payment the farmer is willing to make which satisfies UOPyquOPyqu ),()(),()(10 1 00 11 1 11 (6-4) 14 To avoid confusion between the notation used in this chapter for inputs and that used for farm and farmer attributes in Chapter 4, we suppress x from the derivation.

PAGE 160

160 At any offered price P the farmer will adopt the technology if the expected utility under adoption equals or exceeds expected utility under no adoption. Formally this is given by ),()(),()(),()(),()(00 0 00 01 0 11 10 1 00 11 1 11yquyquPyquPyqu (6-5) In contrast to the model presented in Chap ter 4 where the argument entering utility was expected profits y here profits are known to the farmer at each state of nature and adoption choice. What is uncertain is the state of nature that will occur. The expectation is taken over states of nature using subject ive probabilities as weights. As noted by Chambers and Quiggin (2006), empiri cal applications of the state contingent model have proven challenging. The main obstacle is that nature reveals only one of its states so that only one outcome is observable. This is the same problem we found in Chapter 4 when we pointed out that expected changes in profits are u nobservable. However, at a very general level, we could express the expected utility as U uE],|(.)[10 which highlights the dependence of exp ected utility on the probabilities of each state of nature. This representation goes in line with the models estimated in Chapter 4 giving us some intuition on why specific variables (i.e., mD2) were found to be significant.

PAGE 161

161 Figure 6-1. Representation of a state-general input and its transformation curve. Figure 6-2. Representation of a state-specific input and its transformation curve. Figure 6-3. Representation of a perfectly stat e-general input and its transformation curve. A 1x2z abc 1z2z azbzcz0 C B 1x1za bc01z 2zazbzcz A C B 1x1za bc 1x1za bc A C B 1x1za bc 1x2z a bc 1z2z azbzcz 0

PAGE 162

162 CHAPTER 7 CONCLUSIONS Biotechnology has permitted the rapid production of GM seeds products (i.e., biotech crops) that are differentiated by trait (e.g., RoundupReady with he rbicide tolerance, YieldGard with insect resistance). Since first commercialized in 1996, biotech crops have taken the U.S. and the global seed industries by storm As of 2008, only 12 years into commercialization, cumulatively more than two billion acres have been planted at the global level with GM crops. In 2008 only, GM crops accounted for 125 million hect ares planted in 25 different nations. Four crops (soybean, maize, cotton, and canola) and two traits (herbicide tolerance and insect resistance) have traditionally represented the bulk of the market. However, the dominance of these two traits may change in future years with the release and approval of several new traits which are currently at advanced development stag es in the R&D pipelines of seed companies. Two new specific traits are of current inte rest in maize: FS and DT. An important contribution of this study is the production of estimates of farmers WTP for these two forthcoming traits. The estimates we obtained ar e for corn farmers in Minnesota and Wisconsin; however, our results at the local level should prov ide valuable insights which may guide further studies evaluating adoption of th ese traits on other regions. A nove l design in our study was the comparison of GM and nonGM versions of the same tr ait. In total, four traits were presented for farmers valuation: (i) FS-GM, (ii) FS-nonGM, (iii) DT-GM, and (iv) DT-nonGM. Our results (Chapter 4) show that the agronomic benefits of these tr aits are recognized by farmers who assign them a positive value in dollars For the FS trait, in average, farmers were willing to pay around $19.72 per acre for the nonGM version of the trait and $17.25 per acre for the GM version. For the DT trait we found a mean WTP of $20.87 pe r acre for the nonGM version and $18.73 per acre for the GM version.

PAGE 163

163 In both FS and DT traits, in average, fa rmers are willing to pa y more for the nonGM version compared to the GM version $2.47 a nd $2.14, respectively. At first glance, an important farmer characteristic driving this wedge appears to be whether the farmer is specialized in nonGM production or not. Estimation results show that, whether due to personal preferences or because of associated market risks, nonGM farmers consistently penalize the trait if it is GM. On the other hand, being a nonG M specialist has no influence on the farmers valuation of nonGM traits. Other than the asymmetric effect of nonGM sp ecialization, we find th at the factors with statistically significant effects on valuation and adoption, and th e direction and magnitude of those effects, are similar for the nonGM and GM ve rsions of both the FS and the DT traits. Some important factors increasing farm ers WTP for these traits are: purpose of production, type of adopter and familiarity, farm inco me, and costs of substitute in puts. Some interesting results relating to the estimates obtaine d for the effects of these fact ors on WTP should be highlighted. For example, those farmers identified as early adopters showed a much larger WTP for all of the new traits compared to non-early adopters. The effect was in fact the largest among the positive effects, ranging between $9.61 and $14.96 per acre across the tr aits. The effect was consistently found to be larger for the GM versio ns of each trait suggestin g familiarity with GM technologies plays a role in farm ers valuations. These results ma y have interesting implications for seed companies pricing strategies, for example, suggesting the potential to offer new products at introductory marked-up prices to cap ture the significantly larger rents that early adopters are willing to pay. Another interesting result relates to the eff ects on WTP associated with substitute inputs. For example, farmers with higher fertilizer costs showed willing to pay more for a FS trait; while

PAGE 164

164 farmers that pay more in crop insurance showed willing to pay more fo r the DT trait. More interestingly, the substitution effects in both trai ts for both substitute inputs were found to be less than perfect (i.e., less than a dollar for a dollar) In Chapter 4 we argued that such imperfect substitution may be due to the fact that seed traits are ex ante inputs. This argument was reinforced in Chapter 6 by fram ing production uncertainty under the state contingent production model of Chambers and Quiggin (2000). There, a careful treatment of the time component and of the characteristics of the different inputs used in agricultural production show ed that seed traits are, by their nature, necessarily ex ante inputs; while the ex anteness of every other input in agricultural production depends largely on th e farmers ability to engage in perfect supplementation (i.e., identifying a nd perfectly supplementing deficien cies (water, nutrients, etc) on a real-time basis). In Chapter 4 we were also able to plot adoption curves using the individual predicted WTP values obtained from our model and were able to say something about the adoption potential of GM seeds compared to nonGM seeds. Results from non-parametric tests suggest that nonGM versions of each of the traits studied, in general, hold better adoption potential th an GM versions. For the FS trait, the adoption cu rve of the nonGM version stochast ically dominates that of the GM version, which implies an unambiguous dom inance in adoption potential of the nonGM version. At prices above $20 per acre the adoption potential for both versions (GM and nonGM) of both traits is similar. Moreover, there is a sharp increase in the difference of adoption potentials at the $20 per acre mark (i.e., the curves separate from each other). Finally, as prices drop below this mark, the two adoption curves (GM and nonGM) move separately but in parallel fashion. These shapes of the adoption curves may be indi cation that different market segments are being

