ABSTRACT
Title of dissertation: ESSAYS ON QUALITY CERTIFICATION
IN FOOD AND AGRICULTURAL MARKETS
Aaron Ashok Adalja, Doctor of Philosophy, 2017
Dissertation directed by: Professor Se´bastien Houde
Department of Agricultural and Resource Economics
This dissertation features three essays exploring the market impacts of two
types of quality certification—a voluntary non-GMO label and a mandatory food
safety standard. In the first essay, I use a hedonic framework to examine whether
firms use a voluntary quality certification for non-GMO products to extract rent
from customers. Using U.S. retail scanner data coupled with data from a voluntary
non-GMO label, I find no evidence of price premiums or quantity changes for newly
certified non-GMO products. Instead, the label may induce firms to develop new
non-GMO products targeted to high-valuation consumers. The second essay ex-
amines how voluntary non-GMO food labeling impacts demand in the ready-to-eat
[RTE] cereal industry. I estimate a discrete-choice, random coefficients logit demand
model using monthly data for 50 cereal brands across 100 DMAs. Consumer tastes
for the label are widely distributed, and this heterogeneity plays a substantial role
in individual choices; but, on average, the non-GMO label has a positive impact
on demand. I estimate welfare effects by simulating two labeling scenarios: one in
which all brands use the non-GMO label, and one in which no brands use the label.
The simulation results suggest that non-GMO labeling in the RTE cereal industry
may improve consumer welfare on average. In the final essay, we use data from
an original national survey of produce growers to examine whether complying with
the Food Safety Modernization Act’s Produce Rule will be prohibitively costly for
some growers. We examine how food safety measure expenditures required by the
Rule vary with farm size and practices using a double hurdle model to control for
selectivity in using food safety practices and reporting expenditures. Expenditures
per acre decrease with farm size, and growers using sustainable farming practices
spend more than conventional growers on many food safety practices. We use our
estimates to quantify how the cost burden of compliance varies with farm size. We
also explore the policy implications of exemptions to the Rule by simulating how
changes to exemption thresholds might affect the cost burden of each food safety
practice on farms at the threshold.
ESSAYS ON QUALITY CERTIFICATION
IN FOOD AND AGRICULTURAL MARKETS
by
Aaron Ashok Adalja
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2017
Advisory Committee:
Professor Se´bastien Houde, Chair
Professor Erik Lichtenberg
Professor Roberton C. Williams III
Professor James Hanson
Professor Yogesh Joshi
© Copyright by
Aaron Ashok Adalja
2017
Acknowledgments
The level of support I have received during my graduate study at the Univer-
sity of Maryland, both professionally and personally, cannot be overstated. Each
of the members of my dissertation committee has contributed to my success im-
measurably. I thank my advisor, Se´bastien Houde, for his thoughtful guidance and
advice throughout my dissertation work. His patient attitude and encouragement
while I explored new ideas and worked to overcome obstacles has made me a bet-
ter researcher. I am also indebted to Erik Lichtenberg, who has provided me with
invaluable mentorship and research opportunities that have led to a very fruitful
collaboration over the past three years. I thank Jim Hanson for being a wonderful
research collaborator, mentor, and friend throughout my time at Maryland, and I
am grateful to Rob Williams for providing countless suggestions in the early stages
of my research that helped shape the course of my dissertation. Lastly, I would
like to thank Yogesh Joshi for helping me navigate the sea of theoretical work on
product differentiation.
I would also like to express appreciation for my fellow graduate students in
the Department of Agricultural and Resource Economics for their camaraderie and
for their valuable comments on my research during workshops. In particular, my
cohort of classmates—Joe, Patrick, Magda, Sarah, Shirley, and Xiaoya—have been
a great source of friendship and support.
Additionally, I am truly grateful to the AREC department for providing me
with the financial support necessary to conduct my research. The Bruce and Mary
ii
Ann Gardner Dissertation Enhancement Award enabled me to purchase essential
data, without which my dissertation would not have been possible. The Non-GMO
Project also provided me with data vital to my research, for which I thank them.
Lastly, I thank my parents, Varsha and Ashok, as well as my siblings, Anita
and Amesh, for their endless support and encouragement, not only during my gradu-
ate study, but throughout my life. I also thank my wife Aparna for being a constant
source of happiness and humor in all aspects of my life. And, finally, I owe a debt
to my cat Fred for keeping me sane while writing my dissertation.
iii
Table of Contents
List of Tables vii
List of Figures ix
Introduction 1
1 Who Pays for Voluntary Quality Certification? Evidence from the Non-GMO
Project Verified Label 4
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Institutional Details . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Non-GMO Project Verification . . . . . . . . . . . . . . . . . . 10
1.2.3 Relevant Literature . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Economics of Voluntary Non-GMO Food Labeling . . . . . . . . . . . 15
1.3.1 Baseline Model . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Introduction of the Non-GMO Label . . . . . . . . . . . . . . 18
1.3.3 Incentives for New Product Development . . . . . . . . . . . . 20
1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.1 Nielsen Retail Scanner and Consumer Panel Data . . . . . . . 23
1.4.2 Non-GMO Project Data . . . . . . . . . . . . . . . . . . . . . 24
1.4.3 Selection of Food Categories . . . . . . . . . . . . . . . . . . . 24
1.5 Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.1 Price Premium Regression Model . . . . . . . . . . . . . . . . 27
1.5.1.1 Main Specification . . . . . . . . . . . . . . . . . . . 27
1.5.1.2 Organic Interaction . . . . . . . . . . . . . . . . . . . 28
1.5.2 Quantity Regression Model . . . . . . . . . . . . . . . . . . . 30
1.5.3 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.6.1 Price Premiums . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.6.1.1 Main Results . . . . . . . . . . . . . . . . . . . . . . 32
1.6.1.2 Organic Interaction Results . . . . . . . . . . . . . . 34
1.6.2 Quantity Sold . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.7 Alternative Firm Strategies . . . . . . . . . . . . . . . . . . . . . . . 36
iv
1.7.1 Timing of Certification . . . . . . . . . . . . . . . . . . . . . . 36
1.7.2 Targeting to Non-GMO Consumers . . . . . . . . . . . . . . . 37
1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2 The Impact of Voluntary Non-GMO Labeling on Demand in the Ready-to-
Eat Cereal Industry 53
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 Background Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2.1 Demand Estimation in Markets with Differentiated Products . 55
2.2.2 Ready-to-Eat Cereal Industry . . . . . . . . . . . . . . . . . . 57
2.2.3 Willingness-to-Pay for Non-GMO . . . . . . . . . . . . . . . . 59
2.3 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.1 Consumer Demand with Heterogeneous Preferences . . . . . . 60
2.3.2 Firm Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4.1 Nielsen Retail Scanner and Consumer Panel Data . . . . . . . 64
2.4.2 Non-GMO Project . . . . . . . . . . . . . . . . . . . . . . . . 65
2.4.3 Market Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.4.4 Consumer Demographic Data . . . . . . . . . . . . . . . . . . 67
2.5 Empirical Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.5.1 Demand Estimation . . . . . . . . . . . . . . . . . . . . . . . 68
2.5.2 Price Instruments . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.5.3 Time-Invariant Product Characteristics . . . . . . . . . . . . . 72
2.5.4 Welfare Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.6.1 Logit Specification . . . . . . . . . . . . . . . . . . . . . . . . 76
2.6.2 Full Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6.3 Simulated Welfare Effects . . . . . . . . . . . . . . . . . . . . 80
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3 Produce Growers’ Cost of Complying with the Food Safety Modernization
Act (with Erik Lichtenberg) 103
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.2.1 Relevant Literature . . . . . . . . . . . . . . . . . . . . . . . . 105
3.2.2 The Food Safety Modernization Act and the Produce Rule . . 107
3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.1 Survey Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.2 Survey Administration . . . . . . . . . . . . . . . . . . . . . . 111
3.3.3 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . 112
3.4 Profit Maximizing Choice of Food Safety Practices: Theory and Econo-
metric Specification of Expenditures . . . . . . . . . . . . . . . . . . . 114
3.5 Estimation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.6 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.6.1 Effect of Farm Size on Expenditures on Food Safety Practices 122
v
3.6.2 Effect of Sustainable Farming Practices on Expenditures on
Food Safety Practices . . . . . . . . . . . . . . . . . . . . . . . 124
3.6.3 Effect of Farm Size and Sustainable Farming Practices on Use
of Food Safety Practices . . . . . . . . . . . . . . . . . . . . . 126
3.7 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.7.1 Farm Size and the Cost Burden . . . . . . . . . . . . . . . . . 128
3.7.2 Effect of Changes to FSMA Exemption Thresholds on Food
Safety Cost Burden . . . . . . . . . . . . . . . . . . . . . . . . 129
3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
A Appendix to Chapter 1 147
A.1 Solution for the Baseline Equilibrium . . . . . . . . . . . . . . . . . . 147
A.2 Solution When Only Firm A Labels Products . . . . . . . . . . . . . 148
A.3 Equilibrium When Both Firms Labels Products . . . . . . . . . . . . 149
A.4 Equilibrium When Firms Develop New Products . . . . . . . . . . . . 149
A.5 Price Premium Estimation with Full Sample . . . . . . . . . . . . . . 150
B Appendix to Chapter 2 153
B.1 Additional Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
B.2 Computational Details . . . . . . . . . . . . . . . . . . . . . . . . . . 159
B.2.1 Parallel Cluster Inititalization . . . . . . . . . . . . . . . . . . 160
B.2.2 Parallelized Mean Utility . . . . . . . . . . . . . . . . . . . . . 160
B.2.3 Parallelized Jacobian . . . . . . . . . . . . . . . . . . . . . . . 162
C Appendix to Chapter 3 164
C.1 Additional Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
C.2 Survey Instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References 183
vi
List of Tables
1.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.2 Manufacturer-Category Variation in Certification Timing . . . . . . . 47
1.3 Price Premium Regressions . . . . . . . . . . . . . . . . . . . . . . . . 48
1.4 Price Premium Regressions with Organic Interaction . . . . . . . . . 49
1.5 Quantity Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.6 Non-GMO Certification of New and Pre-Existing Food Products . . . 51
1.7 Average Consumer for Conventional & Non-GMO Products . . . . . 52
2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.2 Descriptive Statistics for Demographic Variables by DMA . . . . . . . 90
2.3 Results from the Logit Specification . . . . . . . . . . . . . . . . . . . 95
2.4 Full Model Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.5 Initial Baseline for Simulation . . . . . . . . . . . . . . . . . . . . . . 97
2.6 Simulated Labeling Scenario Results . . . . . . . . . . . . . . . . . . 99
2.7 Welfare Effects of Simulated Labeling Scenarios . . . . . . . . . . . . 102
3.1 Survey Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 137
3.2 Regional Distributions of Farm Operations . . . . . . . . . . . . . . . 138
3.3 Exemption Status by Farm Size and Grower Organization . . . . . . 139
3.4 Estimated Farm Size Elasticities of Food Safety Practice Expenditures140
3.5 Estimated Impact of Sustainable Farming Practices on Food Safety
Practice Expenditures . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.6 Marginal Effects of Acreage, Marketing Channel, Farming Practices,
and Crop Type on the Probability of Safety Measure Use . . . . . . . 143
3.7 Effects of Changes to Exemption Thresholds on Food Safety Cost
Burden Percentage for Each Food Safety Practice . . . . . . . . . . . 144
A.1 Price Premium Regressions - Unrestricted Sample . . . . . . . . . . . 152
B.1 Median Own- and Cross-Price Elasticities - Brands 1-10 . . . . . . . . 154
B.2 Median Own- and Cross-Price Elasticities - Brands 11-20 . . . . . . . 155
B.3 Median Own- and Cross-Price Elasticities - Brands 21-30 . . . . . . . 156
B.4 Median Own- and Cross-Price Elasticities - Brands 31-40 . . . . . . . 157
B.5 Median Own- and Cross-Price Elasticities - Brands 41-50 . . . . . . . 158
vii
C.1 Grower Conferences and Online Grower Listservs Surveyed . . . . . . 165
C.2 Estimated Coefficients for Double Hurdle Specification With No In-
teractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
C.3 Estimated Coefficients for Double Hurdle Specification With Interac-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
viii
List of Figures
1.1 Non-GMO Project Verified Label . . . . . . . . . . . . . . . . . . . . 41
1.2 Cumulative Monthly Non-GMO Project Verified Products by Organic
Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.3 Growth in Non-GMO Project Verified Products by Product Category 43
1.4 Annual Non-GMO Project Verified Product Sales . . . . . . . . . . . 44
1.5 Product Age When Non-GMO Project Verified . . . . . . . . . . . . 45
2.1 Annual Non-GMO Project Verified RTE Cereal Sales . . . . . . . . . 85
2.2 Distribution of Non-GMO Label Coefficient . . . . . . . . . . . . . . 86
3.1 Empirical Distributions of Fruit and Vegetable Farm Revenue and
Farm Acreage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
ix
Introduction
This dissertation features three essays exploring the market impacts of two
types of quality certification—a voluntary non-GMO label and a mandatory food
safety standard. Quality certification plays an important role in many industries,
especially in markets for credence goods (Darby and Karni 1973) and markets with
adverse selection (Akerlof 1970). In these cases, consumers cannot accurately eval-
uate the quality of goods prior to purchase; therefore, certification can alleviate
informational asymmetries between consumers and firms and increase market effi-
ciency (Dranove and Jin 2010).
Many forms of quality certification exist in food and agricultural markets in
the U.S., ranging from voluntary disclosure to mandatory standards. In the case
of non-GMO food products, certification is based on a voluntary third-party label
managed by the Non-GMO Project. The Non-GMO Project began offering non-
GMO certification and labeling in 2010 for food products that fall under a 0.9%
threshold for GMO1 presence. Products that obtain the certification feature an
easily recognizable label on their package. With regards to produce safety, the
1GMO stands for “genetically modified organism” and refers to plants whose genetic mate-
rial has been altered using genetic engineering techniques, such as recombinant DNA technology.
This term is also used to describe agricultural crops produced from seed stock that employs this
technology and food products that contain ingredients derived from these crops.
1
prevailing “certification” is the Produce Rule implementing the Food Safety Mod-
ernization Act [FSMA], a mandatory government-based minimum quality standard
for growing produce. The Produce Rule requires operational changes to meet stan-
dards associated with agricultural water; biological soil amendments; domesticated
and wild animals; employee training and health and hygiene; and equipment, tools,
buildings, and sanitation. This dissertation explores the economic impact of these
two quality certifications. The first two essays examine how the Non-GMO Project
Verified certification affects demand for food products, and the third essay examines
growers’ costs of complying with the Produce Rule.
The first essay investigates whether firms use a voluntary quality certification
for non-GMO products to extract rent from consumers. Using weekly retail point-of-
sale data from a large sample of supermarkets across the U.S coupled with a unique
dataset from the Non-GMO Project, I find no statistically significant price premi-
ums or quantity changes for newly certified non-GMO food products. I, however,
find support for the hypothesis that the certification may induce other firm strate-
gies such as new non-GMO product development targeted to specific consumers.
Altogether, the findings suggest that certification costs are not passed directly to
consumers of existing products that are reformulated to meet the non-GMO certifi-
cation standard. Instead, firms may pass the costs using newly introduced products.
The second essay builds on the first and investigates the role of voluntary
non-GMO food labeling as a non-price marketing strategy in the ready-to-eat [RTE]
cereal industry. I estimate a discrete-choice, random coefficients logit demand model
with monthly retail point-of-sale data for 50 breakfast cereal brands in 100 DMAs
2
between 2010 and 2014. The results indicate that consumer tastes for the non-
GMO label have a wide distribution, and this heterogeneity plays a substantial role
in individual choices; but, on average, the non-GMO label has a positive impact
on demand. To shed light on the potential welfare effects of non-GMO labeling, I
simulate two labeling scenarios in the RTE cereal industry: one in which all brands
use the non-GMO label over the entire timeframe of the data, and one in which no
brands use the label. The simulation results indicate that non-GMO labeling in the
RTE cereal industry may reduce industry profits but improve consumer welfare on
average.
The final essay, co-authored with Erik Lichtenberg, addresses concerns raised
by small and medium size produce growers that compliance with the FSMA Produce
Rule will be prohibitively costly. We use data from an original national survey of
fruit and vegetable growers to examine that contention. In particular, we examine
how expenditures on food safety measures required by the Produce Rule vary with
farm size and farming practices using a double hurdle model to control for selec-
tivity in both using food safety practices and reporting expenditures. We find that
expenditures per acre decrease with farm size and that growers using sustainable
farming practices spend more than conventional growers on many food safety prac-
tices. We use our estimates to quantify how the cost burden of compliance varies
with farm size. We then explore the policy implications of exemptions to the Rule
by simulating how changes to the exemption thresholds for farm revenue and share
of direct sales might affect the cost burden of each food safety practice on farms at
the threshold.
3
Chapter 1: Who Pays for Voluntary Quality Certification? Evidence
from the Non-GMO Project Verified Label∗
1.1 Introduction
Quality disclosure is an important element of many industries, most notably
in markets for credence goods (Darby and Karni 1973) and markets with adverse se-
lection (Akerlof 1970). In both cases, quality certification corrects an informational
asymmetry between consumers and firms, enabling consumers to ascertain product
quality, which can lead to quality improvements and facilitate vertical sorting (Dra-
nove and Jin 2010). By the same token, depending on market structure, firms may
use quality certification to exercise market power and engage in second degree price
discrimination and extract rent from consumers (Mussa and Rosen 1978), typically
benefiting firms at the expense of consumers. This paper explores the role of volun-
tary quality certification as a means to exercise such market power, using evidence
from a voluntary non-GMO certification in the U.S food industry.
The Non-GMO Project began offering non-GMO certification and labeling in
2010 for food products that fall under a 0.9% threshold for GMO presence. Products
∗Nielsen data is provided by the Data Center at The University of Chicago Booth
School of Business. Information on availability and access to the data is available at
http://research.chicagobooth.edu/nielsen.
4
that obtain the certification feature an easily recognizable label2 on their packaging
that reads, “Non-GMO Project Verified” (See Figure 1.1). The Non-GMO Project
does not restrict the types of products that can be certified, which is to say that a
product is eligible for certification regardless of whether or not it contains ingredi-
ents for which commercially produced GMO variants currently exists. Furthermore,
organic products, which are prohibited from containing GMO ingredients based on
the National Organic Program standards, are also eligible for certification. As such,
the cost of certification can vary significantly depending the magnitude of product
changes required (e.g., product reformulation, sourcing of new ingredients, etc.) to
meet the non-GMO certification standard.
The goal of this paper is to first determine whether firms use a voluntary,
non-GMO quality certification to extract price premiums3 or increase market share
for newly certified food products, and whether these strategies evolve over time.
Specifically, I use a hedonic framework to estimate price premiums and changes
to quantity sold for newly certified non-GMO food products using the Non-GMO
Project Verified label. The estimation is carried out on a large sample of weekly,
product-level retail point-of-sale data for 18 product categories from U.S. super-
markets from 2009 to 2014. The transaction data is coupled with a unique dataset
from the Non-GMO Project that contains non-GMO certification dates for products
throughout the label’s history. I find no evidence of price premiums or changes to
quantity sold for newly certified non-GMO food products. I then investigate alter-
2Throughout the paper, I use the terms “label” and “certification” interchangeably with regards
to the Non-GMO Project verification standard.
3Depending on the product, price premiums may reflect pass-through of certification costs, rent
extraction, or a combination of both.
5
native strategies by which firms could extract rent and pass the certification cost
to consumers. I find suggestive evidence that the certification may induce firms to
develop new non-GMO products for specific types of consumers.
The paper is structured as follows. Section 1.2 provides institutional details
and discusses the literature on quality disclosure and labeling as well as previous
empirical work on willingness-to-pay for non-GMO products. Section 1.3 presents
a simple theoretical model on the economics of labeling to help guide the empirical
analysis. Section 1.4 describes the data sources I employ to implement this study.
Section 1.5 presents the empirical model for the analysis of non-GMO price pre-
miums and quantities sold, with accompanying results in Section 1.6. Section 1.7
explores alternative strategies firms may use to extract rent and pass the certifica-
tion cost to consumers. Lastly, Section 1.8 offers concluding remarks as well as next
steps to better understand firm behavior and consumer preferences.
1.2 Background
1.2.1 Institutional Details
In the U.S., over 90% of canola, corn, cotton, soybeans, and sugar beets are
GMO.4 Most GMO seed varieties are modified to carry several input-traits designed
to benefit producers, the most common of which are herbicide tolerance5 and insect
4GMO stands for “genetically modified organism” and refers to plants whose genetic material
has been altered using genetic engineering techniques, such as recombinant DNA technology. In the
literature, this term is used interchangeably with GM (“genetically modified”) and GE (“genetically
engineered”) to describe agricultural crops produced from seed stock that employs this technology
and food products that contain ingredients derived from these crops.
5Monsanto’s “Roundup Ready” seeds for canola, corn, soybeans, and sugar beets are resistant
to glyphosate, the active chemical in their popular herbicide Roundup.
6
resistance.6 While genetically engineered seeds exist for additional crops and input
traits, these crops represent the vast majority of the total area of GMO crop varieties
planted in the U.S. Many common ingredients used in processed foods are derived
from these GMO crops, such as aspartame, flavorings, high-fructose corn syrup, oils,
starches, and various additives and preservatives; and the Grocery Manufacturers
Association estimates that 70-80% of food eaten in the U.S. contain GMOs (Bren
2003).
The FDA asserts that approved GMO food products are not significantly dif-
ferent from or less safe than their non-GMO produced counterparts and, thus, do
not require additional labeling. Nonetheless, 64 countries around the world require
labeling of GMO food, and labeling has become a mainstream debate in the U.S. The
U.S. Organic Standard prohibits the use of GMOs, thus providing an indirect form of
non-GMO labeling for Organic food products in the U.S., but conventionally-grown
food has no such restriction. Nonetheless, a voluntary verification and labeling
scheme for non-GMO products called the Non-GMO Project emerged in the U.S.
beginning in 2010. As of December 2015, nearly 35,000 products from over 1,900
brands use the label, accounting for over $13.5 billion in annual sales.
Twenty U.S. states introduced mandatory GMO labeling legislation in 2014,
by which time mandatory GMO labeling laws had already been passed in Maine,
Connecticut, and Vermont. The labeling laws in Maine and Connecticut contained
trigger clauses that required additional states to pass similar laws before theirs would
6Monsanto seeds for corn, cotton, and soybeans express genes for insecticidal proteins from
Bacillus thuringiensis (Bt).
7
go into effect, but the Vermont law contained no such clause and became effective on
July 1, 2016. In the meantime, Congress passed the National Bioengineered Food
Disclosure Standard (2016), creating a national mandatory GMO labeling standard.
The bill, which became law on July 29, 2016, preempts any mandatory state GMO
labeling laws and calls for the creation of a federal labeling standard within two
years of its enactment. Notably, the law allows food manufacturers a choice of
labeling including on-package text, a symbol, or a digital link (e.g., a QR code) that
provides access to an internet website containing information about the product’s
GMO content (Hall 2016). Critics of the new law insist that the labeling options
are too lenient and will allow food manufacturers to hide the GMO content of their
products behind a QR code, effectively preventing consumers without smartphones
from accessing that information (Lowe 2016).
The institutional incentives for non-GMO food labeling are well established in
the economics literature. In the context of information economics, non-GMO food
products are differentiated by a vertical process attribute unobservable to the con-
sumer, even after consumption, which makes them a type of credence good (Darby
and Karni 1973). The commonly prescribed mechanism for dealing with this infor-
mation asymmetry is to implement some type of third-party monitoring or labeling,
much like the Organic standard (McCluskey 2000). GMO labeling schemes vary
across countries, with the U.S.7 and Canada employing a voluntary non-GMO la-
beling regime, while the European Union, Australia, New Zealand, and Japan use
7In the case of the U.S., mandatory labeling will take effect once rulemaking is finalized for the
National Bioenginerred Food Disclosure Act.
8
a mandatory GMO labeling scheme. The typical economic argument for voluntary
labeling is that, in the absence of market failures, this regime yields the socially opti-
mal outcome while avoiding any unnecessary costs to society. One of the arguments
commonly promulgated by the food industry against mandatory GMO-labeling is
that such a law would cause a large increase in food prices as food manufactur-
ers reformulate their products to be non-GMO to avoid the stigma that a “contains
GMO” label would create, which proved to be a very effective argument in defeating
a patchwork of state legislation, most notably Prop 37 in California in 2012 (Carter
et al. 2012).
As a corollary to such a cost argument, one might also argue that food man-
ufacturers who choose to use a non-GMO label would also increase food prices,
passing on the costs associated with ingredient reformulation as well as certification
and labeling fees to the consumer. However, if the market for existing products
that become non-GMO certified is very price competitive, already commands high-
margins, or is subject to low retailer pass-through rates,8 firms may not necessarily
be able to pass these costs on to consumers. Despite these limitations, if firms can
increase market share by adopting the label, incentives may still exist to seek out
non-GMO certification. On the other hand, firms may have an opportunity to use
new product development as a means to extract a non-GMO price premium. That
is, food manufacturers may certify new products prior to market entry and launch at
a higher price point. In this case, we may observe firms behaving more strategically,
8Besanko et al. (2005) analyze retailer’s pass-through behavior of a major U.S. supermarket
chain for 78 products across 11 categories and find that pass-through varies substantially across
products and across categories.
9
choosing to price non-GMO products certified within their product life differently
from new products certified before market entry. In this paper, my empirical anal-
ysis focuses primarily on the first group—products certified within their product
life—but I also provide some descriptive analysis to help characterize the second
group of products.
1.2.2 Non-GMO Project Verification
The Non-GMO Project is a nonprofit organization that offers third-party ver-
ification and labeling for products that fall under a 0.9% threshold for GMO pres-
ence, which aligns with the mandatory labeling standards in Europe. The Non-GMO
Project Standard defines the program’s core requirements including traceability, seg-
regation, and testing of high-risk ingredients at critical control points (Non-GMO
Project 2014c). The verification process is handled by one of three technical ad-
ministrators: FoodChainID, NSF International, and IMI Global. Products that
contain any high GMO risk ingredients9 require an onsite inspection for verifica-
tion, whereas products with low-risk ingredients may only require a review of the
ingredient specification sheets, and therefore verification costs can vary considerably
between products (Non-GMO Project 2014a). The Non-GMO Project Standard also
requires ongoing testing of all at-risk ingredients10 as well as rigorous traceability
9The Non-GMO Project classifies the following crops as high GMO risk: alfalfa, canola, corn,
cotton, papaya, soy, sugar beets, and zucchini and yellow summer squash. Inputs derived from
these crops and animals fed these crops or their derivatives are also considered high-risk. They also
maintain a list of monitored crops for which suspected or known incidents of accidental comingling
have occurred that are regularly tested (Non-GMO Project 2014e).
10The Non-GMO Project Standard requires testing of individual ingredients, not finished prod-
ucts, because the latter is not a reliably accurate measure of GMO presence (Non-GMO Project
2014c).
10
and segregation practices, both of which are maintained through annual audits and
on-site inspections for high-risk products (Non-GMO Project 2014b). On average,
the verification process takes 4 to 6 months, and upon completion the Non-GMO
Project provides the producer with a licensing agreement to use their name and
verification mark on the verified product.
The Non-GMO Project also verifies products for which no commercially pro-
duced GMO variant currently exists, which they refer to as low-risk. Their rationale
for doing so involves four distinct considerations (Non-GMO Project 2014d): (1)
Some low-risk products may still contain high-risk ingredients, such as the oil some-
times used in packaged dried fruit; (2) Incidents of accidental comingling of GMO
material have occurred with seemingly low-risk products such as rice and flax, so
verification can help mitigate these issues; (3) The organization believes that only
verifying high-risk products may place a large burden on consumers to know which
products are at risk of containing GMO ingredients, and this lack of understanding
may provide an unfair marketing advantage to products with high-risk ingredients
carrying the label; and (4) The organization believes that verifying low-risk prod-
ucts helps raise awareness and build consumer interest for non-GMO food products
as a whole, which can help set norms as new GMO products are developed.
Usage of the Non-GMO Project Verified label has grown significantly over the
past five years, with sales of labeled products in 2014 totaling $11 billion (Non-GMO
Project 2014a). Figure 1.2 shows the monthly cumulative growth in products using
the label by Organic status. The label launched with about 200 products in 2010
and includes over 15,000 products as of January 2015. Products using the label are
11
close to evenly split between Organic and conventionally grown. Non-GMO Project
Verified products span a wide range of product categories as well. Figure 1.3 shows
the annual growth by product category for products using the label. As of December
2014, the largest category was snack foods, desserts, and sweeteners, accounting for
over 2,800 products. Other large categories include beverages; breads, grains, and
beans; fruits and vegetables; and packaged/prepared foods, each of which comprises
over 1,500 products.
As of 2015, the Non-GMO Project Verified program accounts for the largest
share of non-GMO food labeling in the U.S., but in recent years other voluntary
labeling efforts have also emerged. Whole Foods Market, the top specialty grocer
in the U.S., has vowed to label all GMO products in their stores by 2018, and
the FDA recently finalized their industry guidance for voluntary non-GMO labeling
(FDA 2015). In May 2015, at the request of a major non-GMO grain dealer in
the U.S., the USDA developed a voluntary non-GMO certification and labeling
program through their existing “USDA Process Verified” program (Jalonick 2015).
Similar to other USDA-sponsored voluntary food labels such as “humanely raised”
or “grass fed”, the program is administered through the department’s Agriculture
Marketing Service and is available to companies for a paid fee. Also in mid-2015,
NSF International launched another private label option called the Non-GMO True
North program, which offers certification and labeling of non-GMO intermediate
and retail products (Greene et al. 2016).
12
1.2.3 Relevant Literature
The concept of a credence good was first discussed by Darby and Karni (1973)
as an extension of search and experience goods (Nelson 1970). In the context of a
vertically differentiated good,11 the consumer knows what she needs ex ante, but
she neither observes the utility nor the type of good she receives ex post. Because
consumers cannot verify quality even after consumption, a market for credence goods
will theoretically fail in the absence of third-party monitoring.
More broadly, credence goods are simply a manifestation of asymmetric infor-
mation between consumers and producers, a topic widely discussed in the literature
on quality disclosure. Perhaps the most fundamental and oft-cited result in this
area is the well-known “unraveling result” (Viscusi 1978; Grossman 1981). Accord-
ing to the theory, all but the lowest quality seller in a market have an incentive to
voluntarily disclose quality information, thus eliminating the need for mandatory
disclosure. However, this result is based on several strong assumptions, so in real-
ity, we observe incomplete voluntary disclosure in many markets.12 Milgrom and
Roberts (1986) show that if the consumer is unsophisticated or not well informed,
full voluntary disclosure will generally fail. This is particularly applicable to non-
GMO labeling given that genetic engineering is a relatively new technology, and
the average consumer may be unaware of its proliferation in the conventional food
system. Another important consideration is interactions between different quality
11In the case of non-GMO food products, this vertical differentiation takes the form of a process
attribute.
12For an extensive review of the literature on the failure of unraveling and, more broadly, on the
theory and practice on quality disclosure, see Dranove and Jin (2010).
13
labels (e.g., interaction between Organic and non-GMO food labels). Bonroy and
Constantatos (2014) review the theoretical literature on quality labels and discuss
how a new label may interact with existing market distortions, identifying a number
of effects that may cause the industry not to set a socially optimal label. From a
relevant policy perspective, Roe and Sheldon (2007) examines the tradeoffs between
different labeling regimes (private versus government, discrete versus continuous
quality, mandatory versus voluntary) and shows that firms tend to prefer discrete
labels certified by private firms.
Most empirical studies of GMO labeling employ hypothetical surveys and in-
centivized lab experiments to analyze consumer preferences for GMO products. Lusk
et al. (2005) identifies 25 separate studies that together provide 57 estimates of con-
sumers’ willingness-to-pay (WTP) for GMO food products. Significant variation
exists in the valuation estimates across studies. Percentage premiums for non-GMO
food ranged from -68% to 784%, with an average of 42%, and are significantly
affected by elicitation method.
More recent studies tend to focus on the issue of GMO labeling more directly
and attempt to quantify the effects of different labels. Onyango et al. (2006) uses
a nationwide survey to analyze U.S. consumer’s choice of cornflakes in five different
labeling scenarios. They find that consumers place a 10% premium on food labeled
as non-GMO and 6.5% discount on food labeled as GMO; but, interestingly, con-
sumers also attach a 5% premium for food labeled GMO if the label also specifies
“USDA approved” or “to reduce pesticide residues in your food.” Roe and Teisl
(2007) further explores the nuances of GMO labeling content by eliciting consumer
14
reactions to 80 different GMO label variations through a survey. A key finding of
the study is that labels with simple claims and claims certified by the FDA are
most credible. Dannenberg et al. (2011) uses an experimental auction to compare
mandatory versus voluntary labeling of GMO food and finds that when two labels
exist, one for GMO and one for non-GMO, both schemes generate a similar level of
uncertainty about unlabeled products. Costanigro and Lusk (2014) conducts a series
of choice experiments and finds evidence that consumer WTP to avoid GMO food
is 140% higher with a mandatory “contains” GMO label compared to a voluntary
“does not contain” GMO label.
1.3 Economics of Voluntary Non-GMO Food Labeling
Food products tend to be differentiated both horizontally and vertically in
product space. The vertical dimension reflects quality-based attributes, which share
a “more is better” property. Therefore, while consumers may differ in their valua-
tion of quality, these attribute levels can be consistently ranked by all consumers.
The horizontal dimension reflects non-quality taste attributes, which each consumer
will rank differently based on their taste preferences. For example, nutritional con-
tent and freshness could be vertical quality attributes, whereas texture, flavor, and
brand could be considered horizontal taste attributes. Given the observed level of
segmentation in nearly all food product categories, any realistic model of non-GMO
food labeling should account for heterogeneity in consumer preferences along both
dimensions. Additionally, in terms of market structure, the food manufacturing in-
15
dustry tends to be dominated by multiproduct firms operating under oligopolistic
competition.13
1.3.1 Baseline Model
Desai (2001) develops a multiproduct firm duopoly model in which the market
consists of two consumer segments that differ in their willingness to pay for quality,
and consumers in each segment are distributed across a linear city a la Hotelling
(1929). Each consumer’s location reflects her specific taste preference, and the
“transportation cost” for each segment represents that segment’s strength of taste
preferences. Each firm produces up to two products (one for the low-valuation
segment and one for the high-valuation segment), and has a fixed location on the
line within a given segment that reflects consumer perceptions of the firm’s product
taste attributes (i.e. a firm’s location may differ across consumer segments). I use
this model as a baseline by which to analyze the economics of voluntary non-GMO
food labeling and present a summary of its original specification below.
Formally, as presented in Desai (2001), the market consists of a high-valuation
and a low-valuation consumer segment (i = H,L), each of which containsNH andNL
consumers, respectively. Consumers in each segment gain utility θiq from consuming
a product of quality q. Furthermore, consumers in each segment are uniformly
distributed along [0, 1] in a linear city and incur transportation cost ki to travel
away from their location x. As such, a consumer of type i derives utility U(θi, x) =
13Take, for example, the market for ready-to-eat cereals, in which the top four firms, each
producing dozens of different brands, account for upwards of 80% of the market (Nevo 2001).
16
θiq − kit − p from consuming a product of quality q, price p, located a distance
t from the consumer’s location. The supply side consists of two symmetric firms
(j = A,B), each of which produces a product of quality qij for consumer segment
i. In each segment i, Firm A is located at a distance ai, and Firm B is located at a
distance bi from the left,
14 where ai < 1/2 and bi = 1− ai > 1/2. Costs are identical
across firms, with each firm incurring marginal cost c(q) = γq2/2 for producing
a product of quality q. Firms maximize profits by choosing prices (pLj, pHj) and
quality levels (qLj, qHj) for their respective product lines.
In contrast to the classic monopoly self-selection models with only vertical
differentiation (Mussa and Rosen 1978; Moorthy 1984), in this model firms provide
each segment with its preferred quality level and the typical self-selection constraints
are not binding in equilibrium (under some conditions).15 In this scenario, when
both market segments are fully served, the location xi of the consumer in segment
i indifferent between buying products of quality qiA and qiB offered by each firm is
xi =
θi(qiA − qiB) + ki(ai + bi)− (piA − piB)
2ki
if ai ≤ xi ≤ bi. (1.1)
We can thus calculate demand for each product of Firm A (diA) and Firm B (diB) as
diA = Nixi and diB = Ni(1− xi), respectively. The equilibrium prices and qualities
14See Figure 1 in Desai (2001) for a visual representation of the model.
15The appendix in Desai (2001) provides a full derivation of the equilibrium for the duopoly
model.