PAGE 165

165 captured at prices below or above the $20 per acre mark. Further investiga tion of this assertion should be an interesting topic for future studies. If it turns out to be correct seed companies could benefit by understanding how these two market segments diffe r in their characteristics (age, education, income, location, etc). In Chapter 5 we investigated the possibility that sample selection bias may be affecting our estimates of farmers WTP for the traits. Testing for sample selection in our results from Chapter 4 was important because if detected it would suggest inconsistent estimates for the parameters of the population initia lly targeted. The application of a sample selection model was motivated by the observation, in our sample, of a considerably larger number of non-respondents to the CV questions concerning the GM versi on of a trait compared to the number of nonrespondents to CV questions concerning the nonGM ve rsion of the same trait. Testing for sample selection in a DB CV is not trivial, and has in fact escaped the CV literature until recently. We estimated the DB-DC sample selection model proposed by Yoo and Yang (2001) and tested for sample selection bias in the WTP estimates for the FS GM trait. The model assumes a bivariate normal distribution for the normalized errors in the two equations forming the model. Mathematically, sample selection is present if th ere is correlation betwee n these error terms (i.e., 0 ). Results from both a Likelihood Ratio Test an d a t-test suggest no evidence of sample selection bias in our data. However, as was the case with our WTP estimates, being a nonGM specialist also played an interesting role in our analysis of sample selection. NonGM specialists were less likely to participate in the hypothetical market presente d in the Contingent Valuation exercise. This finding does not necessarily contradict our assertion of no sample selection since nonGM farmers who chose not to participate in the hypothetical market may have done so simply because of their unfamiliarity regardi ng the complexity associated with GM crop

PAGE 166

166 production. For example, the non-responding nonG M farmers may have been unaware (or improperly informed) of all the possible tran saction costs and produc tion requirements (e.g., fees, refuge practices) associated with GM crops and may have found themselves unable to construct their valuations. Two important notes should be made about the estimation and inte rpretation of these results. First, as pointed by St razzerra et al. (2003), the likel ihood function of sample-selection models may behave in an irregu lar manner, but is usually well behaved for fixed values of Convergence of our sample selection model was ach ieved in three steps. In the first step, the model was allowed to iterate but the estimate of was fixed at a value of zero. This first step obtained the estimates for the univariate models (Table 5-4). The second step used the univariate estimates as initial values in the sample sel ection model but restricted them so that only could vary in each iteration. The last step allowe d all parameter estimates to vary freely. Second, while the fact that our sample sele ction model was empirically estimable is a good indication; optimally, we would have liked a larger sample size. In our interpretation of the DBDC sample-selection model results we have assumed that sample size and variation have been sufficient, but we cannot provide definitive evidence that this is true given the complex nature of the model. In fact, an interesting research que stion for future studies could be determining appropriate sample sizes for the DB-DC Samp le Selection model by use of Monte Carlo methods, for example. While the prospect of higher yields and hi gher profitability remains one of the most important factors determining a doption (Chapter 4), another im portant factor that should influence adoption is the prospect of more consistent yields and profits (i.e., reducing uncertainty) (Chapter 6). Nevertheless, the unc ertainty inherent in GM crop adoption has

PAGE 167

167 traditionally been oversimplified and sometimes even neglected. In Chapter 6 we presented an analysis of the GM crop adoption scenario faced by the farmer while emphasizing the uncertainty in agricultural pr oduction throughout the discussion We treated uncertainty by framing it under the state contingent production model of Chambers and Quiggin (2000). We also identified two main strategies that farm ers use to obtain desire d outcomes with more predictability: (i) taking actions to control and modify the e nvironment surrounding the plant, and (ii) taking actions to contro l and modify the metabolism and genetic material of the plant itself. Some interesting conclusions fr om Chapter 6 are worth mentioning. For example, as discussed earlier, the ex anteness of seed traits seems to be more pervasive than the ex anteness of other agricultural inputs. For other ag ricultural inputs, the farmer may eliminate uncertainty and the need to act ex ante (i.e., before nature reveals its state of nature) by perfectly monitoring and perfectly supplementing f ound deficiencies on a real-time basis although he would incur in potentially large co sts. The type of t echnologies the farmer needs to achieve this feat, are those affine to the goals of precision agriculture. On the other hand, with seed traits, the farmer necessarily acts before nature because the seed decision is made once at the beginning of the growing seas on. The farmer may buy a drought resistant seed but he must make this decision long before know ing if the drought will ac tually happen; in the meanwhile, irrigation (if av ailable) may be adjusted on the go as drought severity varies during the growing season. Another point we made was that some GM tr aits are better descri bed as state-general inputs while others are best desc ribed as state-specific. Whether an input classifies as stategeneral or state-specific depends on which states of nature are c onsidered and on the number of states in which the input influences output. For a good analysis, the states of nature considered

PAGE 168

168 should exclude those that are not plausible. State-general inputs, since they obtain similar yields under a larger number of states of nature, are less risky than statespecific inputs. As for the GM traits currently in the market, the glyphosate tole rant, for example, is state-general while a Bt trait is more of a state-specific input. Also, a st acked Bt trait would be mo re state-general than a single Bt trait. The DT and FS traits, as defined in our CV questionnaires, would be more correctly classified as state-specif ic and state-general, respectively. We defined a perfectly-state-general input as on e that obtains the same yield (all else fixed) in all plausible states of nature. While not very realistic, this type of i nput helped us visualize that even as an ex ante measure such seed trait would eliminat e uncertainty in a similar way that precision agriculture aims to do. The conclusion was that making such trait available to farmers would substantially modify the types of data sets that farmers need to and actually gather. For example, a trait that completely protected the plant against all plausible levels of pest pressure in a given region would eliminate the farmers ne ed to monitor pest populations. As such, we argued that GM seed traits are somewhat subst itutes for precision agriculture methods. It makes sense that the two strategies used to reach the same goal of reducing uncer tainty (i.e., modifying the plant or modifying the surroundings of the plant) are substitutes to each other. As a final note, while biotech crops bring with them important benef its, some argue they also bring potential risks. Most proponents of biotech crops cite enhanced crop yields, more environmentally friendly food production, and mo re nutritious food as the major potential benefits. Skeptics cite uncertain effects on the environment, uncertain long-term effects on human health, ethical concerns, and distribution of benefits across developed and developing nations as the major potential risk s. The uncertainty regarding the long term net effects (benefits vs. risks) of the technology has implications on all levels of the s eed (or crop) industry

PAGE 169

169 consumers, farmers, seed companies and polic y makers. Whether advocates or skeptics, all agents across the industry are to benefit from being better informed (less uncertain) about the different aspects of emerging biotech seed products. This study and the results obtained will likely be of high value to the U.S. agricultural sector. Understanding th e factors influencing producers WT P presents a great opportunity for improved efficiency in the market via appropriate pricing strategies and market segmentation by seed companies. Also, farmers are bound to be better serviced if be tter understood by the companies serving them. Finally, premarket studies like this one provide valuable information to policy makers; WTP elicited directly from farmers provides a benchmark for future monitoring of overly monopolisti c pricing and for future st udies on GM crop adoption.

PAGE 170

170 APPENDIX A GM CROP REGULATION Two contrasting types of regulatory sc hem es exist worldwide among governmental agencies and other stakeholders when it comes to GM foods (James and Krattiger 1996). The first scheme considers GM foods are like any ot her food and therefore same regulations apply. This view sees biotech engineering as a part of a natural continuum fo llowing selective breeding techniques and believes current sc reening and testing procedures used on conventional foods are sufficient. The second scheme views GM foods as intrinsically different from conventional foods and thus argues that separate laws and regulations are needed. Despite opposing views, there is widespread agreement on the use and need of regulation of some kind. Existing regulations for GM crops/foods may be classified into two types (Dale 1995): (i) Under contained conditions (i.e., labo ratory procedures and health sa fety of workers); and (ii) On field trials (i.e., assessments of risk to, or likel y impact on, human health and the environment) Why regulate? The justification is largely ba sed on unfamiliarity (Dale 1995; OECD 1993; James and Krattiger 1996). Plant selection, of one or another kind, has been around for some 10,000 years. We have practiced conventional selective breeding techniques for most of the 20th century and became familiarized with its product s. This is not to say we understand all the possible outcomes of sexual genetic recombinat ion that happens in conventional selective breeding. For example, the release of new pot ato varieties obtained vi a conventional breeding still requires prior analysis for high levels of toxi c substances (glycoalkaloids). This toxicity as a consequence of conventional breedi ng may be perceived as a danger ous one, but at this point has become familiar to us giving us the ability to prevent or manage possible damages (safety or risk management). As mentioned earlier, genetic engineering techniques expand the possibilities of gene tr ansfer by allowing for transference between unrelated species (genes

PAGE 171

171 outside the normal sexual gene pool) and even pe rmit the introduction of synthetic genes. The unfamiliarity associated with GM crops comes from the increase in the number and diversity of traits that can be possibly transferred, and thus, the increased possible unfamiliar outcomes that might follow.1 In consequence, risk assessments of potential unfamiliar outcomes must precede any release of new GM crops into the environment, or their use as food, feed, fiber, or other. Such risk assessments warrant oversight and approval from regulatory authorities. Governments worldwide have adopted one of two types of regulatory agendas (Dale 1995): vertical or horizontal. The U.S. and Canada have adopted vertical regulation which treats new crop varieties on a case-by-case basis defining characteristics of crops that require them to be regulated; with no a priori requirement that all GM crops/foods be regulated. The European Union (EU) takes a horizontal ap proach requiring all transgenic f oods to be regulated (James and Krattiger 1996). In the U.S., both experimentation with GM crops/foods and their approval for commercialization are vertically regulated by government agencies. The main roles played by regulations are to confirm perfor mance, evaluate characteristics of food, evaluate risk to human health (allergies) and, evaluate environmental effects. Elements presented in risk assessments include (OECD 1993; Dale 1995): ch aracterization of the function played by the gene in donor organism and effect on target organism; evidence of toxicity; persistence in natural habitats (weediness); impact on non-target organism s (unintended effects); and likelihood and consequences of undesired gene transfer to other cultivars or wild/weedy species. 1 Some argue that the increase in precision obtained with genetic engineering results in more thoroughly characterized and potentially more predictable organisms (OECD 1993) compared to conventional breeding where genes with unknown and possibly undesirable functions can tag along with the desired trait. For examples of undesired results from conventional breeding see Pauppke (2 001). Others argue that tra it expression is generally governed by a complex interaction among numerous genes and conventional breeding is more likely to pass on all the genes needed for proper expression and metabolic regulation (Palumbi 2001).

PAGE 172

172 Three agencies are responsible for implemen ting a coordinated framework of regulations in the United States (U.S. General Accounti ng Office 2002; James and Kr attiger 1996): (i) the USDA, (ii) the EPA, and the (iii) FDA. Within the USDA, the Animal and Plant Health Inspection Service (A PHIS) is responsible for assessing environmental safety of new GM crops, for issuing field trial permits, and for approving general environmental release.2 If the product contains a modification involving a gene with a pesticide (i.e., resistance to insects, bacteria or viruses) the EPA is also involved in the approval. These genetically in corporated pesticides are subject to EPAs regulations on sale, distribution, and use of such s ubstances. Finally, if the transgenic crop is intended for food or feed use, the FDA is also involved. The FDA has primary authority over safety of most of the food supply in the U.S. The 1938 Federal Food, Drug, and Cosmetic Act (FD&C Act) establishes the FDA as the regulatory agency responsible for protecting the public fr om adulterated and fraudulently labeled foods other than those regulated by USDAs Food Safety and Inspection Service (FSIS).3 Section 402 of the FD&C Act defines an adul terated food as a food which: bears or contains any poisonous or deleterious substance whic h may render it injurious to health; but in case th e substance is not an added s ubstance such food shall not be considered adulterated under this clause if the quantity of such substance in such food does not ordinarily render it in jurious to health [] FDAs view and policy with respect to GM foods was established in its 1992 Policy on Foods Derived from New Plant Varieties policy statement published in the Federal Register in 2 The familiarity concept proposed by the OECD (1993) has been adopted in the U.S. regulatory framework so that certain crops that have become sufficiently familiar and have been recognized as low risk; qualify for a simplified notification system prior to release. Under such notification system, the proposer si mply notifies USDA-APHIS of an intention to release (to the environment) a new transgenic variety and can proceed to do so if no response is received from APHIS within 30 days. 3 The FSIS is responsible for ensuring that meat, poultry, and egg products are safe, wholesome, and correctly labeled.

PAGE 173

173 May of that year (Food and Drug Administra tion 1992). The policy statement explains how foods are regulated under the FD&C Act; it applies equally a nd indiscriminately to foods (including animal feeds) derived from plants modified through all methods of breeding, including genetic engineering (F ood and Drug Administration 1995). The 1992 Policy Statement recommended compan ies with GM food under development to consult with FDA. Even though not required, mo st companies voluntarily complied. In June 1996, the FDA further provided the industry with Consultation Guidelines to streamline the consultation process (Food and Dr ug Administration 1997). Under su ch guidelines, the company meets with FDA and provides a summary of scie ntific and regulatory as sessment of the food. The FDA then evaluates and responds to the submission by letter. In January 2001, the FDA issues a proposed rule published in the Federal Register that would require developers to submit their scientific and regulatory assessments to th e FDA 120 days before the GM food is marketed (Food and Drug Administration 2001). The propose d rule recommended consultation practices should be continued before submission of assessments. Highlights from the 1992 Policy on Foods Derived fr om New Plant Varieties policy statement are presented below (Food a nd Drug Administration 1995):4 Genetic Modification the introduced genetic mate rial should be sufficiently characterized so that it does not encode harmful substances. It should also show a stable insertion in the plant genome to minimize potential future undesired genetic rearrangements. Toxicants Many existing plants are known to produce toxicants or anti-nutritional factors. Many of these factors are found in foods at levels which do not cause acute toxicity or do not affect humans. FDAs 1992 policy statement indicates new varieties 4 FDA's review of the first GM comm ercialized in the U.S. (FLAVR SAVRTM) was conducted consistent with the May 29, 1992 policy statement; showing how the agency inte rprets the Federal Food, Drug, and Cosmetic Act with respect to foods derived from new plant varieties obt ain by genetic engineering methods (Food and Drug Administration 1994a).