17
for the unconstrained problem are
p∗iA =
1
6
(2ki(2 + ai + bi)) +
θ2i
2γ
,
p∗iB =
1
6
(2ki(4− ai − bi)) + θ
2
i
2γ
, q∗ij =
θi
γ
, (1.2)
and this solution satisfies the self-selection constraints when −2γ[(1 +aL−aH)kH −
kL] + (θH − θL)2 ≥ 0. Economically, this condition indicates that firms are more
likely to provide low-valuation consumers with their preferred quality levels when
there is greater heterogeneity in quality valuations than in taste preferences between
segments; and firms are more likely to provide efficient quality to the low-valuation
segment as horizontal differentiation decreases in the high-valuation segment. Ad-
ditionally, price-cost margins are increasing in taste preferences ki; if high-valuation
consumers also have stronger taste preferences, the higher quality product will have
higher price-cost margins as well.
1.3.2 Introduction of the Non-GMO Label
Within the horizontal or vertical product space, food can have multiple at-
tributes, each of which consumers may value differently. In the baseline model,
I represent the attributes within each space as single taste and quality attribute.
When thinking about how to model voluntary non-GMO labeling, consider that
Non-GMO production is a vertical process attribute that is (weakly) more costly
than conventional production, which supports its treatment as a quality attribute in
a vertical differentiation framework. Nonetheless, because non-GMO is a credence
18
attribute, it is plausible that some consumers may value it in a fundamentally dif-
ferent way than other quality attributes, thus warranting a separate treatment in
the consumer utility function.
Prior to the introduction on the non-GMO label, suppose that consumers
derive utility and firms operate in equilibrium as characterized in the baseline model
presented in Section 1.3.1.16 Once the non-GMO label is launched, consumers gain
additional benefit from food products that feature the label such that the a consumer
of type i derives utility U(θi, δi, x) = θiq+δiDng−kit−p, where Dng = 1 if a product
undergoes non-GMO certification and uses the label, and δi is the incremental gain
in utility a consumer of type i gets from consuming a non-GMO product.
When a firm chooses to certify an existing product, it necessitates replacing
any GMO ingredients in the product with non-GMO variants, which often cost more
to produce. I model this new cost as a constant, additive term α in the expression
for marginal cost: c(q,Dng) = γq
2/2 + αDng. That said, it is important to note that
non-GMO variants do not differ from GMO variants of ingredients nutritionally or in
any other way the FDA deems “significant,” so this type of “reformulation” is more
aptly characterized as simply a re-sourcing of certain inputs and may not induce
firms to re-optimize over other quality attributes.17 Therefore, I assume that quality
is fixed at the equilibrium in the baseline model, and firms decide whether or not to
label their products by choosing prices to maximize profits given the new consumer
utility and marginal cost.
16See Appendix A.1 for a brief characterization of the baseline equilibrium.
17In fact, if firms did reformulate a product in an appreciable way, resulting in a change of
ingredients or nutritional content, this would trigger a new UPC and effectively qualify as a new
product.
19
We can show that if Firm A chooses to label its products and Firm B does
not, the consumer indifference location xi shifts to the right by an amount δi/2ki,
enabling Firm A to capture additional market share and make greater profit at Firm
B’s expense, provided that δi > α. By symmetry, the equivalent outcome occurs if
Firm B chooses to label and Firm A does not.18 In equilibrium, both firms’ optimal
strategy is to label their products, in which case xi remains unchanged relative to the
baseline case due to symmetry across firms, and both firms’ prices simply increase by
an amount equal to the additional marginal cost α for non-GMO labeling, resulting
in no change in profits.19 Interestingly, in the duopoly setting, neither firm can
use the non-GMO label to exercise market power and extract additional rents from
consumers due to the nature of competition between the two firms. Additionally,
note that if the incremental marginal cost α associated with non-GMO labeling
is insignificant, firms may not increase prices at all upon labeling their products.
Lastly, due to symmetry of the non-GMO utility term across products of either
quality type for a given consumer, the labeling equilibrium will automatically satisfy
the self-selection constraints when −2γ[(1 + aL − aH)kH − kL] + (θH − θL)2 ≥ 0, as
in the baseline equilibrium.
1.3.3 Incentives for New Product Development
In the previous model of non-GMO labeling, I assume that consuming a non-
GMO product benefits consumers in either segment i through an incremental gain in
18See Appendix A.2 for a complete derivation of the result when only one firm chooses to label
its products as non-GMO.
19See Appendix A.3 for a derivation of the equilibrium when both firms choose to label their
products as non-GMO.
20
utility δi; and once the label launches in the marketplace, firms take quality of their
existing product line as fixed from the pre-labeling regime when deciding whether or
not to label their products. However, one might argue that a more GMO-attentive
consumer segment exists that values the non-GMO label in a fundamentally different
way. For this segment of consumers, the benefit of the non-GMO label may interact
with the quality and taste dimensions in consumer utility, such that it enhances
their valuation of quality or increases the strength of their taste preferences20—i.e.,
brand loyalty.
In this scenario, the label could induce firms to develop new products at the
higher, preferred quality level of this segment because doing so would enable firms to
extract higher price-cost margins on these products. In effect, the label may serve as
a mechanism by which to segment these GMO-attentive customers from other high-
valuation consumers and make it more profitable to serve them separately. For the
sake of comparability, I build off the same two-segment baseline model presented
in Section 1.3.1. Once the non-GMO label is launched, consumers in segment L
derive utility U(θL, δL, x) = θLq + δLDng − kLt − p, just as before. Consumers in
segment H derive utility U(θH , δH , x) = θH(1 + δHDng)q − kH(1 + δHDng)t − p,
as described above. Marginal costs are c(q,Dng) = γq
2/2 + αDng. Lastly, I assume
that quality for products in segment L (i.e. pre-existing products) is fixed at the
equilibrium in the baseline model, but firms now choose new optimal quality levels
for products in segment H (i.e. new product development), in addition to choosing
20Empirical work suggests that not all consumers assign a non-negative valuation to non-GMO
certification (Lusk et al. 2005), in which case the non-GMO attribute may not meet the “more is
better” property for all consumers, thus warranting its treatment as a horizontal taste attribute.
21
prices to maximize profits given the new consumer utility and marginal cost. In
equilibrium, both firms’ optimal strategy is to label their products and increase
quality for segmentH. The equilibrium prices for products in segment Lmatch those
presented in Section 1.3.2, where prices increase by an amount equal to the additional
marginal cost α for non-GMO labeling. The equilibrium prices and qualities for
products in segment H are
p∗HA =
1
6
(2kH(1 + δH)(2 + aH + bH)) +
θ2H(1 + δH)
2
2γ
+ α,
p∗HB =
1
6
(2kH(1 + δH)(4− aH − bH)) + θ
2
H(1 + δH)
2
2γ
+ α,
q∗Hj =
θH(1 + δH)
γ
. (1.3)
Clearly, in this model the price-cost margins for segment H are amplified by the
interaction effect of the non-GMO label with taste preferences in consumer utility.
In fact, unlike the previous case, even if both segments have identical transportation
costs (i.e. kH = kL), segment H still commands a higher price-cost margin in this
setting. Additionally, we can show that both firms earn additional profit in the
amount of
∆Πj =
NHkHδH
2
(1.4)
over the baseline equilibrium and, by extension, the equilibrium in which both firms
simply label their pre-existing product line.21
21See Appendix A.4 for a derivation of the equilibrium when both firms choose to introduce a
new product for segment H.
22
1.4 Data
1.4.1 Nielsen Retail Scanner and Consumer Panel Data
Because the Non-GMO Project Verified label has only been in use since 2010,
and the majority of its growth has occurred from 2012 onward, it is crucial that
this research is conducted with the most recent demand data available. The Nielsen
Retail Scanner (Nielsen RMS) data contains weekly purchase and pricing data from
retail store point-of-sale systems for over 2.6 million UPCs. The data includes 35,000
retail stores across the U.S., representing over 90 major retail chains in 52 markets.
It includes all 1,100 reported Nielsen product categories, which span 125 product
groups and 10 departments. In addition to demand data, the dataset includes store
demographics and product characteristics.
The Nielsen Consumer Panel (Nielsen HMS) data contains trip-level purchase
and pricing data for an unbalanced panel of 40,000 to 60,000 U.S. households and
spans the same timeframe as the Retail Scanner data. The data is collected using
a handheld scanning device that participants use at home to track all their pur-
chases. Like the Retail Scanner data, the Consumer Panel data includes all 1,100
reported Nielsen product categories for all major retail channels: grocery, drug,
mass merchandise, superstore, club stores, convenience, and health. Along with
purchase data, the panel includes consumer demographics, product characteristics,
and geographic data.
23
1.4.2 Non-GMO Project Data
I have secured a unique monthly dataset of verified products from the Non-
GMO Project that includes product UPC, verification date, product name, product
category, organic status, and producer/brand name. The data spans the entire
label history through 2014. I merge this information with Retail Scanner data from
Nielsen to clearly identify non-GMO food products that use the Non-GMO Project
Verified label. Moreover, the label verification date contained in this dataset allows
me to identify when non-GMO products in the Nielsen data began using the Non-
GMO Verified label.
1.4.3 Selection of Food Categories
For this analysis, I focus on 18 food product categories primarily consisting
of snack foods, dry goods, and other processed foods. Table 1.1 presents summary
statistics for each product category of the Nielsen Retail Scanner Data from 2009
to 2014. The decision to focus on these categories is based on common and distinct
factors for each category. First, all 18 categories are comprised of a non-negligible
share of Organic products. Because Organic products do not contain GMO ingredi-
ents, their presence ensures the existence of products that are eligible for non-GMO
certification without reformulation within a given food category, and these prod-
ucts may serve as a reliable counterfactual to help identify the effect of non-GMO
labeling. Additionally, each category exhibits good variation in non-GMO labeling
over time and has exhaustive coverage in the Nielsen data, helping ensure that the
24
empirical analysis will have reasonable identification power to provide meaningful
results.
Each category also has unique features that will aid in uncovering nuances
in the analysis. Ready to eat cereal has a long history of study in the empirical
industrial organization literature, beginning with the work of Scherer (1979) on
optimal product variety through the work of Richards and Hamilton (2015) on
variety pass-through. This may provide a benchmark for our analysis and help
guide future avenues of exploration. Snack chips have distinct varieties that are
either more or less likely to contain GMO ingredients, and this feature is likely
more salient to consumers than in other product categories. For example, tortilla
chips are primarily corn-based. Over 90% of corn in the U.S. is GMO, so these
products present a much more salient GMO “risk” to consumers. On the other
hand, potato chips are made mostly of potatoes, for which no commercially available
GMO varieties currently exist, and thus pose a lower “risk” to consumers.
Baby food represents a product category for which consumers may have a
heightened sensitivity to GMO presence; and, therefore, we may expect to see dif-
ferent purchasing behavior in this category. In particular, parents that perceive
GMO ingredients as posing some sort of health risk may pay a higher premium for
non-GMO in this category, since these food products are intended for their chil-
dren. On the other hand, baby food has long been dominated by a small number
of well-established conventional brands, and the reputations these firms have built
over time may overshadow non-GMO labeling. Other product categories pose dif-
fering levels of GMO risk as well. For example, products from categories such as
25
rice, chocolate, dried fruit, olive oil, nuts, tea, pasta, and dry seasoning, have no
commercially available GMO variants; however, in some cases the additives used in
processing may contain GMOs (e.g., soy lecithin used in chocolate, etc.). Lastly,
cooking oils are typically made from corn, soybean, and canola, all of which are
predominantly GMO in the U.S.
Figure 1.4 shows total national sales for Non-GMO Project Verified products
between 2010 and 2014, based on the retail scanner data for the selected product
categories. The figure also shows sales of products each year that became Non-
GMO Project Verified in a future calendar year, denoted “To Be Verified,” which
helps distinguish growth in the Non-GMO market from newly introduced products
versus existing products that become Non-GMO Project Verified. Sales on Non-
GMO Project Verified products more than doubled in 2011 and 2012, largely due
to growth in labeling among existing products. 2013 and 2014 also saw double-digit
percentage growth, attributable to both expansion of the overall Non-GMO market
as well more labeling among pre-existing products.
1.5 Empirical Model
For each product category in the analysis, the Nielsen Retail Scanner Data
contains weekly prices and quantities sold across the U.S. at the store and UPC
level. For each estimation, I restrict the sample to products that obtained the
Non-GMO Project Verified label between 2010 and 2014, with at least 6 months of
26
sales data prior to being certified and 12 months of sales data after certification.22
The rationale for this approach is based on two requirements. First, the empirical
specification relies on pre- and post-treatment indicators that I construct using
the product verification dates; therefore, it is critical that sufficient data exists
before and after the labeling event to estimate the effect of the label on prices and
quantities. Second, the sample needs to remain relatively stable to minimize the
confounding effects of product entry and exit on the model estimates. Restricting
the sample as I have done helps ensure both these conditions are met.23
1.5.1 Price Premium Regression Model
1.5.1.1 Main Specification
I use scanner data from 2009 to 2014 aggregated to the national level and
calculate a sales-weighted price per ounce pjklt, where j is a product UPC, k is
a manufacturer, l is a category, and t represents a particular week. Using the
verification dates for non-GMO products, I construct multiple treatment indicators
based on the timeframe before and after a product receives the non-GMO label to
estimate the average effect of labeling on prices for non-GMO food products and
22As a robustness check, I explore several specifications using the full sample, which includes
conventional products that were never certified, and found no significant deviations. Those results
are available in Appendix A.5.
23As a caveat to the subsequent analysis, note that post-treatment indicators beyond 12 months
are subject to changes in product mix.
27
explore dynamic effects of the label in greater detail,
log(pjklt) = ψ1PRE612jklt + ψ2PRE06jklt + ψ3POST06jklt + ψ4POST612jklt
+ ψ5POST1224jklt + ψ6POST24pjklt + ξj + ξt × ξk × ξl + jklt (1.5)
where each treatment indicator is a dummy variable that equals 1 if observation
week t is, respectively, 6 to 12 months prior, 0 to 6 months prior, 0 to 6 months
after, 6 to 12 months after, 12 to 24 months after, or greater than 24 months after
the verification date for product j; ξj, ξk, ξl, and ξt are product UPC, manufacturer,
category, and week fixed effects, respectively; and jklt is a random error term. To
control for any potentially confounding manufacturer- and category-level pricing
decisions, I allow the weekly intercepts to vary across manufacturer and category by
including Week×Manufacturer×Category fixed effects. The coefficients of interest,
ψ, measure the average price effect of non-GMO labeling in each time period. Since
I use a log-linear specification, the coefficients ψ can be interpreted as a percentage
change in the product price in each time period.
1.5.1.2 Organic Interaction
The National Organic Program, established in 2000, also prohibits the use of
GMO ingredients, effectively making USDA Certified Organic products a subset of
Non-GMO products. Therefore, a Non-GMO Project Verified label on an organic
product is somewhat redundant and does not necessarily provide new information,
so I would not expect to observe a price premium associated with it. Nonetheless,
28
nearly half of all Non-GMO Project Verified products are also Certified Organic,
suggesting that firms believe that consumers are not fully informed about the organic
standard or that the Non-GMO label bestows some additional value. There may
also be favorable cost considerations that influence a firm’s decision to seek out a
non-GMO label for organic products: firms have already invested in a Non-GMO
supply chain and incurred any associated reformulation costs for these products.
Furthermore, the supply chain has already been vetted to minimize adventitious
presence of GMO ingredients, so the likelihood of incurring any unforeseen costs
during the certification process is lower for organic products. Therefore, we expect
the cost of non-GMO certification for organic products to be less than that of non-
organic products; and, to the extent that certification costs are passed through to
the consumer, this will be reflected in price premiums.
Both of these factors support the hypothesis that Certified Organic products
will command lower price premiums after non-GMO certification than non-organic
products. I explore this with an additional price premium specification that includes
an organic indicator interacted with the treatment indicators:
log(pjklt) = ψ1PRE612jklt + ψ2PRE06jklt + ψ3POST06jklt + ψ4POST612jklt
+ ψ5POST1224jklt + ψ6POST24pjklt + ψ7PRE612jklt ×Orgj
+ ψ8PRE06jklt ×Orgj + ψ9POST06jklt ×Orgj + ψ10POST612jklt ×Orgj
+ ψ11POST1224jklt ×Orgj + ψ12POST24pjklt ×Orgj
+ ξj + ξt × ξk × ξl + jklt (1.6)
29
where Orgj is an indicator variable that equals one if product j is Certified Organic.
1.5.2 Quantity Regression Model
Depending on market conditions, firms may not be able to extract a price
premium by using the label; however, firms may use the non-GMO Project Veri-
fied label to capture market share from other products. To test for this possibility, I
regress weekly product sales quantities on the treatment indicators using a specifica-
tion similar to that used for the price premium regressions. I use scanner data from
2009 to 2014 aggregated to the national level and calculate weekly sales quantity
qjklt,where j is a product UPC, k is a manufacturer, l is a category, and t represents
a particular week. I construct the same time-period-based indicator variables based
on when a product receives the non-GMO label to estimate the average effect of
labeling on the sales quantity for non-GMO food products,
log(qjklt) = ψ1PRE612jklt + ψ2PRE06jklt + ψ3POST06jklt + ψ4POST612jklt
+ ψ5POST1224jklt + ψ6POST24pjklt + ξj + ξt × ξk × ξl + jklt (1.7)
where the treatment indicators are the same as in Equation 1.5; ξj is a product UPC
fixed effect; and ξt × ξk × ξl is a Week×Manufacturer×Category fixed effects. The
coefficient of interest, ψ, measures the average quantity effect of non-GMO labeling
in each time period. Since I use a log-linear specification, the coefficients ψ can be
interpreted as a percentage change in the weekly sold quantity in that time period.
30
1.5.3 Identification
Each of the specifications introduced above includes fixed effects to control for
unobserved heterogeneity across product UPC and week-manufactuer-category. The
product-UPC fixed effect controls for unobserved differences in product attributes
across UPCs. The week-manufacturer-category fixed effects essentially create weekly
intercepts to control for manufacturer-category level pricing changes. Therefore,
the identification strategy relies on variation in timing of non-GMO certification
for UPCs within each manufacturer-category. In other words, if every UPC for a
manufacturer-category is certified in the same week, the treatment effect cannot be
identified. I provide a measure of this variation in Table 1.2. For each product cat-
egory, I calculate the average number of weeks between the first and last non-GMO
product certification for each manufacturer. With the possible exception of olive
oil, there is sufficient variation in certification timing in all product categories for
identification. Of course, the standard identifying assumption also applies: unob-
served factors that could simultaneously affect price or quantity sold and non-GMO
certification are time-invariant.