PAGE 174

174 should not contain levels of such toxicants that exceed levels found in current existing varieties. Nutrients Genetic modifications mi ght unintentionally (and in some cases undesirably) affect nutrient content or bioavailability of nutrients. Nutrient content of new/modified varieties should not be reduced in co mparison with existin g/conventional foods. New Substance In some cases, genetic modification may encode substances that are substantially different from those originally found in food. In some of such cases, premarket approval as food additives will be requ ired for these substances; in other cases, proper labeling will be sufficient. Allergenicity Genetic modification may result in introduction of genetic material that produce allergic reactions in pe ople, especially when the don or organism is known to be commonly allergic. FDA believes that prot eins derived from commonly allergenic sources should be presumed to be allerg ens and special labeli ng would be required, unless scientific evidence demonstrates otherwise. Antibiotic Resistance Markers M-genes (discussed earlier in this chapter) form part of the DNA cocktail used in the genetic e ngineering process. Both the (i) gene responsible for the desired tra it and (ii) a marker gene which provides the modified plant cell with resistance to a given antibiotic; are jointly transferred to the target organism. Successful transference is not guaranteed, so engineers treat all resulting cells with antibiotics. Only antibiotic-resistant cells (those that were successfully modified) survive. Once this selection process is finished, the marker gene is no longer needed but remains as a DNA residue from the transformation process and keeps producing the protein responsible for inactivation of the antibiotic subs tance (i.e., antibiotic -resistance). The most commonly used marker traits are for resistance to antibiotics kanamycin and neomycin. The use of marker genes raises co ncerns about possible inactivation of oral doses of antibiotic due to human consump tion of GM foods containing marker genes. Some questions have also been raised a bout possible transferen ce of the antibioticresistance gene to pathogenic microbes in the human gastrointestinal tract. Whereas the FDA states that there are no known mechanisms by which a gene can be transferred from a plant to a microbe (Food and Drug Ad ministration 1995), more recent evidence (Netherwood et al. 2004) shows existence of such mechanis ms. The FDA also found that kanamycin and neomycin have very limited use as oral antibiotics and concentrations of antibiotic-resistant proteins in GM foods are too little to degrade a significant amount of antibiotic. Animal Feeds Under the FD&C Act, feeds grown for animals raised as human food sources must meet same safety standards as human food. Some additional points to consider when performing risk assessment on ne w animal feeds are that : (i) in contrast to the human diet which consists of many plants, some animal diets may consist of a single plant; (ii) animals consume parts of the plant which are not consumed by humans; and (iii) nutritional composition of the plant is essential for efficient production and profitability.

PAGE 175

175 Labeling The FD&C Act requires labeling to be truthful and not misleading. The 1992 Policy Statement makes labeling of GM foods /crops voluntary unless the composition of the GM food differs significantly from its c onventional counterpart. For example, if a modified food contains a potential allergen which is not expected in that food, the consumer must be informed in the label. If the allergen has potentia lly serious associated effects, then the FDA evaluates if labeling is sufficient for consumer protection. One issue causing controversy (even though no food of this type has been produced yet), is whether a plant which has been conferred an animal gene must be labeled such that people with different ethical views are inform ed (e.g., vegetarians, specific religions).

PAGE 176

176 APPENDIX B RESULTS FOR AUXILIARY ESTIMATIONS Table B-1. Estim ation results for drought tolerance tra it (insured only) Drought nGM Drought GM Estimate p-value Estimate p-value constant 11.359 (0.320) 4.623 (0.664) f_age -0.014 (0.924) 0.043 (0.763) f_educ -0.193 (0.885) -0.211 (0.867) f_nfjob 1.674 (0.598) 0.312 (0.919) f_income 1.628** (0.017) 2.174*** (0.001) earlyad 10.099** (0.046) 13.636** (0.013) farmsize -0.004 (0.244) -0.002 (0.520) nonGM100p 8.513* (0.059) -3.234 (0.433) insurcost 0.227* (0.078) 0.265** (0.046) mD2 0.369** (0.025) 0.271** (0.039) sigma 10.127*** (0.000) 9.669*** (0.000) N 81 77 Log-likelihood -95.284 -91.192 Mean WTP i 22.657*** (15.895) 20.283*** (15.139) C.I. 95 L 19.863 17.657 C.I. 95 U 25.451 22.909 LR Index 0.63 0.61 LR statistic ii 85.216*** (0.000) 79.781*** (0.000) i For mean WTP the t-statistic is reported in parent hesis. Standard error for mean WTP was calculated using the Delta Method. ii LR statistic is for likelihood ratio test with 0... :21 oH, sigma is left unconstrained. *** indicates significance at 0.01 leve l, ** indicates significance level of 0.05, and indicates significance level of 0.1.

PAGE 177

177 LIST OF REFERENCES Adam owicz, W., J. Louviere, and M. Williams 1994. Comparing Revealed and Stated Preference Methods for Valuing Environmental Amenities.Journal of Environmental Economics and Management 26:271. Alberini, A., B.J. Kanninen, and R.T. Carson. 1997. Modeling Response Incentives in Dichotomous Choice Conti ngent Valuation Data. Land Economics 73:309. Aldridge, S. 1994. Ethically Sens itive Genes and the Consumer. Trends in Biotechnology 12:71. Alexander, C., J. Fernandez-Cornejo, a nd R. E. Goodhue. 2003a. Farmers Adoption of Genetically Modified Varietie s with Input Traits. Giannini Foundation Research Report 347, University of California, Davis. 2003b. Effects of the GM Controversy on Iowa Corn-Soybean FarmersAcreage Allocation Decisions. Journal of Agricultural and Resource Economics 28:580. Alexander, C. and T. Van Mellor. 2005. Det erminants of Corn Rootworm Resistant Corn Adoption in Indiana. AgBioForum 8:197. Allenby, G.M., N. Arora, and J.L. Ginter. 1995. I ncorporating Prior Knowledge Into the Design and Analysis of Conjoint Studies. Journal of Marketing Research 32:152. lvarez-Farizo, B., N. Hanley, and R. Wright. 1996. Sample Selection Bias and the Estimation of Bid Curves in Open-Ended Contingent Valuation Studies. Di scussion Paper in Ecological Economics No 96/5, Department of Economics, University of Stirling. Amemiya, T. 1984. Tobit Models: A Survey. Journal of Econometrics 24:3. Argenbio. 2006. Los Cultivos Transgenicos en el Mundo. Argenbio website. Available at http://www.argenbio.org/h/biotecnologia/16.php accessed May 2009. Arrow, K. 1953. Le Des Valeurs Boursiers Pour la Repartition la Meillure Des Risques. Cahiers du Seminair dEconomie, Centre Nationa le de la Recherche Scientifique (CNRS), Paris. Arrow, K. 1964. The Role of Securities in the Optimal Allocation of Risk-Bearing. Review of Economic Studies 31:91. Arrow, K., R. Solow, P. Portney, E. Leamer, R. Radner, and H. Shuman. 1993. Report of the National Oceanic and Atmospheric Administ ration Panel on Conti ngent Valuation. Federal Register 58:4601. Azevedo, C. D., J.A. Herriges, and C.L. Kling. 2003. Combining Revealed and Stated Preferences: Consistency Tests and Their Interpretations. American Journal of Agricultural Economics 85:525.

PAGE 178

178 Bateman, I.J., R.T. Carson, B. Day, M. Hane mann, N. Hanley, T. Hett, M. Jones-Lee, G. Loomes, S. Mourato, E. zdemiroglu, D.W. Pearce, R. Sugden, and J. Swanson. 2002. Economic Valuation with Stated Preference Techniques. Northampton MA: Edward Elgar. Bates, G. 2009. Corn Silage. Report SP434-D, Ag ricultural Extension Service, University of Tennessee. Available at www.utextension.utk.edu/publ ications/spfiles/sp434d.pdf accessed Jan uary 2009. Bergstrom T. C. and R.P. Goodman. 1973. Private Demands for Public Goods. The American Economic Review 63(3):280-296. Bishop, R.C. and T. A. Heberlein. 1979. Measuring Values of Extra Market Goods. American Journal of Agricultural Economics 61: 926. Black, D. 1948. On the Rationale of Group Decision-making. The Journal of Political Economy 56: 23. Borcherding, E. and R.T. Deacon. 1972. The Demand for the Services of Non-Federal Governments. The American Economic Review 62(5):891-901. Borlaug, N. 1970. Nobel Lecture: The Green Revolution, Peace, and Humanity. In F. W. Haberman, ed. Nobel Lectures, Peace 1951-1970. Amsterdam: Elsevier Publishing Company. Available at http://nobelprize.org/nobel_pri zes/peace/laureates/1970/borlauglectu re.html. Bowen, H.R. 1943. The Interpretation of Voti ng in the Allocation of Economic Resources. Quarterly Journal of Economics 58:27. Cameron, T.A. 1992. Combining Contingent Valuat ion and Travel Cost Data for the Valuation of Nonmarket Goods. Land Economics 68:302. Cameron, T.A. and M.D. James. 1987a. Effici ent Estimation Methods for Closed-Ended Contingent Valuation Surveys. Review of Economics and Statistics 69:269. 1987b. Estimating Willingness to Pay from Surv ey Data: An Alternative Pre-Test Market Evaluation Procedure. Journal of Marketing Research 24: 389. Cameron, T.A. and J. Quiggin. 1994. Estima tion Using Contingent Va luation Data from a Dichotomous Choice with Follow-up Questionnaire.Journal of Environmental Economics and Management 27: 218. Carpenter, J. and L. Gianessi. 2001. Agricultu ral Biotechnology: Updated Benefit Estimates. Washington DC: National Center fo r Food and Agricultural Policy. Carson, R.T. 1985. Three Essays on Contingent Valuation. Ph.D. dissert ation, University of California, Berkeley.

PAGE 179

179 Carson, R.T. and W.M. Hanemann. 2005. Continge nt Valuation. In Karl-Goran Mler and Jeffrey Vincent, eds. Handbook of Environmental Economics: Volume 2. New York: Elsevier. Center for Disease Control. 2001. Investigati on of Human Health Eff ects Associated with Potential Exposure to Genetica lly Modified Corn. A repor t to the U.S. Food and Drug Administration, Atlanta GA. Chambers, R.G. and J. Quiggin. 2000. Uncertainty, Production, Choice and Agency: The StateContingent Approach. New York: Cambridge University Press. Chen, S. and M. Ravallion. 2008. The Developi ng World is Poorer than We Thought, But No Less Successful in the Fight Against Povert y. Working paper No. WPS4703, World Bank Policy Research. CIA. 2008. CIA World Fact Book Database. Washington DC. Available at https://www.cia.gov/library/publications/the-world-factbook/index.htm l accessed April 2009. Ciriacy-Wantrup, S.V. 1947. Capital Retu rns from Soil-Conservation Practices. Journal of Farm Economics 29:1181. 1952. Resource Conservation: Economics and Policies. Berkley CA: University of California Press. Clark, G. 2007. A Farewell to Alms: A Brief Economic History of the World. New Jersey: Princeton University Press. Clawson, M. and J. Knetsch. 1966. Economics of Outdoor Recreation. Baltimore MD: Johns Hopkins University Press. Conley, T. G. and C. R. Udry. 2007. Learni ng About a New Technology: Pinapple in Ghana. Unpublished manuscript, Yale University. Cooper, J.C. 1997. Combining Actual and Cont ingent Behavior Data to Model Farmer Adoption of Water Quality Protection Practices. Journal of Agricultural and Resource Economics 22:30. Cooper, J. and W.M. Hanemann. 1995. Refere ndum Contingent Valuation: How Many Bounds Are Enough? Working paper, USDA Economic Research Service, Food and Consumer Economics Division. Daberkow, S.D. and W.D. McBride. 2003. Far m and Operator Charact eristics Affecting the Awareness and Adoption of Precision Ag riculture Technologies in the U.S. Precision Agriculture 4:163. Dale, P. 1995. R&D Regulation and Field Trialling of Transgenic Crops. Trends in Biotechnology 13:398.

PAGE 180

180 Davis, R.K. 1963a. Recreation Planning as an Economic Problem. Natural Resources Journal 3: 239. 1963b. The Value of Outdoor Recreation: an Economic Study of the Maine Woods. Ph.D. dissertation, Harvard University. Dawe, D., R. Robertson and L. Unnevehr. 2002. Golden Rice, What Role Could It Play in Alleviation of Vitamin A Deficiency? Food Policy 27:541. Debreu, G. 1952. A Social Equilibrium Existence Theorem. Proceedings of the National Academy of Science of the USA 38:886. Department of Soil and Crop Sciences. 2004. T ransgenic Crops. Colorado State University. Available at http://www.cls.casa.colostate.edu/TransgenicCrops/how.htm l accessed April 2009. DeShazo, J.R. 2002. Designing Transactions W ithout Framing Effects in Iterative Question Formats. Journal of Environmen tal Economics and Management 43:360. Diamond, J. 1999. Guns, Germs and Steel: The Fates of Human Societies. New York: W.W. Norton & Company. Dubin, J.A. and D. Rivers. 1989. Selection Bi as in Linear Regression, Logit and Probit Models. Sociological Methods and Research 18:360. Duflo, E., M. Kremer, and J. Robinson 2006. U nderstanding Technology A doption: Fertilizer in Western Kenya. Evidence from field experiments. Unpublished manuscript. Available at www.povertyactionlab.com/papers/kenyafield.pdf Dupuit, J. 1844. On the Measurem ent of the Utility of Public W o rks. Reprinted in K.J. Arrow and T. Scitovsky, eds. Readings in Welfare Economics. Homewood IL: Richard D. Irwin and Nobleton, 1969. Edwards, F.E. and G.D. Anderson. 1987. Overlooke d Biases in Contingent Valuation Surveys: Some Considerations. Land Economics 63:168. Eklf, J. and S. Karlsson. 1999. Testing and Corr ecting for Sample Selection Bias in Discrete Choice Contingent Valuation Studies. Working paper No. 171, Stockholm School of Economics, Sweden. Feder, G., R. J. Just, and D. Zilberman. 1985. Adoption of Agricultural Innovations in Developing Countries: A Survey. Economic Development and Cultural Change 33:255 98. Feitelson, J.S. 1993. The Bacillus thuringiensis Family Tree. In L. Kim, ed. Advanced engineered pesticides. New York: Marcel Dekker, pp. 63.