31
1.6 Results
1.6.1 Price Premiums
1.6.1.1 Main Results
Table 1.3 presents the price premium regression results based on the model
specified in the Equation 1.5. Columns I and II present alternate specifications with
a progression of fixed effects, and Column III is the preferred specification. In the
first specification with UPC and Week×Category fixed effect, I estimate coefficients
for the pre- and post-treatment indicators that are consistently negative and sta-
tistically significantly different from zero. The estimates for pre-certification 6-12
months and pre-certification 0-6 months indicate about a 1% decrease in price lead-
ing up to the certification event. After certification, the price decreases by about a
3% in the first 0-6 months and becomes more negative over time.24 In the second
specification with UPC and Week×Manufacturer fixed effects, the point estimates
for the coefficients are negative as well, but most of them are not statistically signif-
icantly different from zero; and, furthermore, the estimates are heavily attenuated.
The fact that the Week×Manufacturer fixed effect absorbs much of the significant
negative treatment effect observed in the first specification suggests that firms may
be engaging in manufacturer-level pricing decisions that were biasing the previous
results.
24Based on the data construction, the coefficient estimates for the post-certification 12-24 months
and 24+ months indicators may be biased by changes in product mix, since the sample only
guarantees 12 months of post-certification data for a given UPC.
32
In the final specification with the full suite of UPC and Week×Category×Manufacturer
fixed effects, the treatment effect vanishes, and none of the coefficient estimates are
statistically significantly different from zero. Moreover, while the point estimates are
still slightly negative, they are further attenuated towards zero and lack economic
significance. These results show no evidence of dynamic pricing effects, either; which
is to say that the treatment effect does not evolve over the post-certification time
period. The progression of results across specifications suggests that firms may en-
gage in manufacturer and category specific pricing strategies; but once we control
for this behavior, we do not find evidence that firms are using the Non-GMO Project
Verified certification to extract price premiums on pre-existing products.
There are a number of reasons firms may not use the Non-GMO Project Ver-
ified label to extract price premiums for existing products, some of which were
discussed in prior sections of this paper. The stylized data presented in Section 1.4
suggests that non-GMO food products occupy a higher-priced food segment to begin
with, so it is possible that firms already enjoy large profit margins on these product
and cannot increase prices without losing market share. Additionally, we observe
that a significant portion of Non-GMO Project Verified products receive certifica-
tion prior to market entry, and these products may launch at a higher price point
on average, relative to existing Non-GMO Project Verified products. Therefore,
another possibility is that firms are recouping costs and exercising market power
through new product entry.
33
1.6.1.2 Organic Interaction Results
Table 1.4 presents the price premium regression results based on Equation 1.6
that includes an organic indicator interaction term with each of the treatment indi-
cators. Once again, Column I and II contain results for an alternate specifications
with UPC and Week×Category fixed effects and with UPC and Week×Manufacturer
fixed effects, respectively. The preferred specification in Column III employs the
full suite of fixed effects from Equation 1.6. The progression of results across
specifications is very similar to that presented in the previous section, with the
Week×Manufacturer fixed effect absorbing some manufacturer-level pricing behav-
ior in the second specification.
Focusing on the final specification, the main treatment indicator coefficient es-
timates are not statistically significantly different from zero, and the point estimates
are very close to zero, which is consistent with our results from the main specifica-
tion. To interpret the organic interaction, the coefficient estimates for the main and
interaction terms should be added together.25 The point estimates for the organic in-
teraction terms are all slightly negative, but only the post-certification 12-24 month
interaction term is statistically significantly different from zero.26 Therefore, I can-
not conclude that organic products command smaller price premiums for non-GMO
certification than non-organic products.
25Because I include UPC fixed effects, a separate, time-invariant organic indicator term cannot
be estimated.
26Given the potentially confounding product mix effects after 12 months of certification, this
result warrants some skepticism.
34
1.6.2 Quantity Sold
While our results do not indicate that firms use the non-GMO certification to
extract price premiums for existing products, firms may use the certification to sell
more units of non-GMO products. For single-product firms, any increase in quantity
sold directly increases profits so long as the product has a positive profit margin.
In the case of multi-product firms, if these firms seek non-GMO certification for
products that command higher profit margins, then any increased market share for
these products will also lead to increased profits.
Table 1.5 presents results for the quantity regression estimates based on Equa-
tion 1.7. As with the price premium regressions, Column I, II, and III contain results
for a progression of fixed effects, with the preferred specification contained in Col-
umn III. In the first specification, the estimates for the post-certification 0-6 months
and 6-12 months treatment indicators are positive and statistically significantly dif-
ferent from zero, suggesting that firms may increase quantity sold after certification
for non-GMO products. However, to the extent that firms engage in manufacturer-
level marketing strategies, this specification will produce biased results. Once we
control for Week×Manufacturer fixed effects in the second specification, the results
lose statistical significance, although the point estimates are still large and positive.
In the preferred specification, while the point estimates for the post-certification
treatment indicators remain positive, none of the estimates are statistically signifi-
cantly different from zero. As such, our results do not provide conclusive evidence
that firms use the non-GMO certification to increase the quantity of products sold.
35
1.7 Alternative Firm Strategies
In the preceding results, I find no evidence that firms use non-GMO certifi-
cation to extract price premiums or increase quantities sold for pre-existing, newly
certified products. The certification may, however, induce other firm strategies such
as new non-GMO product development targeted to specific consumers by which
firms could extract rent and pass the certification cost to consumers. To explore
this possibility, I first show that a significant portion of non-GMO products obtain
the Non-GMO Project Verification prior to appearing in the retail scanner data and
therefore represent new product introductions. I augment this with some descriptive
statistics that may support the notion that firms price non-GMO products certified
within their product life differently from new products certified before market entry.
Then I provide descriptive statistics for consumer demographics to highlight differ-
ences between consumers that purchase pre-existing non-GMO products, newly in-
troduced non-GMO products, and non-certified products, which suggests that firms
may target new non-GMO product introductions to a specific type of consumers.
1.7.1 Timing of Certification
Figure 1.5 illustrates the number of months a non-GMO product is on the mar-
ket prior to receiving the Non-GMO Project Verification. A negative value indicates
that a product obtained Non-GMO Project Verification prior to appearing in the
Nielsen retail scanner data. A significant portion of products in each food category
(20% on average) receive certification before entering the market, suggesting that
36
firms may use the label to facilitate new product development and increase product
diversity, thereby exercising market power through second-degree price discrimina-
tion.
To delve more into Figure 1.5, Table 1.6 presents a comparison of average
prices by product category for Non-GMO Project Verified UPCs, based on whether
the product already existed in the Nielsen data prior to certification or was newly
introduced after certification. “Pre-Existing” products consists of post-certification
data for products that are Non-GMO certified and have at least 6 months of sales
history prior to certification and 12 months after certification. “New Entry” prod-
ucts consists of the first 3 months of post-certification data for products that are
Non-GMO certified and became certified prior to appearing in the Nielsen data. The
second column shows the percentage of manufacturers in each category for which
the mean price of their new entry products exceeded the price of their pre-existing
products. The data indicates that, for many product categories, the majority of
manufacturers introduced new products at prices greater than those of pre-existing
products, further suggesting that firms may use new product entry as a means to
extract rent and pass the certification costs for Non-GMO Project Verified products
to consumers.
1.7.2 Targeting to Non-GMO Consumers
If firms are potentially developing new non-GMO products and introducing
them at higher price points than their existing product line, are these products
37
being targeted to a specific type of consumer? From a future policy standpoint, it is
important to understand whether voluntary non-GMO labeling disproportionately
affects a particular consumer segment, and whether that impact is beneficial or
harmful.27 To provide some context, I explore how non-GMO consumers differ, on
average, from other consumers for the food products in this study.
I use Nielsen Consumer Panel data from 2009 to 2014 for all purchases made
in the relevant product categories. Each record represents a household’s purchase
of a particular product on a specific trip to a store. I calculate mean values for
several household demographic variables (income, household size, graduate educa-
tion, presence of children) across the data, by product non-GMO certification and
new entry status (i.e. products certified prior to entering the market, as dicussed
in the previous section). The summarized data reflect product- and trip-weighted
statistics for household demographics.
Table 1.7 presents results for the consumer demographic analysis. In aggregate,
across all 18 food categories, households that consume non-GMO food products
tend to be wealthier, smaller, more educated, and less likely to have children. These
trends are even more pronounced for consumers of new entry, non-GMO products,
further suggesting that firms may target different market segments for new and
pre-existing non-GMO products. This evidence, while suggestive, is consistent with
the hypothesis that firms strategically introduce new non-GMO certified products
to facilitate second-degree price discrimination and pass the non-GMO certification
27As a concrete example involving another food policy issue, similar concerns have been raised
regarding the soda tax in New York City, which many regard as a regressive tax that is unduly
burdensome to households of low socioeconomic status.
38
cost to consumers.
1.8 Conclusion
In recent years, GMO food labeling has become a mainstream debate in the
U.S. While the U.S. organic Standard provides an indirect form of non-GMO labeling
for Organic food products by prohibiting the use of GMOs, a voluntary verification
and labeling scheme for non-GMO products called the Non-GMO Project emerged in
the U.S. in 2010. It has grown rapidly to include 35,000 products, accounting for over
$13.5 billion in annual sales as of 2015. In this paper, I investigate whether firms use
a voluntary, third-party quality certification to exercise market power by extracting
price premiums or increasing quantity sold on newly certified products, and whether
those effects persist over time. In particular, I use a hedonic framework to estimate
price premiums and quantity changes for newly certified non-GMO food products
using the Non-GMO Project Verified label in the U.S. I exploit a unique dataset from
the Non-GMO Project that contains verification dates for products throughout the
label’s history, coupled with weekly retail point-of-sale data from 2009 to 2014 for a
large sample of supermarkets across the U.S. I find no statistically significant price
premiums or quantity changes for newly certified non-GMO food products in the
food categories examined. I, however, find suggestive evidence that the label may
induce other firm strategies such as new non-GMO product development targeted
to specific consumers by which firms could extract rent and pass the certification
cost to consumers.
39
The findings in the paper warrant more in-depth analysis along two fronts.
First, while firms do not appear to extract price premiums or increase quantities sold
for newly certified non-GMO products that already exist in their product line, some
evidence suggests that firms may exercise market power through new non-GMO
product introduction. Exploring this type of behavior is best suited to a rigorous
structural model that can account for market structure and firm branding strategy.
Second, consumers clearly differ in their preferences for non-GMO products, and,
therefore, the behavioral effect of the Non-GMO Project Verified certification is
not entirely straightforward. Furthermore, if firms behave strategically, perhaps
exploiting the non-GMO certification to increase profits through second degree price
discrimination, the welfare implications of the certification are also unclear. To
quantify these effects, a structural demand model that captures heterogeneity in
consumer preferences for non-GMO products is essential.
40
Source: The Non-GMO Project, 2016.
Figure 1.1: Non-GMO Project Verified Label
41
Note: Product counts are not unique by package specification.
Figure 1.2: Cumulative Monthly Non-GMO Project Verified Products by Organic
Status
42
Note: Product counts are not unique by package specification.
Figure 1.3: Growth in Non-GMO Project Verified Products by Product Category
43
Figure 1.4: Annual Non-GMO Project Verified Product Sales
44
Note: Product ages are capped at 60 months based on the time span of the data.
Figure 1.5: Product Age When Non-GMO Project Verified
45
Table 1.1: Summary Statistics
Product Category Total
UPCs
Mfrs. Organic
UPCs
Non-
GMO
Verified
UPCs
Mean
Price
($/oz)
BABY FOOD - STRAINED 939 26 442 198 0.19
CANDY-CHOCOLATE 12868 1237 393 153 0.35
NUTS - BAGS 5655 626 86 153 0.44
SNACKS - POTATO CHIPS 5871 252 13 154 0.29
CEREAL - GRANOLA & NATURAL 914 197 106 126 0.24
CEREAL - READY TO EAT 2923 137 186 240 0.20
COOKIES 15886 1655 180 172 0.23
FRUIT-DRIED AND SNACKS 3840 496 320 262 0.34
FRUIT DRINKS-OTHER CONTAINER 6433 815 310 141 0.03
GRANOLA & YOGURT BARS 3264 310 334 229 0.37
OLIVE OIL 1811 487 103 72 0.28
PASTA-SPAGHETTI 1335 289 114 41 0.09
RICE - PACKAGED AND BULK 1418 375 84 151 0.07
SALAD AND COOKING OIL 1062 351 85 90 0.08
SEASONING-DRY 12184 1422 633 410 0.84
SNACKS - TORTILLA CHIPS 2353 346 54 154 0.24
TEA - BAGS 2482 300 379 117 0.09
TEA - HERBAL BAGS 1996 263 340 185 0.17
46
Table 1.2: Manufacturer-Category Variation in Certification Timing
Product Category Average Weeks
b/t Certification
BABY FOOD - STRAINED 46.51
CANDY-CHOCOLATE 24.87
NUTS - BAGS 9.51
SNACKS - POTATO CHIPS 7.32
CEREAL - GRANOLA & NATURAL 2.13
CEREAL - READY TO EAT 23.48
COOKIES 21.75
FRUIT-DRIED AND SNACKS 14.49
FRUIT DRINKS-OTHER CONTAINER 20.53
GRANOLA & YOGURT BARS 8.70
OLIVE OIL 0.41
PASTA-SPAGHETTI 5.33
RICE - PACKAGED AND BULK 9.56
SALAD AND COOKING OIL 28.38
SEASONING-DRY 32.86
SNACKS - TORTILLA CHIPS 20.55
TEA - BAGS 20.93
TEA - HERBAL BAGS 23.70
47
Table 1.3: Price Premium Regressions
I II III
Pre-Cert. 6-12 Mos. −0.010∗∗ −0.012∗ −0.010
(0.003) (0.006) (0.007)
Pre-Cert. 0-6 Mos. −0.013∗∗ −0.007 −0.001
(0.005) (0.008) (0.009)
Post-Cert. 0-6 Mos −0.033∗∗∗ −0.018· −0.013
(0.006) (0.011) (0.011)
Post-Cert. 6-12 Mos. −0.038∗∗∗ −0.028∗ −0.021
(0.007) (0.013) (0.013)
Post-Cert. 12-24 Mos. −0.045∗∗∗ −0.022 −0.011
(0.009) (0.017) (0.018)
Post-Cert. 24+ Mos. −0.062∗∗∗ −0.036 −0.018
(0.012) (0.022) (0.022)
UPC FEs Yes Yes Yes
Week FEs Yes Yes Yes
× Category Yes No Yes
× Manufacturer No Yes Yes
Adj. R2 0.986 0.989 0.989
Num. obs. 351052 351052 351052
Note: Each column represents a separate regression. Standard
errors are clustered at the product level in parentheses: ∗∗∗p <
0.001, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1.
48
Table 1.4: Price Premium Regressions with Organic Interaction
I II III
Pre-Cert. 6-12 Mos. 0.006 −0.003 −0.001
(0.006) (0.010) (0.011)
Pre-Cert. 0-6 Mos. 0.007 −0.002 0.005
(0.007) (0.012) (0.012)
Post-Cert. 0-6 Mos −0.013 −0.009 −0.004
(0.008) (0.014) (0.015)
Post-Cert. 6-12 Mos. −0.027∗∗ −0.015 −0.008
(0.009) (0.016) (0.017)
Post-Cert. 12-24 Mos. −0.034∗∗ 0.003 0.014
(0.011) (0.020) (0.022)
Post-Cert. 24+ Mos. −0.053∗∗∗ −0.024 −0.010
(0.014) (0.024) (0.025)
Pre-Cert. 6-12 Mos. × Organic −0.028∗∗∗ −0.018 −0.017
(0.008) (0.011) (0.012)
Pre-Cert. 0-6 Mos. × Organic −0.035∗∗∗ −0.008 −0.010
(0.009) (0.011) (0.012)
Post-Cert. 0-6 Mos × Organic −0.035∗∗∗ −0.016 −0.016
(0.009) (0.012) (0.012)
Post-Cert. 6-12 Mos. × Organic −0.019· −0.022· −0.021
(0.010) (0.013) (0.013)
Post-Cert. 12-24 Mos. × Organic −0.019· −0.042∗∗ −0.043∗∗
(0.010) (0.013) (0.014)
Post-Cert. 24+ Mos. × Organic −0.015 −0.017 −0.013
(0.011) (0.013) (0.014)
UPC FEs Yes Yes Yes
Week FEs Yes Yes Yes
× Category Yes No Yes
× Manufacturer No Yes Yes
Adj. R2 0.986 0.989 0.989
Num. obs. 351052 351052 351052
Note: Each column represents a separate regression. Standard errors are
clustered at the product level in parentheses: ∗∗∗p < 0.001, ∗∗p < 0.01,
∗p < 0.05, ·p < 0.1.
49
Table 1.5: Quantity Regressions
I II III
Pre-Cert. 6-12 Mos. 0.044 −0.020 −0.048
(0.037) (0.054) (0.056)
Pre-Cert. 0-6 Mos. 0.062 −0.011 −0.040
(0.049) (0.077) (0.078)
Post-Cert. 0-6 Mos 0.138∗ 0.060 0.027
(0.062) (0.098) (0.097)
Post-Cert. 6-12 Mos. 0.144· 0.151 0.120
(0.077) (0.123) (0.119)
Post-Cert. 12-24 Mos. 0.116 0.218 0.187
(0.103) (0.164) (0.160)
Post-Cert. 24+ Mos. 0.164 0.300 0.277
(0.143) (0.224) (0.211)
UPC FEs Yes Yes Yes
Week FEs Yes Yes Yes
× Category Yes No Yes
× Manufacturer No Yes Yes
Adj. R2 0.858 0.901 0.903
Num. obs. 351052 351052 351052
Note: Each column represents a separate regression. Standard
errors are clustered at the product level in parentheses: ∗∗∗p <
0.001, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1.
50
Table 1.6: Non-GMO Certification of New and Pre-Existing Food Products
Product Category Pct. Mfrs.
New Entry >
Pre-Existing
Mean Price
(%)
New
Entry
UPCs
Pre-
Existing
UPCs
BABY FOOD - STRAINED 100.0 57 81
CANDY-CHOCOLATE 25.0 46 67
NUTS - BAGS 80.0 30 85
SNACKS - POTATO CHIPS 66.7 31 83
CEREAL - GRANOLA & NATURAL 100.0 42 56
CEREAL - READY TO EAT 70 68 121
COOKIES 55.6 59 91
FRUIT-DRIED AND SNACKS 44.4 65 87
FRUIT DRINKS-OTHER CONTAINER 60.0 38 80
GRANOLA & YOGURT BARS 33.3 55 87
OLIVE OIL 50.0 12 24
PASTA-SPAGHETTI 0.0 14 19
RICE - PACKAGED AND BULK 75.0 52 73
SALAD AND COOKING OIL 60.0 25 52
SEASONING-DRY 33.3 26 228
SNACKS - TORTILLA CHIPS 50.0 25 84
TEA - BAGS 100.0 22 75
TEA - HERBAL BAGS 62.5 33 96
51
Table 1.7: Average Consumer for Conventional & Non-GMO Products
Non-GMO Product Mean
Inc.