PAGE 181

181 Fernandez-Cornejo, J. and W.D. McBride. 2002. Adoption of Bioengineered Crops. Washington DC: U.S. Department of Agri culture, Economic Research Service, Agricultural Economic Report AER810, May. Fernandez-Cornejo, J., C. Alexander, a nd R.E. Goodhue. 2002. Dynamic Diffusion with Disadoption, The case of Crop Biotechnology in the USA. Agricultural and Resource Economics Review 31:112. Fernandez-Cornejo J., C. Hendricks, and A. Mishra. 2005. Technology Adoption and Off-Farm Household Income. Journal of Agricultura l and Applied Economics 37:549. Flachaire, E. and G. Hollard. 2005. Controll ing Starting-Point Bias in Double Bounded Contingent Valuation Surveys. Work ing paper, Universite Paris. Food and Drug Administration. 1994a. Biotechnology of Food. FDA website. Available at http://www.cfsan.fda.gov/~lrd/biotechn.html accessed April 2009. 1995. FDA'S Policy for Foods Developed by Biotechnology, Em erging Technologies Biotechnology. FDA website. Available at http://www.cfsan.fda.gov/~lrd/bioeme.html accessed April 2009. 1994b. First Biotech Tom ato Marketed. FDA website. Available at http://www.cfsan.fda.gov/~lrd/biotech.html accessed April 2009. 1997. Guidance on Consultation Procedures. FDA website. Available at http://www.cfsan.fda.gov/~lrd/consulpr.html accessed May 2009. 2001. Prem arket Notice Concerning Bioengineered Foods. Federal Register 66:4706. 1992. Statement of Policy: Foods Derived from New Plant Varieties. Federal Register 57:22984. Freeman.A.M. 1993. The Measurement of Environment and Resource Values: Theory and Method. Washington DC: Resources for the Future. Gianessi, L., C.S. Silvers, S. Sankula, and J. Carpenter. 2002. Plant Biotechnology: Current and Potential Impact for Improving Pest Ma nagement in U.S. Agriculture. Washington DC: National Center for F ood and Agricultural Policy. Graham, D.A. 1981. Cost-Benefit Analysis Under Uncertainty." The American Economic Review 71:715. Gray, M. and K. Steffey. 1999. European Corn Bo rer Population in Illinoi s Near Historic Low. Pest Management and Crop Development Bulleti n No. 24, Extension Service, University of Illinois. Green P. and V.R. Rao.1971. Conjoint Meas urement for Quantifying Judgmental Data. Journal of Marketing Research 8:355.

PAGE 182

182 Griliches, Z. 1957. Hybrid Corn: An Exploration in the Econom ics of Technological Change. Econometrica 25:501. Haavelmo, T. 1944. The Probability Approach in Econometrics.New Haven CT: Cowles Foundation for Research in Economics. Hanemann, W.M. 1978. A Methodological and Em pirical Study of the Recreation Benefits from Water Quality Improvement. Ph.D. dissertation, Harvard University. 1985. Some Issues in Continuousand Disc rete-Response Contingent Valuation Studies. Northeastern Journal of Agricultural Economics 14:5. 1984. Welfare Evaluations in Continge nt Valuation Experiments with Discrete Responses.American Journal of Agricultural Economics 66:332. 1991. Willingness to Pay and Willingness to Accept: How Much Can They Differ? The American Economic Review 81(3): 635-647. Hanemann, W.M., J.B. Loomis, and B.J. Kannine n. 1991. Statistical Efficiency of DoubleBounded Dichotomous Choice Contingent Valuation. American Journal of Agricultural Economics 73:1255. Hausman, J.A., ed. 1993. Contingent Valuation: A Critical Assessment. Amsterdam: NorthHolland. Heckman, J. 1979. Sample Selectio n Bias as a Specification Error. Econometrica 47:153. Herriges, J.A. and J.F. Shogren. 1996. Startin g Point Bias in Dichot omous Choice Valuation with Follow-up Questioning. Journal of Environmental Economics and Management 30:112. Hicks, J. 1956. A Revision of Demand Theory. Oxford: Clarendon Press. 1941. The Rehabilitation of Consumers Surplus. Review of Economic Studies 8:108. 1943. The Four Consumer Surpluses. Review of Economic Studies 11:31. Hfte, H. and H.R. Whiteley. 1989. Insecticidal Crystal Proteins in Bacillus thuringiensis. Microbiology Review 53:242. Huang, W., W. McBride, and U. Vasavada. 2009. Recent Volatility in U. S. Fertilizer Prices: Causes and Consequences. Amber Waves 7(1):28 31. Hubbell, B.J., M.C. Marra, and G.A. Carlson. 2000. Estimating the Demand for a New Technology: Bt Cotton and Insecticide Policies. American Journal of Agricultural Economics 82:118. James, C. 2006. Global Status of Commercialized Biotech/GM Crops: 2006. Brief No. 35. Ithaca, NY: International Service for the Acquisition of Agri-biotech Applications.

PAGE 183

183 2008. Global Status of Commercialized Biotech/GM Crops: 2008. Brief No. 39. Ithaca, NY: International Service for the Acquisition of Agri-Biotech Applications. James, C. and A.F. Krattiger. 1996. Global Review of the Field Testi ng and Commercialization of Transgenic Plants, 1986 to 1995: Th e First Decade of Crop Biotechnology. ISAAA Brief No. 1. Ithaca, NY: International Service for the Acquisition of Agri-biotech Applications. Johnson, R.M. 1974. Trade-Off Analysis of Consumer Values. Journal of Marketing Research 11:121. Just, R.E., D.L. Hueth, and A. Schmitz. 1982. Applied Welfare Econom ics and Public Policy. Englewood Cliffs NJ: Prentice-Hall. 2004. The Welfare Economics of Public Policy. Northampton MA: Edward Elgar. Khanna, M. and D. Zilberman. 1997. Incentiv es, Precision Technology, and Environmental Protection. Ecological Economics 23:25. Knudson, M.K. 1991. Incorporating Technol ogical Change in Diffusion Models. American Journal of Agricultural Economics 3:724. Krattiger, A.F. 1997. Insect Resistance in Crops: A Case Study of Bacillus thuringiensis (Bt) and its Transfer to Developing Countries. Brief No. 2. Ithaca, NY: International Service for the Acquisition of Agri-biotech Applications. Krutilla, J. 1967. Conservation Reconsidered. The American Economic Review 57(4):777-786. Loomis, J.B. 1987. Expanding Contingent Valu e Sample Estimates to Aggregate Benefit Estimates: Current Practice and Proposed Solutions. Land Economics 63:396. Luce, D. 1959. Individual Choice Behavior. New York: John Wiley and Sons. Mler, K-G. 1974. Environmental Economics: A Theoretical Inquiry. Baltimore: Johns Hopkins University Press. Malthus, T. 1798. An Essay on the Principle of Population. London. Available at http://www.econlib.org/library/Malthus/malPop.html Marschak, J. 1960. Binary Choice C onstraints on Random Utility Indications In K. Arrow, ed. Stanford Symposium on Mathematical Methods in the Social Sciences. Stanford CA: Stanford University Press. Marshall, A.1930. Principles of Economics. London: Macmillan and Co. Martineau, B. 2001. First Fruit: The Creation of the Flavr Savr Tomato and the Birth of Biotech Foods. New York: McGraw-Hill.

PAGE 184

184 McConnell, K.E. 1990. Models for Referendum Data : the Structure of Di screte Choice Models for Contingent Valuation. Journal of Environmental Economics and Management 18:19 34. McFadden, D. 1974. Conditional Logit Analysis of Qualitative Choice Behavior. In P. Zarembka, ed. Frontiers in Econometrics. New York: Academic Press. Meng, C.-L. and P. Schmidt. 1985. On the Cost of Partial Observability in the Bivariate Probit Model. International Economic Review 26:71. Mishra, A.K., H.S. EI-Osta, MJ. Morehart, J. D. Johnson, and J.W. Hopkins. 2002. Income, Wealth, and the Economic Well-Being of Fa rm Households. Washington DC: U.S. Department of Agriculture, Economic Resear ch Service, Agricult ural Economic Report 812. Mitchell, R.C. and R.T. Carson. 1989. Using Surveys to Value Public Goods: The Contingent Valuation Method. Baltimore MD: Johns Hopki ns University Press. Monroe, K. 2003. Pricing: Making Profitable Decisions, 3rd. ed. New York: McGraw-Hill. Monsanto. 2005. Borlaug Notes Proven Success. Available at http://videos.howstuffworks.com/monsanto/ 511-borlaug-notes-prove n-success-video.htm accessed May 2009. 2009. Com pany History. Monsanto website. Available at http://www.monsanto.com/who_we_are/history.asp accessed April 2009. National Drought Mitigation Center, The Drought Monitor. 2006. Drought Monitor Archives Database. Drought Monitor Website. Available at http://drought.unl.e du/dm /archive.html accessed May 2009. Netherwood, T. S.M. Martin-Orue, A.G. ODonnell, S. Gockling, J. Graham, J.C. Mathers, and H.J. Gilbert. 2004. Assessing the Survival of Transgenic Planic Plant DNA in the Human Gastrointestinal Tract. Nature Biotechnology 22:204. Oates, W. 1994. Comments on Estimating th e Demand for Public Goods: The Collective Choice and Contingent Valuation Approach es. Paper presented at the DOE/EPA Workshop on Using Contingent Valuation to Measure Non-Market Values, Hemdon, VA. OECD. 1993. Safety Considerations for Biotech nology: Scale-up of Crop Plants. Group of National Experts on Safety in Biotechnology, Paris. Orme, B. 2005. Getting Started with Conj oint Analysis: Strategi es for Product Design and Pricing Research. Madison WI: Research Publishers LLC. Palumbi, S.R. 2001. High Stakes Battle over Brute-Force Ge netic Engineering. The Chronicle of Higher Education, April 13.

PAGE 185

185 Payne, J., J. Fernandez-Cornejo, and S. Dabe rkow. 2002. Factors Affecting the Likelihood of Corn Rootworm Bt Seed Adoption. AgBioForum 6:79. Pearce, D. and E. zdemiroglu. 2002. Economic Valuation with Stat ed Preference Techniques: Summary Guide. London: Queens Printer. Pew Initiative on Food and Biotechnology. 2004. Feeding the World: A Look at Biotechnology and World Hunger. Agricultural Biotec hnology Report, Washington DC. Available at http://www.pewtrusts.org/uploadedFiles/wwwpewtrustsorg/Reports/Food_and_Biotechnol ogy/pew_agbiotech_feed_world_030304.pdf accessed May 2009. 2001. Harvest on the Horizon: Future Us es of Agricultural Biotechnology. Agricultural Biotechnology Report, Wash ington DC. Available at http://www.pewtrusts.org/uploadedFiles/wwwpewtrustsorg/Reports/Food_and_Biotechnol ogy/hhs_biotech_harvest_report.pdf accessed May 2009. Pueppke, S.G. 2001. Agricultural B iotechnology and Plant Improvement. American Behavorial Scientist 44: 1233. Qaim, M. and A. DeJanvry. 2003. Genetically M odified Crops, Corporat e Pricing Strategies, and Farmers Adoption: the Case of Bt Cotton in Argentina. American Journal of Agricultural Economics 85:814. Qaim, M., A. Subramanian, G. Naik, and D. Zilberman. 2006. Adoption of Bt Cotton and Impact Variability: Insights from India. Review of Agricultural Economics 28:48. Quiggin, J. and R.G. Chambers. 2004. D rought Policy: A Graphical Analysis. Australian Journal of Agricultural and Resource Economics 48:225. 2006. The State-Contingent Approach to Production Under Uncertainty. Australian Journal of Agricultural and Resource Economics 50:153. Randall, A. and J. R. Stoll. 1980. Consumer's Surplus in Commodity Space. The American Economic Review 70(3): 449. Randall, A., J. Hoehn, and G. Tolley. 1981. The Structure of Contingent Markets: Some Results of a Recent Experiment. Unpublished manuscript, Department of Agricultural Economics, University of Kentucky. Randall, A., B.C. Ives, and C. Eastman. 1974. Bidding Games for the Valuation of Aesthetic Environmental Improvements. Journal of Environmental Economics and Management 1:132. Rasmussen, S. 2003. Criteria for Optimal Produc tion Under Uncertainty: The State-Contingent Approach. Australian Journal of Agricu ltural and Resource Economics 47:447. Ravallion, R., S. Chen, and P. Sangraula. 2007. New Evidence on the Urbanization of Global Poverty. Working paper No. 4199, World Bank Policy Research.