Median Inc. HH
Size
Grad
Edu.
Child
No All $65607 [$50K, $60K) 2.70 0.16 0.30
Yes Pre-Existing $68508 [$50K, $60K) 2.60 0.20 0.28
Yes New $77277 [$60K, $70K) 2.53 0.25 0.26
52
Chapter 2: The Impact of Voluntary Non-GMO Labeling on De-
mand in the Ready-to-Eat Cereal Industry∗
2.1 Introduction
In markets with asymmetric information, quality disclosure can lead to effi-
ciency gains that benefit consumers and producers (Grossman 1981; Dranove and Jin
2010). Firms may also use quality certification to influence perceived product qual-
ity and exercise market power. This paper examines the impact of voluntary quality
certification on demand in the ready-to-eat [RTE] cereal industry, using evidence
from the Non-GMO Project Verified food label. While past studies have investigated
firms’ use of non-price marketing strategies such as advertising, couponing, and new
product introductions in the RTE cereal industry (Thomas 1999; Nevo 2001; Nevo
and Wolfram 2002), this paper is the first to examine the role of voluntary quality
certification as a marketing strategy in this industry.
Prior hedonic analysis does not find evidence of price premiums or quantity
changes for newly certified non-GMO food products across 18 food categories, in-
cluding RTE cereal (Adalja 2016). In this study, I estimate a discrete-choice, random
∗Nielsen data is provided by the Data Center at The University of Chicago Booth
School of Business. Information on availability and access to the data is available at
http://research.chicagobooth.edu/nielsen.
53
coefficients logit demand model (Berry et al. 1995; Nevo 2001) with monthly Nielsen
Retail Scanner data for 50 breakfast cereal brands in 100 DMAs between 2010 and
2014. I use the model to examine the impact of voluntary non-GMO labeling on
demand for RTE cereal and to characterize heterogeneity in consumer tastes for non-
GMO labeling. The results indicate that consumer preferences vary significantly for
the non-GMO label, and this heterogeneity affects individual choices. In aggregate,
the non-GMO label positively impacts demand on average.
I use the structural parameters recovered from the demand estimation along
with an assumed model of firm behavior to calculate price-cost margins. I use these
results to simulate welfare effects for two different labeling regimes in the RTE
cereal industry: one in which all brands use the non-GMO label and one in which
no brands use the label. I analyze changes in producer and consumer welfare by
analyzing changes in firm profit and individual compensating variation, respectively.
The simulation results suggest that non-GMO labeling in the RTE cereal industry
may improve consumer surplus but reduce industry profit on average.
Several factors make the RTE cereal food category well suited for estimating
the effects of voluntary non-GMO food labeling on demand. First, there exists
substantial variation in non-GMO labeling across time and products in this category.
Second, the data that I use have an exhaustive coverage of the purchases for these
products. Finally, RTE cereal has a long history of study in empirical industrial
organization, so parameter estimates are readily available in the literature with
which to benchmark my model estimates.
The paper is structured as follows. Section 2.2 discusses the literature on
54
the RTE cereal industry, willingness to pay for non-GMO, and demand estimation.
Section 2.3 presents the theoretical framework for the structural economic model
and welfare analysis. Section 2.4 describes the data sources I employ to implement
this study. Section 2.5 highlights the empirical strategy I use to estimate the demand
system. Section 2.6 presents parameter estimates recovered from the model and the
simulated welfare effects of non-GMO labeling. Lastly, Section 2.7 offers concluding
remarks as well as opportunities for future extensions to the analysis.
2.2 Background Literature
2.2.1 Demand Estimation in Markets with Differentiated Products
Demand estimation has a long history in economics research dating back to
Stone (1954), but estimating structural demand models for differentiated products
has historically posed several obstacles. First, due to the large number of products,
simply estimating a system of demand equations is empirically intractable due to the
large number of parameters to estimate. Another problem that must be addressed
is consumer preference heterogeneity, of which the proliferation of differentiated
products is a manifestation. Discrete choice models such as the logit demand model
(McFadden 1973) circumvent the dimensionality issue by transforming each product
into a bundle of relevant attributes, thereby shrinking the dimensions of the system.
However, due to the model assumptions, the standard logit model carries with it
strong restrictions on consumer substitution patterns.
Of the many advances that have occurred since then, the methodology devel-
55
oped in Berry (1994) and Berry et al. (1995) [BLP] represents a significant inno-
vation for demand estimation in differentiated-products markets. In this seminal
work, BLP estimates a random-coefficients discrete choice demand model by us-
ing a fixed-point contraction mapping to solve the system numerically. The BLP
model incorporates preference heterogeneity and allows for more flexible substitu-
tion patterns, but these features come at the expense of significant computational
complexity. Thanks in part to Nevo (2000b), who made MATLAB code for BLP
estimation publicly available, the BLP model has seen wide use in empirical indus-
trial organization, underlying much of the work in demand estimation conducted
over the past two decades.
Since then, several additional improvements have been made within this esti-
mation framework. Nevo (2001) extended this model by incorporating panel data
techniques such as brand fixed effects and partially relaxing parametric assumptions
for preference heterogeneity by drawing from an empirical distribution of consumers.
Petrin (2002) uses average consumer data relating demographics to purchase prob-
ability to fit additional moment restrictions, thus obtaining more precise estimates
of structural parameters. Berry et al. (2004) uses detailed consumer-level data on
consumers’ second choices to fit three sets of moments, thus providing an alterna-
tive source of identification. Despite its widespread use, the BLP estimation tech-
nique presents significant numerical challenges. Dube´ et al. (2012) and Knittel and
Metaxoglou (2014) both conduct comprehensive analyses of the algorithms used to
estimate random-coefficients logit demand models and show that results vary widely
depending on the algorithm, starting values, and tolerances. As such, it is critical
56
that researchers exercise extreme caution when estimating these types of models.
2.2.2 Ready-to-Eat Cereal Industry
There has been sustained interest among economists in the RTE cereal in-
dustry since the 1970s, when the FTC brought an antitrust suit against Kellogg,
General Mills, and Post. The industry exhibits some of the classic traits of a differ-
entiated oligopoly—high concentration, enormous brand proliferation, and frequent
new product introductions. The early literature on the RTE cereal industry directly
addresses the antitrust concerns raised by the FTC. Schmalensee (1978) use’s the
Hotelling model to analyze firm conduct in the RTE cereal industry and argues that
frequent new product introductions by incumbent firms serve to protect profits and
deter new entry in the industry.2 Scherer (1979) looks at new product introduc-
tions as well, but from a welfare perspective. He provides evidence suggesting that
product variety is overstimulated and, based on launching costs, very likely welfare
reducing at the margin.
More recent literature on the RTE cereal industry tends to focus on either
price or non-price marketing strategies. Thomas (1999) examines firm response to
entry and finds that incumbent response depends on the scale of entry, and firms use
advertising and new product introductions for entry deterrence in addition to price.
Nevo (2001) uses a BLP approach to measure market power in the RTE cereal indus-
try. He finds that the high price-cost margins in the industry are largely explained
2This paper is based upon the author’s expert testimony as a government witness in the afore-
mentioned FTC antitrust case.
57
by product differentiation and multi-product firm pricing rather than collusive be-
havior, suggesting that any market power is attributable a firm’s product portfolio
and advertising. Nevo (2000a) uses the same model to analyze the effects of mergers
in the industry by using the structural parameters recovered from BLP estimation
to simulate new price equilibria and welfare changes of various merger scenarios (two
of which actually occurred). Nevo and Wolfram (2002) analyze the relationship be-
tween shelf prices and coupons in the RTE cereal industry. Interestingly, they find a
negative correlation between prices and availability of manufacturer coupons. They
present evidence that this behavior is driven by strategic interaction between firms,
manager incentives, and the effects of coupons on repeat purchase.
In an effort to address both price and non-price strategies, Richards and Pat-
terson (2006) uses a dynamic setting to examine strategic interaction between firms.
They find that firms tend to price and choose product lines cooperatively in the
static setting; but, with dynamic interactions, firms behave more competitively
along both dimensions. Chidmi (2012) examines vertical relationship between re-
tailers and manufacturers in the RTE cereal industry to shed light on retail pricing
decisions. He estimates different supply models using demand parameters recov-
ered from BLP estimation with data from four supermarket chains in the Boston
area. The results imply that manufacturers make pricing decisions and retailers do
not intervene (i.e. retailer margins are zero), thus avoiding double-marginalization.
Richards and Hamilton (2015) examines pass-through of wholesale price changes
into retail prices and product lines of firms in the RTE cereal industry. By account-
ing for the endogeneity of product line decisions for multi-product firms, they find
58
evidence that wholesale price changes are passed through one-to-one to retail prices.
2.2.3 Willingness-to-Pay for Non-GMO
Empirical studies of willingness-to-pay [WTP] for non-GMO labeling have de-
cidedly mixed findings. Most studies employ surveys and lab experiments to analyze
consumer preferences for GMO products. Lusk et al. (2005) identifies 25 separate
studies that together provide 57 estimates of consumers’ WTP for GMO food prod-
ucts and finds significant variation in the estimates. Price premiums for non-GMO
food ranged from -68% to 784%, with an average of 42%, and are significantly
affected by elicitation method.
Recent studies attempt to shed light on the source of this variation. Using
data from a nationwide survey, Onyango et al. (2006) finds that consumers place
a 10% premium on food labeled as non-GMO and 6.5% discount on food labeled
as GMO; but, interestingly, consumers also attach a 5% premium for food labeled
GMO if the label also specifies “USDA approved” or “to reduce pesticide residues
in your food.” Roe and Teisl (2007) uses a survey to elicit consumer reactions
to 80 different GMO label variations and finds that labels with simple claims and
claims certified by the FDA are most credible. Costanigro and Lusk (2014) conducts
a series of choice experiments and finds evidence that consumer WTP to avoid
GMO food is 140% higher with a mandatory “contains” GMO label compared to a
voluntary “does not contain” GMO label. Lastly, Adalja (2016) uses national retail
scanner data for 18 food categories coupled with labeling data from the Non-GMO
59
Project Verified voluntary label and finds no evidence of price premiums for non-
GMO products; however, suggestive evidence supports the hypothesis that the label
induces incumbent firms to introduce new products.
2.3 Conceptual Framework
The conceptual approach as well as the empirical strategy for the structural
model follows very closely with the framework of Nevo (2001), so I present an
abbreviated treatment here using the same notation.
2.3.1 Consumer Demand with Heterogeneous Preferences
Consider an economy in which we observe t = 1, . . . , T markets, each with
i = 1, . . . , It consumers and j = 1, . . . , J products with average prices pjt. The
indirect utility of consumer i from consuming product j in market t is
uijt = xjtβi − αipjt + ξjt + εijt, (2.1)
where xjt is a K-dimensional vector of observable product characteristics, ξjt is the
unobserved product characteristic, (βi, αi) are K+ 1 individual-specific coefficients,
and εijt is a mean-zero stochastic term. The unobserved product characteristic
ξjt can be further decomposed as ξjt = ξj + ξt + ∆ξjt, where ξj and ξt can be
captured empirically with brand and time dummies, respectively, in which case xjt
only contains time-varying product characteristics. The indirect utility from the
outside option is normalized to zero.
60
Consumer preferences depend on individual demographics D and unobserved
individual characteristics v, which are formally modeled as a
 αi
βi
 =
 α
β
+ ΠDi + Σvi, vi ∼ Pv(v), Di ∼ PD(D), (2.2)
where Di is a d × 1 vector of demographics that follow the distribution PD, vi is
a K + 1 vector of mean-zero normally distributed unobservables that follow the
distribution Pv, Π is a (K + 1) × d matrix of coefficients that measure how tastes
(for observable characteristics) vary with demographics, and Σ is a (K+1)×(K+1)
matrix of parameters. If we observe individual demand data, we can use such data
to characterize Di nonparametrically by drawing from an empirical distribution PˆD
such as the Current Population Survey [CPS] or the Nielsen Consumer Panel.
The set of individual characteristics that lead to product choice j are implicitly
defined by
Ajt(x, p·t, δ·t; θ2) = {(Di, vi, εit)|uijt ≥ uilt}, ∀ l = 1, . . . , J.
If we assume that D, v, and ε are independent, the market share for product j is
the integral
sjt(x, p·t, δ·t; θ2) =
∫
Ajt
dP (D, v, ε) =
∫
Ajt
dPε(ε)dPv(v)dPD(D), (2.3)
which can be computed either analytically or numerically depending on the distri-
61
butional assumptions made on D, v, and ε.
2.3.2 Firm Behavior
Assume there are f = 1, . . . , F firms, and each firm produces a subset Ff of
the J products. Profit is calculated as
Πf =
∑
j∈Ff
(pj −mcj)Msj(p)− Cf ,
where sj(p) is the market share of product j, M is the market size, and Cf is the
fixed cost. Assuming a Bertrand-Nash equilibrium in prices, the first order condition
with respect to price for each product j is
sj(p) +
∑
r∈Fr
(pr −mcr)∂sr(p)
∂pj
= 0.
We can construct the J × J price derivative matrix S where each element Sjr =
−∂sr(p)
∂pj
and the J × J ownership matrix Ωo where each element Ωjr = 1 if products
r and j are owned by the same firm or zero otherwise. If I define the matrix Ω as
the Hadamard product of Ωo and S and express for s, p, and mc as J × 1 vectors, I
can solve for the price-cost margins as
p−mc = Ω−1s(p). (2.4)
62
Once demand parameters are recovered, Equation 2.4 can be used estimate marginal
costs for each brand.
2.4 Data
Estimating a demand system for differentiated products requires, at a mini-
mum, marketing data with prices, market share, and product characteristics across
several markets in the U.S. These data typically consist of consumer panel data,
aggregate market-level data, or both. Models that use individual data account for
consumer heterogeneity and allow for a high level of product differentiation. They
also circumvent price endogeneity issues common to aggregate demand models (e.g.,
Goldberg 1995). That said, an aggregate industry model explicitly addresses supply
side and equilibrium considerations. Ideally, an estimation strategy that combines
both approaches can enrich the analysis by addressing demand, supply, and market
equilibrium together (e.g., Nevo 2001; Petrin 2002; Berry et al. 2004).
I use month-DMA-brand-level data between 2010 and 2014 (each market is
a DMA-month, for a total of 5,988 markets) on prices, market shares, and brand
characteristics from the Nielsen Retail Scanner Data; and I combine it with non-
GMO labeling data from the Non-GMO Project. Additionally, to complete the
demand system, I must specify the market share for the outside good in each mar-
ket, which requires an estimate of overall market size. I use data from the U.S.
Census Bureau’s Annual Estimates of the Resident Population for Counties along
with household sales data for RTE cereal from the Nielsen Consumer Panel to con-
63
struct this estimate. Lastly, I draw consumer demographics from the CPS Annual
Supplement.
2.4.1 Nielsen Retail Scanner and Consumer Panel Data
The Nielsen Retail Scanner data contains weekly, UPC-level quantity and
price data from retail store point-of-sale systems for 35,000 retail stores covering
more than half the total sales volume of grocery stores across the U.S. The full
dataset contains 2.6 million UPCs, representing 1,100 Nielsen product categories.
RTE cereal represent one such product category. I aggregate the RTE cereal data
by month, DMA, and brand (e.g., General Mills Honey Nut Cheerios is one brand,
etc.). Volume sales data is standardized by converting quantity to ounces sold, and
market share is calculated by dividing ounces sold by potential market size (see
Section 2.4.3 for details on the derivation of market size). A standardized price
variable is calculated as total dollar sales divided by ounces sold, and real prices are
adjusted using the U.S. average monthly urban CPI for breakfast cereal.
If one was simply interested in estimating demand for the major brands in
the RTE cereal industry, an appropriate way to choose brands would be to select
the J brands with the top national market share. However, since I am primarily
interested in examining how a voluntary non-GMO label affects demand for RTE
cereal, it is critical that the brands chosen accurately reflect the market for non-
GMO RTE cereal. Accordingly, the brands used in the demand estimation must
include those which started using the non-GMO label between 2010 and 2014 and
64
unlabeled brands that may be considered viable substitutes for these products.
I use information from Nielsen Consumer Panel to select brands to use in
the demand estimation. The Nielsen Consumer Panel data contains trip-UPC-level
purchase and pricing data for a nationally-representative panel of 40,000 to 60,000
U.S. households, covering the same product categories as the Retail Scanner data
for all major retail channels. To determine the relevant brands, I first identify
households that purchased newly launched Non-GMO Project Verified RTE cereal
brands in 2014.3 I then examine the portfolio of RTE cereal brands previously
purchased by these same households, and I choose the top 50 brands based on
projected volume purchased between 2010 and 2014. I then restrict the total number
of markets in my final dataset by selecting the 100 DMAs with the highest total
volume of sales for these 50 brands. Table 2.1 presents summary statistics by brand
for the variables used in the estimation.
2.4.2 Non-GMO Project
To estimate the effect of a non-GMO food label on demand for RTE cereal, I
have secured a unique, UPC-level monthly dataset of non-GMO products from the
Non-GMO Project4 that identifies the date each product began using the Non-GMO
Project Verified label. The Non-GMO Project began offering third-party verification
and labeling for food products in 2010, and the dataset spans 2010 to 2014. Usage
of the label has grown rapidly since 2010; the Non-GMO Project currently verifies
3These are brands that entered the market between 2010 and 2014 with Non-GMO Project
Verification already in place prior to appearing in the Nielsen data.
4See Adalja (2016) for a comprehensive background on the Non-GMO Project.
65
over 3,000 brands that represent more than 43,000 products and $19.2 billion in
sales (Non-GMO Project 2017). Figure 2.1 shows total national sales for Non-GMO
Project Verified RTE cereal products between 2010 and 2014, based on the Nielsen
Retail Scanner data. To help distinguish between growth from newly introduced
products and existing products, the figure also show sales for products verified in
a future calendar year, denoted “To Be Verified.” I merge this data by UPC with
Nielsen Retail Scanner data to clearly identify the month in which RTE cereals
began using the Non-GMO Project Verified label.
2.4.3 Market Size
We do not directly observe the market share for the outside good—it is calcu-
lated as the total market size less the market share of the inside goods. We must,
therefore, define the total market size. This is generally done by choosing an observ-
able variable to which market size is proportional as well as a proportionality factor
to calculate actual market size (Nevo 2000b). In the case of the RTE cereal indus-
try, we ultimately need a measure of the monthly market size, in terms of quantity
of RTE cereal consumed, for each DMA in the sample. I assume the market size
in each DMA-month is proportional to the population size. I construct monthly
estimates of the population size for each DMA between 2010 and 2014 using the
U.S. Census Bureau’s Annual Estimates of the Resident Population for Counties. I
assume the population is constant across all months in a given year. DMAs con-
sist of non-overlapping counties, so aggregating these population estimates to the
66
DMA-level is straightforward.
To estimate the proportionality factor, I use the national, trip-level Nielsen
Consumer Panel data from 2010 to 2014 to calculate the total volume (in ounces)
of RTE cereal consumed per household, per year. I then use each household’s size
to calculate the total number of individuals in the sample and construct a yearly
measure of average daily per capita consumption. On average, daily per capita cereal
consumption between 2010 and 2014 is about 0.5 ounces (about half a serving per
person per day for a typical breakfast cereal); however, it declines slightly each
year over this time period, suggesting that the RTE cereal market is shrinking. To
accurately capture the changing market size, I let the daily per capital consumption
estimate vary by year. Multiplying by the number of days in each month, I construct
the final proportionality factor to use with monthly population estimates above to
define market size: average monthly per capita RTE cereal consumption in ounces.
I use market size along with the quantity data from the Nielsen Retail Scanner data
to calculate brand market shares to use in demand estimation.
2.4.4 Consumer Demographic Data
To construct an empirical distribution of consumer demographics, I use data
from the Annual Social and Economic Supplement to the Current Population Survey
for 2010 through 2014. Using county information and the sample weights provided
in the survey, I sample 50 individuals for each DMA each year (the data is the same
across all months in a given year). The variables include age, household income, and
67
household size. I calculate individual income as household income divided by house-
hold size, and I also define a child indicator variable that equals one if age < 16.
The final demographic variables used in the estimation are logarithm of income, log-
arithm of income-squared, age, and child. For stability in the estimation procedure,
log income and age are demeaned and scaled by their standard deviations; and log
income-squared and child are demeaned. Table 2.2 contains summary statistics by
DMA for the demographic variables used in estimation.
2.5 Empirical Approach
2.5.1 Demand Estimation
By defining θ = (θ1, θ2) as a vector of all the parameters in the structural
demand model, where θ1 = (α, β) are the linear parameters and θ2 = (Π,Σ) are the
nonlinear parameters, we can combine Equations 2.1 and 2.2 to express utility as
uijt = δjt(xjt, pjt, ξjt; θ1) + µijt(xjt, pjt, vi, Di; θ2) + εijt, (2.5)
δjt = xjtβ − αpjt + ξj + ξt + ∆ξjt, µijt = [−pjt, xjt]′ ∗ (ΠDi + Σvi).
In this formulation, δjt contains only linear parameters and is devoid of any individual-
specific parameters, so it represents the mean utility common to all consumers.
The terms µijt + εijt represent a mean-zero deviation from the mean, capturing
the individual random coefficients. Consumer tastes are distributed as multivariate
normal, conditional on demographics, such that vi ∼ N(0, IK+1). The vector of
68
demographics D is sampled from the CPS and includes variables for log income, log
income-squared, age, and a child indicator. Because I include brand-specific dummy
variables (ξj), xjt only contains a time-varying indicator for non-GMO certification
that equals one when a product receives Non-GMO Project Verification and zero
otherwise.
We assume the εijt is distributed i.i.d. according to a Type I extreme value
distribution, but allow for correlation between choices through the term µijt. Under
these assumptions, I calculate individual purchase probabilities as
sijt =
exp(δjt + µijt)
1 +
∑K
k=1 exp(δkt + µikt)
(2.6)
and product market shares as
sjt =
∑It
i=1 sijt
It
. (2.7)
To address correlation between prices p and the structural error term ∆ξjt, we
introduce a set of price instruments Z = [z1, . . . , zM ] and use the estimation method
developed by Berry (1994) to construct a nonlinear GMM estimator based on the
moment condition:
E[Zm ω(θ
∗)] = 0, m = 1, . . . ,M,
where ω is the structural error term (see below) and θ∗ are the true parameter
values. The estimation routine entails minimizing the GMM objective function to
69
calculate an estimate of θ∗ such that
θˆ = argmin
θ
ω(θ)′ZA−1Z ′ω(θ) (2.8)
where A is an appropriate weight matrix—i.e., a consistent estimate of E[Z ′ωω′Z].
To express the structural error term as a function of the parameters, we must first
calculate the vector of mean utilities δ·t for each market t. To do so, we equate the
calculated market shares from Equation 2.7 with observed market shares from the
data:
s(δ·t; θ2) = S·t (2.9)
and solve for δ·t by inverting the system of market share equations numerically using
the BLP contraction mapping:
δ
(k+1)
·t = δ
(k)
·t + lnS·t − ln s(δ(k)·t ; θ2), (2.10)
where k denotes the fixed-point iteration. Once δ is computed, the error term can
be calculated as
ωjt = δjt(S·t; θ2)− xjtβ − αpjt (2.11)
and used directly in Equation 2.8. The elements of θ1 in Equation 2.11 are obtained
using linear instrumental variables regression. Nevo (2000b) and Nevo (2001) pro-
70
vide additional details on the estimation strategy, and Appendix B.2 documents
several improvements and deviations in my own computational strategy.
Once I have estimated the structural demand parameters, I can calculate the
partial derivatives of market shares with respect to prices as
∂sj(p)
∂pk
=

1
It
∑It
i αisijt(1− sijt) if j = k,
− 1
It
∑It
i αisijtsikt otherwise.
(2.12)
I calculate the price elasticities of the market shares sjt as
ηjkt =

pjt
sjt
1
It
∑It
i αisijt(1− sijt) if j = k,
−pkt
sjt
1
It
∑It
i αisijtsikt otherwise.
(2.13)
These values are used to calculate price-cost margins and to simulate welfare effects.
2.5.2 Price Instruments
While brand and time fixed effects eliminate the unobserved brand-specific
and month-specific deviations from the structural error term in Equation 2.5, the
DMA-specific component ∆ξjt remains. If firms account for this deviation, then
DMA-specific valuations will be correlated with the error term, creating a price
endogeneity problem and biasing estimates of α. To correct this problem, I use
a similar approach to Nevo (2001) to construct price instruments by exploiting
the panel structure of the data. The identifying assumption is that DMA-specific
valuations are independent across DMAs after controlling for brand, month, and
71
consumer demographics, but prices across DMAs are correlated due to common
marginal costs. Under this assumption, for a given DMA, prices of brand j in all
other DMAs and across all months are valid instruments. I implement this strategy
for each brand j in DMA-month t by constructing monthly average prices for all
directly neighboring DMAs and using prices for the twelve nearest months (including
the current month) as instruments.
2.5.3 Time-Invariant Product Characteristics
By employing brand fixed effects, taste coefficients for time-invariant prod-
uct characteristics that may be of interest cannot be recovered directly from the
main estimation. For example, organic certification may be seen as a substitute
for non-GMO certification, but due to its time-invariant nature in the data sam-
ple, its direct effect cannot be estimated. Other time-invariant characteristics of
interest include: organic certification interacted with final non-GMO status, new
product indicator,5 kids’ cereal indicator, and sugar content. To recover estimated
coefficients for these variables, I regress the brand fixed effects recovered from the
main estimation on these characteristics using the minimum-distance procedure of
Chamberlain (1982). The estimation procedure consists of a GLS regression where
the estimated covariance matrix from the main estimation is used as a weight ma-
trix to adjust for correlation in the dependent variable. These results are presented
alongside the full model results in Section 2.6.2.
5In essence, did the product launch during the time period of the sample or did it exist prior
to that?
72
2.5.4 Welfare Analysis
From a policy standpoint, measuring changes in consumer welfare is a critical
component in evaluating different labeling schemes. Using the structural parame-
ters recovered from demand estimation, I simulate the effects of two counterfactual
scenarios on consumer welfare. In the first scenario, I assume that the government
completely bans the use of GMO ingredients in food and thus requires all 50 brands
to undergo non-GMO certification and use the label across all markets in the sample.
In the second scenario, I assume the government outlaws non-GMO labeling on food
products and thus bans any brands from using the label across all markets in the
sample. These scenarios are not entirely unreasonable and warrant consideration.
For example, many countries in Europe currently place heavy restrictions on the
use of GMO ingredients in food products in what amounts to a ban. As such, most
food products undergo non-GMO certification and very few contain GMOs.6 Ad-
ditionally, in the U.S., the FDA asserts that approved GMO food products are not
significantly different from or less safe than their non-GMO produced counterparts
and, thus, do not require additional labeling. Based on that, industry groups have
spent years lobbying Congress to pass a law banning any form GMO labeling on the
grounds that it would mislead consumers.7
In a perfect information environment, one can estimate the changes to con-
6At the very least, in many European countries if a food product contains GMOs, it must be
clearly labeled as such, creating a stigma that food manufacturers try to avoid.
7This effort ultimately lead to the passage of the National Bioengineered Food Disclosure Stan-
dard in 2016, which tasks USDA with establishing a national voluntary non-GMO labeling stan-
dard.
73
sumer welfare using a simple measure of compensating variation based on the tradi-
tional random utility model. We can calculate the compensating variation for each
individual in a given market t as
CV Pi =
1
αi
[
ln
J∑
j=0
exp(V˜ij)− ln
J∑
j=0
exp(Vij)
]
, (2.14)
where Vij = δj + µij and the terms with a tilde are evaluated after the policy
change. Taking the average of this result across all It individuals yields the aver-
age compensating variation for market t. However, non-GMO food products are
credence goods, differentiated by a vertical process attribute unobservable to the
consumer, even after consumption (Adalja 2016). Accordingly, in the presence of
imperfect information, the traditional multinomial logit measure for compensating
variation is biased due to the discrepancy between consumers’ decision utility and
experience utility (Houde 2016). When the utility function for consumers’ purchase
decisions does not coincide with the utility function for consumers’ post-purchase
experiences, the change in consumer surplus for each individual in a given market t
can be expressed as
CV Ii =
1
αi
[
ln
J∑
j=0
exp(V˜ij) +
J∑
j=0
s˜ij(V˜
E
ij − V˜ij)
]
−
1
αi
[
ln
J∑
j=0
exp(Vij) +
J∑
j=0
sij(V
E
ij − Vij)
]
, (2.15)
where the terms with a tilde are evaluated after the policy change, V Eij denotes
experience utility, and Vij denotes decision utility. The expression in Equation 2.15
74
differs from the perfect information welfare measure in Equation 2.14 due to the
two correction terms of the form
∑J
j=0 sij(V
E
ij − Vij). These terms account for the
discrepancy between consumers’ perceptions that guide decision making and what
they actually experience (see Leggett 2002, for a full derivation). Note that if no
discrepancy exists between the two utility functions, then the expression in 2.15
reduces to the perfect information welfare measure in 2.14.
Given the credence good nature of the non-GMO product attribute, one might
argue that a non-GMO label affects decision utility, but it does not impact expe-
rience utility due to the fact that the attribute cannot be physically experienced,
even after consumption. Such an argument suggests the use of Equation 2.15 for
calculating welfare effects of a policy change. The possibility also exists that con-
sumers who purchase non-GMO products experience a warm glow (Andreoni 1990),
or the certification may affect social status in some way, such that the label also
impacts experience utility, despite the fact that it cannot be physically sensed. If
the non-GMO label’s impact on decision and experience utility are aligned, then
Equation 2.14 is an appropriate measure for welfare analysis. Given these compet-
ing arguments, I present welfare estimates using both expressions for changes in
consumer surplus in the results.
75
2.6 Results
2.6.1 Logit Specification
In Equation 2.5, if we assume that consumer heterogeneity only enters the
model through the error term εijt (such that θ2 = 0, βi = β, and αi = α for all
consumers), and we assume that εijt is distributed as i.i.d. Type I extreme value;
then the model distills to a logit specification. In this case, Equation 2.9 can be
solved analytically as δjt = ln(Sjt) − ln(S0t), where S0t is the observed outside
market share for market t, and the estimation procedure simplifies to 2SLS. The
logit specification places strong restrictions on the model—it implies that cross-
price elasticities are only a function of market share; however, it can serve as a
useful starting point for the full model.
Table 2.3 presents results for the logit specification of the model. The the
first column is estimated using standard OLS without instrumenting for price, while
the second column is estimated using 2SLS with the price instruments described
in Section 2.5.2. As expected, the parameter estimate for price is negative in both
cases; but the 2SLS estimate is larger in magnitude. This suggests, at the very
least, that failing to address the price endogeneity issue results in an attenuation
bias when measuring own-price elasticities. Interestingly, the coefficient on the non-
GMO labeling indicator is negative and statistically significantly different from zero
in both cases, indicating that use of the label reduces the mean utility of consumers.
This result is consistent across both regressions and does not change significantly
76
when we instrument for price. It is worth noting, however, that the point estimate
for the label indicator is about one-twentieth the magnitude of price; so while the
effect is negative, it may not be economically meaningful. In the full model, we will
explore this possibility in more detail by simulating the economic effects of different
labeling scenarios.
2.6.2 Full Model
The results for the full random coefficients logit model are presented in Ta-
ble 2.4. The specification includes brand and time (month) fixed effects in addi-
tion to the demand parameters listed in the table. The first column presents the
mean parameter estimates (β) as well as the taste coefficients estimated using the
minimum-distance procedure. The estimate for price is negative, as expected, and
about twice the magnitude of the estimate from the logit specification. This indi-
cates higher own-price elasticity, on average. The estimate for the non-GMO labeling
indicator is positive and about one-tenth the magnitude of the price estimate. This
suggests that use of the label has a slightly positive effect on mean utility and, thus,
a brand’s market share, which means that firms may have an incentive to seek out
the label. Anecdotally, this result seems to coincides with the label’s rapid growth
between 2010 and 2014. Furthermore, while Adalja (2016) finds no effect of the
non-GMO label on price premiums for newly certified non-GMO food products, this
result suggests that the effect of the non-GMO label is transmitted via changes in
market share.
77
The additional taste coefficients estimated from the brand fixed effects provide
further insight on the drivers of demand. First, the coefficient on organic certifi-
cation is also positive and of similar magnitude to the non-GMO label coefficient,
indicating that both certifications have a similar impact on demand. However, the
Non-GMO×Organic interaction term, while similar in magnitude to the organic and
non-GMO estimates, is of the opposite sign (negative). This finding would suggest
that the two certifications are effectively substitutes in terms of driving demand,
and the presence of both certifications on a product does not have an appreciably
greater impact on demand.8 The coefficient estimate for the kids’ cereal indicator
is positive, as is that for sugar content, which is consistent with past findings in the
literature. Interestingly, the estimate for the new product indicator is negative and
of similar magnitude to the kids’ indicator. To some extent, the extensive product
promotion and advertising that tends to accompany the launch of a new RTE cereal
product may serve as an effort to overcome this negative effect.
The subsequent columns provide model estimates that characterize individual
heterogeneity around the means. The demographic interactions used in the final
specification include price interaction terms for income, income squared, and child;
label interaction terms for income and age; and constant terms for income and age.
Additionally, I estimate standard deviation (σ) for each of the parameters. The
signs for the price interaction coefficient estimates indicate that individuals with
higher incomes are generally more sensitive to price, which is counterintuitive based
8Given that the National Organic Standard prohibits the use of GMO ingredients, it is not
terribly surprising that these certifications act as substitutes to some extent; however, the organic
certification involves much more than simply using non-GMO ingredients.
78
on economic theory; however, the impact of an increase of one standard deviation
in log income from the average is roughly half that of the price coefficient, so the
effect may not be economically meaningful. Individuals under 16 years old are much
more sensitive to price than adults, as expected. Lastly, the parameter estimate for
the standard deviation of price captures unobserved heterogeneity not explained by
demographics.9
The signs for the non-GMO label interaction coefficient estimates are straight-
forward: wealthier individuals as well as older individuals value the non-GMO label
less, all else equal. Furthermore, the magnitude of the income interaction estimate
is on the same order as the mean parameter estimate for the label, and the estimate
for the age interaction is an order of magnitude larger. Therefore, a one-standard
deviation increase in log income or age from the sample averages effectively cancels
out the positive mean valuation of the label, and beyond that the valuation for
non-GMO may become negative. We observe this in the frequency distribution of
the individual-specific non-GMO label coefficients in Figure 2.2. Consumer tastes
for the non-GMO label coefficient have a wide distribution; and, while the mean
valuation is slightly positive, the distribution is rather evenly distributed around
zero, suggesting that consumer willingness to pay for non-GMO certified RTE ce-
real varies significantly.
9I also calculate own- and cross-price elasticities from the model using Equation 2.13. The
results are intuitive and generally as expected. Since price response is not the primary focus of this
paper, median own- and cross-price elasticities are presented in Appendix Tables B.1, B.2, B.3, B.4,
and B.5.
79
2.6.3 Simulated Welfare Effects
To shed light on the potential welfare effects of non-GMO labeling, I simulate
two counterfactual labeling scenarios in the RTE cereal industry as outlined in Sec-
tion 2.5.4: one in which all brands use the non-GMO label over the entire timeframe
of the data, and one in which no brands use the label. To establish an initial baseline
for comparing the simulations, I first use the demand parameters recovered from the
model to construct the price derivative matrix S using Equation 2.12. Then I con-
struct the ownership matrix Ωo for the 50 brands used in the estimation to calculate
price-cost margins using Equation 2.4. Initial mean values for market share, price,
marginal cost, price-cost margin, revenue, and profit for each brand, across all 5,988
markets, are presented in Table 2.5.
To simulate each scenario, I simply update the vector xjt to reflect the new
labeling scenario, and then calculate new values for the individual component of
utility µijt using Equation 2.5 and new product market shares using Equation 2.7.
With updated market shares, new revenue and profit can be calculated for each
brand to assess how each labeling scenario impacts firm profitability. Table 2.6
provides updated mean market share, revenue, profit, and change in profit for each
brand, across all 5,988 markets.
The results indicate that complete labeling (Scenario 1) makes most brands
worse off than in the initial case. A partial rationale for this result is as follows.