PAGE 186

186 Rogers, E. 2005. Diffusion of Innovation, 5th ed. New York: Free Press. Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. Journal of Political Economy 82:34. Runge, C.F. and B. Ryan. 2004. The Global Diff usion of Plant Biotechnology: International Adoption and Research in 2004. A repor t prepared for the Council on Biotech Information, Washington, D.C. Available at http://www.apec.umn.edu/faculty/frunge/globalbiotech04.pdf Sachs, J. 2008. Common Wealth: Economics for a Crowded Planet. New York: The Penguin Press. Samuelson, P. 1954. The Pure Th eory of Public Expenditures. Review of Economics and Statistics 36:387. Schattenberg, P. 2009. Nobel Peace Prize Laur eate Dr. Norman Borlaug Calls for Second Green Revolution. AgNews. Available at http://agnews.tamu.edu/showstory.php?id=1054, accessed May 2009. Schelling, T. 1968. The Life You Save May Be Your Own. In S. Chase, ed. Problems in Public Expenditure Analysis. Washington DC: Brookings Institution, pp. 143-144. Schmitz, T., A. Schmitz, and C.B. Moss. 2005. The Economic Impact of StarLink Corn. Agribusiness 21: 391. Sommer, A.1997. Vitamin A Prophylaxis. Archives of Disease in Childhood 77:191. Strazzera, E., R. Scarpa, C. Pinuccia, G.D. Garrod, and K.G. Willis. 2003. Modeling Zero Values and Protest Responses in Contingent Valuation Surveys. Applied Economics 35:133. Sunding, D. and D. Zilberman. 2001. The Ag ricultural Innovation Process: Research and Technology Adoption in a Changing Agricultural Sector. In B. Gardner and G. Rausser, eds. Handbook of Agricultural Economics Volume 1. Amsterdam: Elsevier, pp. 207. Turner, K. 2002. The Place of Economic Values on Environmental Valuation. In I.J. Bateman and K.G. Willis, eds. Valuing Environmental Preference s: Theory and Practice of the Contingent Valuation Method in th e US, EU, and D eveloping Countries. London: Oxford University Press. United Nations. 2008. Millennium Development Goals Website. Available at http://www.un.org/millenniumgoals/bkgd.shtml accessed May 2009. United Nations, Population Division. 2007. World Population Prospects: The 2006 Revision Executive Summary. Rome.

PAGE 187

187 United Nations, World Water Assessment Programme. 2009. Third World Water Development Report: Water in a Changing World. Paris: UNESCO, a nd London: Earthscan. U.S. Department of Agriculture, Risk Management Agency. 2008a. Pilot Biotech Yield Endorsement. Washington DC, January. Available at http://www.rma.usda.gov/pilots/2008/bye.pdf 2008b. Pilot Biotechnology Endorsement. Washington DC, November. Available at http://www.rma.usda.gov/policies/2009/09-be.pdf U.S. Departm ent of Commer ce, National Oceanic and Atmospheric Administration. 1993. Proposed Rules: Natural Resource Damage Assessment. Federal Register 58:4601. U.S. Department of the Interior.1986. Final Rule for Natural Resources Damage Assessments Under the Comprehensive Environmenta l Response Compensation and Liability Act.Federal Register 51:27674. U.S. General Accounting Office. 2002. Genetically Modified Foods: Experts View Regimen of Safety Tests as Adequate, but FDA's Evaluation Process Could Be Enhanced. GAO-02566, Washington DC, May. U.S. Water Resources Council. 1979. Proce dures for Evaluation of National Economic Development: Benefits and Costs in Water Resources Planning (Level C), Final Rule. Federal Register 44:72892. Vakis,R., E. Sadoulet, A.DeJanvry, and C.Cafiero. 2004. Testing fo r Separability in Household Models with Heterogeneous Behavior: A Mi xture Model Approach. Working paper No. 990, Department of Agricult ural and Resource Economics, University of California, Berkeley. van de Ven, W.P.M.M. and B.M.S. van Praag. 1981, The Demand for Deductibles in Private Health Insurance: A Probit Model with Sample Selection. Journal of Econometrics 17:229. Weisbrod, B.A. 1964. Collective Consumption Services of Individual Consumption Goods. Quarterly Journal of Economics 78:471. Whitehead, J.C. 2002. Incentive In compatibility and Starting-Point Bias in Iterative Valuation Questions. Land Economics 78:285. Whitehead, J.C., P.A. Groothuis, and G.C. Blomquist. 1993. Testing for Non-response and Sample Selection Bias in Contingent Valua tion: Analysis of a Combination Phone/Mail Survey. Economic Letters 41:215. Whitehead, J.C., P.A. Groothuis, T.J. Hoban, and W.B. Clifford. 1994. Sample Bias in Contingent Valuation: A Compar ison of the Correction Methods. Leisure Science 16:249.

PAGE 188

188 Willig, R. 1976. Consumer's Surplus Without Apology. American Economic Review 66:589 97. Wooldridge, J. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge MA: MIT Press. World Bank. 2007. Key Development Data & Statistics. Available at http://go.worldbank.org/1SF48T40L0 accessed April 2009. 1993. World Development Report 1993: Investing in Health. Washington DC: Oxford University Press. 2008. World Development Indicators Online. Available at http://web.worldbank.org/WBSITE/EXTER NAL/DATAS TATISTICS/0,,contentMDK:20 398986~menuPK:64133163~pagePK:64133150~piPK:64133175~theSitePK:239419,00.ht ml Wright, D., J. Marrois, J. Rich, and R. Sprenkel. 2008. Field Corn Production Guide. Publication SS-AGR 85, Institute of Food and Agricultural Sciences (IFAS), University of Florida,. Available at http://edis.ifas.ufl.edu/document_ag202 Ye X., S. Al -Babili, A. Klti, J. Zhang, P. Lucca, P. Beyer, and I. Potrykus. 2000. Engineering the Provitamin A (Beta-Carote ne) Biosynthetic Pathway Into (Carotenoid-Free) Rice Endosperm. Science 287:303. Yoo, S.-H. 2007. Estimation of Household Tap Water Demand Function with Correction for Sample Selection Bias. Applied Economic Letters 14: 1079. Yoo, S.-H. and H.-J. Yang. 2001. Application of Sample Selection Model to Double-Bounded Dichotomous Choice Contingent Valuation Studies. Environmental and Resource Economics 20:147. Yoo, S.-H., H.-J. Lim, and S.-J. Kwak. 2009. E stimating the Residential Demand Function for Natural Gas in Seoul with Correct ion for Sample Selection Bias. Applied Energy 86:460 65.

PAGE 189

189 BIOGRAPHICAL SKETCH Paul Esteban Jaram illo Vega was born in Qu ito, Ecuador in 1979. Hi s research interests are applied econometrics, pre-market and mark et valuation of new products, barriers to technological adoption, economic development, and behavioral and expe rimental economics. He received his Associate of Science degr ee in agricultural production in December 2000 from the Escuela Agricola Panamericana El Zamorano in Tegucigalpa, Honduras, where he was awarded the Board of Trustees Scholarship for academic achievement. He graduated from the University of Florida with honors in August 2 002, receiving his Bachelor of Science degree with a specialization in agribus iness, from the Food and Resource Economics Department in the College of Agriculture and Life Sciences. He co ntinued his graduate college education at the University of Florida receiving his Master of Science degree in 2004 and was awarded an Alumni Fellowship to complete his Ph.D. de gree in food and resource economics in 2009. Alumni Fellowships are the highest graduate stud ent award available at University of Florida providing complete funding for four years to promising graduate students. His Ph.D. research focused on corn farmers willingness to pay and adoption of corn seeds containing genetically modified (GM) traits that reduce fertili zation requirements and increase plants tolerance to droughts. He is also involve d in research in Ecuador studying the barriers to adoption of profitable and environmentally friend ly technologies faced by small and low-income rice growers.