Full labeling erodes the market power of firms that previously used the non-GMO
80
label in the initial case such that these brands lose their niche of high-valuation
non-GMO consumers to other, less expensive brands that start using the label. At
the same time, since the non-GMO label has only a slightly positive effect on mean
utility, and many consumers have a net negative valuation for the non-GMO label,
complete labeling causes more individuals to choose the outside option. As a result,
the original non-GMO RTE cereal brands as well as the newly-labeled conventional
brands tend to lose market share and become worse off on average. On the other
hand, no labeling (Scenario 2) tends to have an ambiguous effect, with firms both
better and worse off. In this scenario, firms that previously used the non-GMO label
may be able to maintain some portion of their high-valuation consumer base while
also capturing market share of lower-valuation consumers from no-label brands. To
the extent that this is possible, some non-GMO RTE cereal brands may become
better off at the expense of conventional brands.
To estimate changes to consumer surplus, I use the new values of utility, µijt,
to calculate the compensating variation [CV] for each scenario, as defined in Equa-
tions 2.14 and 2.15 based on our stance regarding decision vs. experience utility, for
each individual i in market t. I then take the mean of CV over all 50 individuals
for each market t to calculate a market-level mean CV attributable to each labeling
scenario. In Table 2.7, I present both the volume-weighted and population-weighted
average of mean CV over all 5,988 markets for both CV calculations. If we assume
no discrepancy between decision and experience utility, then complete non-GMO
labeling in the RTE cereal industry (Scenario 1) reduces consumer welfare across
all markets on average. However, when we account for the credence good aspect
81
of non-GMO labeling and incorporate the Leggett (2002) correction, complete non-
GMO labeling improves consumer welfare on average. This results indicates that
the “cost” of misperception after the policy change is less than the “cost” before the
policy change. On the other hand, the results for Scenario 2 indicate that consumers
would be worse off with no non-GMO labeling relative to the current baseline, re-
gardless of which calculation is used. In fact, the point estimates are very similar
with and without the Leggett correction, since consumers’ decision and experience
utility effectively converge after the policy change wherein no non-GMO labeled
products exist. Given the positive demand parameter estimate for the non-GMO
label on the mean utility valuation, this result is rather intuitive.
2.7 Conclusion
In this paper, I investigate how voluntary non-GMO food labeling impacts
demand in the RTE cereal industry by estimating a discrete-choice, random coeffi-
cients logit demand model with Nielsen Retail Scanner data for 50 breakfast cereal
brands in 100 DMAs between 2010 and 2014. The results indicate that consumer
tastes for the non-GMO label have a wide distribution, and this heterogeneity plays
a substantial role in individual choices; but, on average, the non-GMO label has a
positive impact on demand. Organic certification has a similar impact on demand
to that of the non-GMO label; however, in combination, the two certifications are
effectively substitutes, and the presence of both certifications on a product does
not have an appreciably greater impact on demand. To shed light on the potential
82
welfare effects of non-GMO labeling, I simulate two labeling scenarios in the RTE
cereal industry: one in which all brands use the non-GMO label over the entire
timeframe of the data, and one in which no brands use the label. The simulation
results indicate that non-GMO labeling in the RTE cereal industry may improve
consumer welfare, but reduce industry profit on average.
The RTE cereal industry has long been a subject of research in empirical indus-
trial organization, with several previous studies investigating the role that non-price
strategies such as advertising, couponing, and new product introductions play in the
industry (Thomas 1999; Nevo 2001; Nevo and Wolfram 2002). This paper builds on
that work and is the first to examine how another non-price marketing strategy—
voluntary quality certification—impacts demand in the RTE cereal industry.
Going forward, there are several other considerations and extensions that may
benefit this work. First, given the emergent nature of the Non-GMO Project Ver-
ified label,10 consumer learning dynamics may factor significantly into the results.
For example, prior to the existence of the non-GMO label, it is quite plausible that
most consumers were uninformed about GMOs. As the label began to appear on
supermarket shelves, those consumers may have gradually become educated about
the label and changed their preferences accordingly. As such, results from a static
demand model that does not account for this may have questionable external va-
lidity. To accurately capture the effect of consumer learning, it may be necessary
to adopt a dynamic framework that incorporates a simple Bayesian learning model
(e.g., Ackerberg 2001, 2003). Additionally, there are several complicating dynamics
10While the organization was founded in 2005, they only began labeling products in 2010.
83
worthy of consideration on the supply side as well, such as new product development.
As consumers become educated about GMOs, firms may develop new products to
meet changing consumer tastes, rather than simply labeling their existing prod-
ucts. While these dynamics are not addressed in the current model, they remain
important extensions for future work.
84
Figure 2.1: Annual Non-GMO Project Verified RTE Cereal Sales
85
Figure 2.2: Distribution of Non-GMO Label Coefficient
86
Table 2.1: Summary Statistics
Brand
Non-
GMO
Price Label
Avg.
Market Share No.
MarketsAvg. StDev Avg. StDev
01 BARBARA’S PUFFINS Yes 0.386 0.064 0.296 0.000 0.000 5306
02 G M CHEERIOS No 0.211 0.019 – 0.013 0.006 5988
03 G M CINNAMON TOAST CRUNCH No 0.195 0.018 – 0.012 0.005 5988
04 G M COCOA PUFFS No 0.209 0.021 – 0.003 0.002 5988
05 G M FIBER ONE No 0.222 0.026 – 0.001 0.001 5988
06 G M HONEY NUT CHEERIOS No 0.198 0.019 – 0.020 0.008 5988
07 G M HONEY NUT CHEX No 0.215 0.029 – 0.002 0.001 5988
08 G M LUCKY CHARMS No 0.209 0.021 – 0.010 0.004 5988
09 G M MULTIGRAIN CHEERIOS No 0.267 0.026 – 0.005 0.002 5988
10 G M REESE’S PUFFS No 0.196 0.019 – 0.004 0.002 5988
11 G M RICE CHEX No 0.238 0.031 – 0.003 0.002 5988
12 KASHI AUTUMN WHEAT PROJECT SPK Yes 0.199 0.022 1.000 0.001 0.000 3131
13 KASHI CINNAMON HARVEST Yes 0.203 0.017 0.749 0.001 0.001 4788
14 KASHI GO LEAN Yes 0.232 0.020 0.017 0.001 0.001 5988
15 KASHI GO LEAN CRISP! Yes 0.220 0.020 0.182 0.002 0.001 5988
16 KASHI GO LEAN CRUNCH! Yes 0.217 0.018 0.182 0.003 0.002 5988
17 KASHI HEART TO HEART No 0.254 0.021 – 0.002 0.001 5988
18 KASHI ISLAND VANILLA Yes 0.206 0.027 0.595 0.000 0.000 5795
19 KASHI ORGANIC PROMISE ATMN WHT Yes 0.199 0.020 0.376 0.001 0.000 3846
20 KASHI ORGANIC PROMISE CN HRVST Yes 0.208 0.022 0.021 0.001 0.001 1226
21 KASHI ORGANIC PROMISE STBY FLD Yes 0.334 0.050 0.588 0.000 0.000 5759
22 KEL APPLE JACKS No 0.214 0.026 – 0.005 0.002 5988
23 KEL CORN FLAKES No 0.187 0.019 – 0.005 0.002 5988
24 KEL FROOT LOOPS No 0.213 0.023 – 0.008 0.003 5988
Continued. . .
87
Table 2.1 – continued from previous page
Brand
Non-
GMO
Price Label
Avg.
Market Share No.
MarketsAvg. StDev Avg. StDev
25 KEL FROSTED FLAKES No 0.172 0.021 – 0.017 0.008 5988
26 KEL FROSTED MINI-WHEATS No 0.161 0.014 – 0.016 0.008 5988
27 KEL FROSTED MINI-WHT LTTLE BTS No 0.193 0.022 – 0.002 0.002 5985
28 KEL RAISIN BRAN No 0.140 0.015 – 0.009 0.004 5988
29 KEL RAISIN BRAN CRUNCH No 0.167 0.016 – 0.005 0.002 5988
30 KEL RICE KRISPIES No 0.222 0.022 – 0.006 0.003 5988
31 KEL SPECIAL K No 0.234 0.018 – 0.004 0.002 5988
32 KEL SPECIAL K RED BERRY No 0.242 0.020 – 0.005 0.003 5988
33 KEL SPECIAL K VANILLA ALMOND No 0.222 0.019 – 0.002 0.001 5988
34 M-O-M FROSTED MINI SPOONERS No 0.126 0.023 – 0.003 0.003 5972
35 MOM’S BEST NATURALS TSTD WT-FS Yes 0.122 0.025 – 0.000 0.000 3444
36 NATURE’S PATH FLAX PLUS Yes 0.271 0.054 0.900 0.000 0.000 5927
37 NATURE’S PATH HERITAGE MLTGN Yes 0.268 0.058 0.901 0.000 0.000 5661
38 NATURE’S PATH OP PW BRKFST BLB Yes 0.258 0.050 0.898 0.000 0.000 5829
39 POST GRAPE-NUTS Yes 0.139 0.017 0.165 0.003 0.002 5988
40 POST HONEY BUNCHES OF OATS No 0.180 0.016 – 0.017 0.009 5988
41 POST RAISIN BRAN No 0.128 0.015 – 0.003 0.002 5988
42 POST SELECTS GREAT GRAINS No 0.201 0.021 – 0.001 0.001 5980
43 POST SHRD WHT ’N BRN SP SZ Yes 0.188 0.022 0.096 0.001 0.001 5768
44 POST SHREDDED WHEAT SPOON SIZE Yes 0.179 0.025 0.116 0.001 0.001 5988
45 QKR CINNAMON LIFE No 0.176 0.021 – 0.004 0.003 5988
46 QKR LIFE No 0.174 0.021 – 0.005 0.003 5988
47 QKR OATMEAL SQUARES No 0.199 0.030 – 0.003 0.002 5984
48 UNCLE SAM TSTD WL-WT FLK&FLXSD Yes 0.310 0.050 0.779 0.000 0.000 5266
Continued. . .
88
Table 2.1 – continued from previous page
Brand
Non-
GMO
Price Label
Avg.
Market Share No.
MarketsAvg. StDev Avg. StDev
49 UNCLE SAM TSTD WWB FLKS&FLSD Yes 0.231 0.060 0.982 0.000 0.000 3336
50 KASHI ORGANIC PROMISE BF PS Yes 0.207 0.018 1.000 0.000 0.000 2454
89
Table 2.2: Descriptive Statistics for Demographic Variables by DMA
Designated Marketing Area (DMA) Outside
Market
Share
Average
Popula-
tion
Average
Income
Average
Age
Average
Child
PORTLAND-AUBURN ME 0.636 989166 24913 39.4 0.18
NEW YORK NY 0.796 21165376 27046 38.0 0.18
MACON GA 0.839 671423 21953 35.5 0.22
PHILADELPHIA PA 0.731 8046354 26739 37.9 0.21
DETROIT MI 0.796 4837253 25722 37.2 0.18
BOSTON (MANCHESTER) MA-NH 0.653 6430730 23929 39.2 0.18
SAVANNAH GA 0.824 915541 24933 35.6 0.24
PITTSBURGH PA 0.726 2840470 28157 41.1 0.18
FT WAYNE IN 0.819 719204 25937 37.9 0.19
CLEVELAND OH 0.789 3836869 24641 39.8 0.20
WASHINGTON DC (HAGERSTOWN MD) 0.685 6627815 35519 35.7 0.24
BALTIMORE MD 0.716 2945870 32093 37.1 0.17
FLINT-SAGINAW-BAY CITY MI 0.843 1164550 25992 37.9 0.19
BUFFALO NY 0.920 1601355 23100 38.4 0.17
CINCINNATI OH 0.688 2337744 25149 38.5 0.23
CHARLOTTE NC 0.732 3036643 23051 35.0 0.24
GREENSBORO-HIGH POINT-WINSTON SALEM NC 0.775 1760343 25763 40.1 0.18
CHARLESTON SC 0.827 829479 20106 34.4 0.23
AUGUSTA GA 0.809 702247 22787 32.4 0.29
PROVIDENCE-NEW BEDFORD RI-MA 0.804 1604318 25035 37.8 0.22
BURLINGTON-PLATTSBURGH VT-NY 0.759 850630 23682 38.5 0.23
ATLANTA GA 0.831 6517469 25114 35.3 0.22
INDIANAPOLIS IN 0.840 2934588 27729 36.5 0.23
Continued. . .
90
Table 2.2 – continued from previous page
Designated Marketing Area (DMA) Outside
Market
Share
Average
Popula-
tion
Average
Income
Average
Age
Average
Child
MIAMI-FT LAUDERDALE FL 0.859 4491338 23845 39.4 0.18
LOUISVILLE KY 0.797 1726671 26683 38.9 0.18
TRI-CITIES TN-VA 0.764 800730 24245 34.9 0.25
ALBANY-SCHENECTADY-TROY NY 0.825 1391414 23921 38.4 0.19
HARTFORD & NEW HAVEN CT 0.843 2657837 26529 38.4 0.19
ORLANDO-DAYTONA BEACH-MELBOURNE FL 0.890 3812030 24164 38.8 0.15
COLUMBUS OH 0.723 2441594 26505 35.4 0.22
YOUNGSTOWN OH 0.807 665768 21974 40.1 0.22
TAMPA-ST PETERSBURG (SARASOTA) FL 0.900 4456924 24682 42.3 0.14
LEXINGTON KY 0.819 1269440 27182 34.1 0.21
DAYTON OH 0.770 1208131 25138 36.6 0.20
NORFOLK-PORTSMOUTH-NEWPORT NEWS VA 0.710 1914660 27071 35.3 0.19
GREENVILLE-NEW BERN-WASHINGTON NC 0.806 805957 19905 28.6 0.30
COLUMBIA SC 0.842 1074481 24916 39.5 0.20
TOLEDO OH 0.813 1068193 22877 37.1 0.20
WEST PALM BEACH-FT PIERCE FL 0.911 1973037 24279 42.2 0.19
WILMINGTON NC 0.755 469138 25637 36.5 0.18
RICHMOND-PETERSBURG VA 0.763 1476812 28114 38.0 0.19
KNOXVILLE TN 0.735 1346498 24070 41.4 0.16
RALEIGH-DURHAM (FAYETTEVILLE) NC 0.726 3011123 27672 33.0 0.24
JACKSONVILLE FL 0.845 1783125 27431 40.1 0.22
CHARLESTON-HUNTINGTON WV 0.841 1160960 24754 34.8 0.26
HARRISBURG-LANCASTER-LEBANON-YORK PA 0.754 1979810 25843 37.1 0.24
Continued. . .
91
Table 2.2 – continued from previous page
Designated Marketing Area (DMA) Outside
Market
Share
Average
Popula-
tion
Average
Income
Average
Age
Average
Child
GREENVILLE-SPARTANBURG SC-ASHEVILLE NC 0.793 2206893 22206 36.3 0.22
FLORENCE-MYRTLE BEACH SC 0.805 752680 19352 36.5 0.23
FT MYERS-NAPLES FL 0.898 1231514 24841 42.6 0.17
ROANOKE-LYNCHBURG VA 0.745 1143505 27664 36.0 0.18
JOHNSTOWN-ALTOONA PA 0.822 759006 25051 41.8 0.16
CHATTANOOGA TN 0.839 948008 24711 36.8 0.21
SALISBURY MD 0.655 414483 23088 43.4 0.19
WILKES BARRE-SCRANTON PA 0.916 1527611 25122 39.3 0.22
CHICAGO IL 0.734 9685854 25338 35.6 0.26
ST LOUIS MO 0.921 3192777 25628 38.1 0.20
ROCHESTER-MASON CITY-AUSTIN MN-IA 0.719 368385 28720 34.6 0.25
SHREVEPORT LA 0.844 1019171 24818 36.0 0.22
MINNEAPOLIS-ST PAUL MN 0.748 4596920 30219 38.3 0.23
KANSAS CITY MO-KS 0.893 2451398 29849 38.1 0.18
MILWAUKEE WI 0.689 2318553 24953 37.9 0.22
HOUSTON TX 0.839 6555421 26218 32.7 0.26
NEW ORLEANS LA 0.898 1707119 22472 35.6 0.24
DALLAS-FT WORTH TX 0.824 7298112 24431 34.4 0.25
AUSTIN TX 0.944 1979407 22766 34.3 0.23
CEDAR RAPIDS-WATERLOO & DUBUQUE IA 0.802 886337 29867 38.0 0.18
MEMPHIS TN 0.808 1811992 26166 37.8 0.17
OMAHA NE 0.751 1100039 29119 35.5 0.24
GREEN BAY-APPLETON WI 0.799 1129587 27551 36.1 0.25
Continued. . .
92
Table 2.2 – continued from previous page
Designated Marketing Area (DMA) Outside
Market
Share
Average
Popula-
tion
Average
Income
Average
Age
Average
Child
NASHVILLE TN 0.830 2702008 26633 37.3 0.21
MADISON WI 0.777 971305 23616 33.2 0.26
PEORIA-BLOOMINGTON IL 0.837 647925 29075 39.3 0.22
WICHITA-HUTCHINSON PLUS KS 0.794 1211270 21324 33.9 0.28
DES MOINES-AMES IA 0.800 1121856 25505 34.3 0.25
DAVENPORT-ROCK ISLAND-MOLINE IA-IL 0.819 771658 27057 35.8 0.26
MOBILE-PENSACOLA (FT WALTON BEACH) AL-FL 0.910 1409843 24135 39.1 0.22
LITTLE ROCK-PINE BLUFF AR 0.856 1467596 24790 35.8 0.23
TYLER-LONGVIEW (LUFKIN & NACOGDOCHES) TX 0.842 743180 19016 34.3 0.30
SIOUX FALLS (MITCHELL) SD 0.811 684839 30218 34.6 0.24
DENVER CO 0.683 4167898 28775 35.3 0.21
COLORADO SPRINGS-PUEBLO CO 0.787 937746 23605 40.4 0.25
PHOENIX AZ 0.713 5116720 27898 40.0 0.19
BOISE ID 0.854 744072 23509 37.3 0.24
SALT LAKE CITY UT 0.835 3036126 20127 30.8 0.33
TUCSON (SIERRA VISTA) AZ 0.695 1171202 23292 38.1 0.21
ALBUQUERQUE-SANTA FE NM 0.876 1933926 28213 36.9 0.26
BAKERSFIELD CA 0.764 857730 17657 31.5 0.26
EUGENE OR 0.787 610633 24399 38.6 0.16
LOS ANGELES CA 0.764 18261036 25024 36.6 0.20
SAN FRANCISCO-OAKLAND-SAN JOSE CA 0.728 7090684 29896 37.9 0.16
YAKIMA-PASCO-RICHLAND-KENNEWICK WA 0.823 690790 18166 35.6 0.27
SEATTLE-TACOMA WA 0.709 4931355 31002 35.8 0.23
Continued. . .
93
Table 2.2 – continued from previous page
Designated Marketing Area (DMA) Outside
Market
Share
Average
Popula-
tion
Average
Income
Average
Age
Average
Child
PORTLAND OR 0.729 3202730 21102 36.9 0.24
SAN DIEGO CA 0.722 3183143 24645 36.6 0.23
MONTEREY-SALINAS CA 0.741 749018 21807 33.9 0.24
LAS VEGAS NV 0.742 2051833 22500 37.6 0.22
SANTA BARBARA-SANTA MARIA CA 0.714 705739 22559 35.1 0.21
SACRAMENTO-STOCKTON-MODESTO CA 0.820 4289848 22477 34.7 0.27
FRESNO-VISALIA CA 0.819 1983349 20782 33.9 0.26
SPOKANE WA 0.849 1126495 27557 39.3 0.19
94
Table 2.3: Results from the Logit Specification
Variable OLS 2SLS
Price −9.218∗∗∗ −9.815∗∗∗
(0.056) (0.075)
NGMO Label −0.442∗∗∗ −0.448∗∗∗
(0.008) (0.008)
Instruments – prices
R2 0.983 0.983
Num. obs. 277, 085 277, 085
Note: Each column represents a separate regression. All regressions in-
clude brand and month fixed effects. Standard errors are in parentheses:
∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1.
95
Table 2.4: Full Model Results
Variable
Means StDev Intxn with Demographic Vars
β σ Income IncomeSq Age Child
Price -17.679 1.531 -8.107 -9.913 -64.246
NGMO Label 1.386 2.904 -1.987 -11.498
Constant -14.902a 2.609 8.995 7.075
(1.228)
Organic 1.293a
(0.043)
Non-GMO×Organic -1.530a
(0.051)
Kids 0.486a
(0.018)
Sugar 0.018a
(0.001)
New Product -0.546a
(0.043)
GMM Obj. 0.147
No. Obs. 277,085
a Estimated using a minimum-distance procedure.
Note: Unless otherwise specified, all parameters are GMM estimates. All regressions
include brand and month fixed effects. Standard errors are in parentheses.
96
Table 2.5: Initial Baseline for Simulation
Brand Market
Share
Price Marginal
Cost
Margin % Margin Revenue Profit
1 0.0003 0.3571 0.6076 -0.2505 -0.8400 0.0001 0.0000
2 0.0161 0.2104 0.0096 0.2009 0.9238 0.0034 0.0033
3 0.0146 0.1910 0.1082 0.0828 0.4350 0.0027 0.0012
4 0.0045 0.2024 0.0787 0.1237 0.5998 0.0009 0.0005
5 0.0020 0.2224 -0.1702 0.3926 1.6803 0.0004 0.0007
6 0.0244 0.1938 0.1087 0.0850 0.4362 0.0047 0.0020
7 0.0026 0.2068 0.1289 0.0779 0.3847 0.0005 0.0002
8 0.0114 0.2047 -0.0451 0.2498 1.1767 0.0023 0.0027
9 0.0057 0.2632 0.1580 0.1052 0.4113 0.0015 0.0007
10 0.0053 0.1911 0.1038 0.0873 0.4527 0.0010 0.0005
11 0.0042 0.2251 0.1393 0.0858 0.3609 0.0009 0.0003
12 0.0008 0.2006 0.1432 0.0575 0.2750 0.0002 0.0001
13 0.0013 0.2009 0.1582 0.0427 0.2141 0.0003 0.0001
14 0.0017 0.2315 0.3026 -0.0711 -0.2681 0.0004 -0.0001
15 0.0022 0.2164 0.1159 0.1005 0.4549 0.0005 0.0002
16 0.0040 0.2151 0.1062 0.1088 0.5038 0.0009 0.0005
17 0.0025 0.2532 0.2519 0.0013 0.0274 0.0006 0.0001
18 0.0005 0.2014 0.1091 0.0923 0.4464 0.0001 0.0000
19 0.0010 0.1991 0.2892 -0.0901 -0.5139 0.0002 0.0000
20 0.0014 0.2024 0.1376 0.0649 0.3269 0.0003 0.0001
21 0.0003 0.3326 0.5123 -0.1797 -0.5431 0.0001 -0.0001
22 0.0059 0.2068 0.0479 0.1589 0.6959 0.0012 0.0008
23 0.0062 0.1836 0.1090 0.0747 0.4021 0.0011 0.0004
24 0.0091 0.2064 0.1070 0.0994 0.4666 0.0018 0.0009
25 0.0207 0.1656 0.1040 0.0617 0.3755 0.0034 0.0012
26 0.0188 0.1587 0.1014 0.0572 0.3529 0.0029 0.0010
27 0.0032 0.1885 0.1087 0.0798 0.4191 0.0006 0.0003
28 0.0116 0.1359 0.0809 0.0550 0.4055 0.0015 0.0006
29 0.0055 0.1646 0.1044 0.0601 0.3565 0.0009 0.0003
30 0.0070 0.2182 0.1061 0.1121 0.5106 0.0015 0.0008
31 0.0056 0.2336 -0.3699 0.6035 2.5115 0.0013 0.0032
32 0.0068 0.2384 0.3597 -0.1214 -0.4742 0.0016 -0.0007
33 0.0028 0.2210 1.0154 -0.7944 -3.1855 0.0006 -0.0024
34 0.0066 0.1145 0.0834 0.0311 0.2978 0.0006 0.0002
35 0.0003 0.1139 0.0853 0.0285 0.2577 0.0000 0.0000
36 0.0003 0.2548 0.4187 -0.1639 -0.5812 0.0001 -0.0000
37 0.0003 0.2497 0.2932 -0.0435 -0.2504 0.0001 0.0000
38 0.0002 0.2310 0.2062 0.0247 0.1461 0.0000 0.0000
39 0.0041 0.1382 0.0973 0.0409 0.2967 0.0005 0.0002
Continued. . .
97
Table 2.5 – continued from previous page
Brand Market
Share
Price Marginal
Cost
Margin % Margin Revenue Profit
40 0.0238 0.1763 0.1288 0.0476 0.2725 0.0042 0.0012
41 0.0038 0.1233 0.0870 0.0364 0.3010 0.0004 0.0001
42 0.0022 0.1954 0.1428 0.0526 0.2776 0.0004 0.0001
43 0.0014 0.1872 0.1267 0.0605 0.3204 0.0003 0.0001
44 0.0022 0.1769 0.1215 0.0554 0.3112 0.0004 0.0001
45 0.0060 0.1688 0.1237 0.0451 0.2673 0.0010 0.0003
46 0.0077 0.1693 0.1260 0.0433 0.2585 0.0013 0.0003
47 0.0051 0.1947 0.1339 0.0608 0.3095 0.0010 0.0003
48 0.0002 0.2944 0.3284 -0.0340 -0.1055 0.0000 -0.0000
49 0.0002 0.2162 0.0564 0.1598 0.7458 0.0001 0.0000
50 0.0006 0.2069 0.1144 0.0925 0.4729 0.0001 0.0001
Note: The values in each column represent the intial volume-weighted mean
values of a given variable for each brand, across all 5,988 markets.
98
Table 2.6: Simulated Labeling Scenario Results
Brand
1: Complete Labeling 2: No Labeling
Market
Share
Revenue Profit Profit
Chg
Market
Share
Revenue Profit Profit
Chg
1 0.0009 0.0004 -0.0001 -0.0001 0.0006 0.0002 -0.0001 -0.0001
2 0.0064 0.0013 0.0012 -0.0021 0.0153 0.0032 0.0032 -0.0000
3 0.0057 0.0011 0.0005 -0.0007 0.0140 0.0026 0.0012 -0.0000
4 0.0017 0.0003 0.0002 -0.0003 0.0043 0.0009 0.0005 -0.0000
5 0.0008 0.0002 0.0002 -0.0005 0.0019 0.0004 0.0007 -0.0000
6 0.0094 0.0018 0.0008 -0.0012 0.0233 0.0045 0.0019 -0.0001
7 0.0009 0.0002 0.0001 -0.0001 0.0024 0.0005 0.0002 -0.0000
8 0.0044 0.0009 0.0010 -0.0017 0.0108 0.0022 0.0026 -0.0000
9 0.0032 0.0008 0.0008 0.0001 0.0054 0.0014 0.0007 -0.0000
10 0.0019 0.0004 0.0002 -0.0003 0.0050 0.0009 0.0004 -0.0000
11 0.0016 0.0003 0.0001 -0.0001 0.0039 0.0008 0.0003 -0.0000
12 0.0003 0.0001 0.0000 -0.0000 0.0040 0.0008 -0.0010 -0.0011
13 0.0005 0.0001 0.0000 -0.0001 0.0052 0.0010 -0.0012 -0.0012
14 0.0008 0.0002 -0.0000 0.0001 0.0016 0.0004 -0.0001 -0.0000
15 0.0008 0.0002 0.0001 -0.0001 0.0025 0.0005 0.0004 0.0002
16 0.0016 0.0004 0.0002 -0.0002 0.0043 0.0009 0.0005 0.0000
17 0.0013 0.0003 0.0001 -0.0000 0.0024 0.0006 0.0001 0.0000
18 0.0002 0.0000 0.0000 -0.0000 0.0016 0.0003 0.0004 0.0004
19 0.0004 0.0001 0.0000 0.0000 0.0016 0.0003 0.0000 0.0000
20 0.0007 0.0001 0.0001 -0.0000 0.0014 0.0003 0.0001 0.0000
21 0.0005 0.0002 -0.0003 -0.0003 0.0006 0.0002 -0.0002 -0.0001
22 0.0023 0.0005 0.0003 -0.0005 0.0056 0.0011 0.0007 -0.0000
23 0.0023 0.0004 0.0002 -0.0002 0.0059 0.0011 0.0004 -0.0000
Continued. . .
99
Table 2.6 – continued from previous page
Brand
1: Complete Labeling 2: No Labeling
Market
Share
Revenue Profit Profit
Chg
Market
Share
Revenue Profit Profit
Chg
24 0.0035 0.0007 0.0003 -0.0005 0.0086 0.0017 0.0008 -0.0000
25 0.0079 0.0013 0.0006 -0.0007 0.0198 0.0032 0.0012 -0.0001
26 0.0073 0.0011 0.0005 -0.0005 0.0180 0.0028 0.0010 -0.0000
27 0.0011 0.0002 0.0001 -0.0001 0.0030 0.0006 0.0002 -0.0000
28 0.0051 0.0007 0.0003 -0.0003 0.0111 0.0015 0.0006 -0.0000
29 0.0021 0.0003 0.0001 -0.0002 0.0053 0.0008 0.0003 -0.0000
30 0.0027 0.0006 0.0004 -0.0005 0.0066 0.0014 0.0008 0.0000
31 0.0023 0.0005 0.0003 -0.0029 0.0054 0.0012 0.0029 -0.0003
32 0.0030 0.0007 -0.0003 0.0004 0.0065 0.0015 -0.0007 0.0000
33 0.0011 0.0002 -0.0017 0.0006 0.0027 0.0006 -0.0024 -0.0000
34 0.0039 0.0003 0.0001 -0.0001 0.0064 0.0006 0.0002 -0.0000
35 0.0002 0.0000 0.0000 -0.0000 0.0003 0.0000 0.0000 -0.0000
36 0.0001 0.0000 -0.0000 0.0000 0.0009 0.0002 -0.0001 -0.0000
37 0.0001 0.0000 0.0000 -0.0000 0.0009 0.0002 -0.0000 -0.0001
38 0.0001 0.0000 0.0000 -0.0000 0.0006 0.0001 0.0000 0.0000
39 0.0019 0.0002 0.0001 -0.0001 0.0050 0.0007 0.0002 0.0001
40 0.0093 0.0016 0.0004 -0.0007 0.0230 0.0040 0.0011 -0.0000
41 0.0021 0.0002 0.0001 -0.0001 0.0036 0.0004 0.0001 -0.0000
42 0.0009 0.0002 0.0001 -0.0001 0.0021 0.0004 0.0001 -0.0000
43 0.0006 0.0001 0.0000 -0.0000 0.0016 0.0003 0.0001 -0.0000
44 0.0009 0.0002 0.0001 -0.0001 0.0023 0.0004 0.0001 0.0000
45 0.0024 0.0004 0.0001 -0.0002 0.0058 0.0009 0.0003 -0.0000
46 0.0031 0.0005 0.0001 -0.0002 0.0074 0.0012 0.0003 -0.0000
Continued. . .
100
Table 2.6 – continued from previous page
Brand
1: Complete Labeling 2: No Labeling
Market
Share
Revenue Profit Profit
Chg
Market
Share
Revenue Profit Profit
Chg
47 0.0019 0.0004 0.0001 -0.0002 0.0049 0.0009 0.0003 -0.0000
48 0.0001 0.0000 -0.0000 -0.0000 0.0006 0.0002 0.0001 0.0001
49 0.0001 0.0000 0.0000 -0.0000 0.0012 0.0002 0.0002 0.0002
50 0.0002 0.0000 0.0000 -0.0001 0.0028 0.0006 0.0002 0.0002
Note: The values in each column represent the mean volume-weighted values of a given
variable in the simulated scenario for each brand, across all 5,988 markets.
101
Table 2.7: Welfare Effects of Simulated Labeling Scenarios
Variable Scenario 1:
Complete Labeling
Scenario 2:
No Labeling
No
Correction
Leggett
Correction
No
Correction
Leggett
Correction
Volume-Weighted Mean CV -0.08708 0.00933 -0.01496 -0.01468
Population-Weighted Mean CV -0.03151 0.03248 -0.01450 -0.01385
Note: To calculate Mean CV, individual compensating variation, as defined in Equa-
tion 2.14 and Equation 2.15, is averaged over all 50 individuals in a given market t.
A weighted-average of Mean CV is then calculated over all 5,988 markets.
102
Chapter 3: Produce Growers’ Cost of Complying with the Food Safety
Modernization Act∗,†
3.1 Introduction
The enactment of the Food Safety Modernization Act [FSMA] gave the Food
and Drug Administration [FDA] authority to regulate the growing, harvesting, pack-
ing, and holding of fresh fruits and vegetables and represents a major shift in the
agency’s approach from outbreak response to prevention-based controls across the
food supply. Data from the Centers for Disease Control and Prevention indicate
that fruits and vegetables accounted for 46% of foodborne illness outbreaks dur-
ing the period 1998-2008, a larger share than any other category of food (Painter
2013). As one of the implementing rules for FSMA, the FDA has implemented a rule
(known popularly as the Produce Rule) intended to reduce health risks associated
with foodborne illness from consumption of fresh produce. That rule, which became
effective in January of 2016, requires operational changes to meet standards asso-
∗This chapter is co-authored with Erik Lichtenberg, Department of Agricultural and Resource
Economics, University of Maryland, College Park.
†This material is based upon research supported by the National Institute of Food and Agri-
culture, Specialty Crop Research Initiative, Award No. 2011-51181-30767. Any opinions, findings,
conclusions, or recommendations expressed in this publication are those of the author(s) and do
not necessarily reflect the view of the U.S. Department of Agriculture.
103
ciated with agricultural water; biological soil amendments; domesticated and wild
animals; employee training and health and hygiene; and equipment, tools, buildings,
and sanitation. Those changes could be costly.
Small farms and sustainable growers have voiced fears that the Rule will have
adverse competitive effects. Small farms worry that the costs of complying with
the new rule may be disproportionately burdensome and could drive them out of
business (Hassanein 2011; Paggi et al. 2013; Knutson et al. 2014; Ribera et al. 2016).
Farms employing sustainable agricultural practices are concerned that the Rule may
make it impossible for them to use the biological soil amendments and livestock
grazing practices in integrated agricultural systems on which they currently rely.
Both concerns suggest that the Rule may adversely affect important segments of
the produce industry (Ribera and Knutson 2011) that are subject to an increasing
focus nationwide in terms of marketing, consumption, and federal policy (Low 2015).
There is very little publicly available information on the likely cost of the
actions required under the new Rule. To help fill that information gap, we use
data from an original national survey of fruit and vegetable growers and construct a
larger sample than most comparable studies. We examine whether compliance with
the standards in the Produce Rule is likely to be disproportionately burdensome to
small and/or sustainable farm operations. To investigate that question, we analyze
how the costs of food safety practices required by the Produce Rule vary with farm
size and use of sustainable farming practices using a double hurdle model to control
for selectivity in both using food safety practices and reporting expenditures on
them. We use our estimates to quantify how the cost burden of compliance varies
104
with farm size. We then explore the policy implications of exemptions to the Rule
by simulating how changes to the exemption thresholds for farm revenue and share
of direct sales might affect the cost burden of each food safety practice on farms at
the threshold.
3.2 Background
3.2.1 Relevant Literature
Several recent studies analyze the costs of implementing practices made manda-
tory by the new on-farm produce safety standards. Hardesty and Kusunose (2009)
use data from a survey of 49 California growers to estimate the compliance costs
for food safety standards imposed by the California Leafy Greens Marketing Agree-
ment [LGMA], which are similar to those required under the proposed Produce
Rule. They find that growers’ seasonal food safety costs more than doubled af-
ter implementation of the LGMA, and the largest growers benefit from significant
economies of scale. Ribera et al. (2012) conduct three case studies of food safety
outbreaks in muskmelon, spinach, and tomatoes and use a survey of producers par-
ticipating in the California LGMA to estimate the compliance costs for new food
safety standards. They find that the costs incurred by producers due to food safety
outbreaks are much greater than LGMA compliance costs, and the most significant
compliance cost increases are attributable to third party audits, staffing, and water
testing. Paggi et al. (2013) use results from several studies to develop an example
of the impact and compliance costs of LGMA-type standards for Florida cabbage
105
producers and find that for a representative grower, the probability of operating
at a net loss (in present value terms) over a 2-year period increased by 17%. To
assess the impacts from FSMA, Ribera et al. (2014) develop representative farms
for cabbage, cantaloupe, citrus, onion, spinach, tomato, and watermelon production
in California, Florida, and Texas. They find that the cost of complying with the
Produce Rule is not size neutral and can have negative impacts on the profitability
of small farms. A University of Minnesota study uses data collected from in-person
and telephone interviews with small and mid-sized vegetable farmers to estimate to-
tal costs incurred for Minnesota vegetable growers to adopt GAPs practices on their
farms (Driven to Discover 2012). The study also finds that compliance costs exhibit
significant economies of scale—small farms in Minnesota would face food safety costs
equal to 10% of gross revenue, while average-sized farm operations would incur costs
around 2% of gross revenue. Lichtenberg and Page (2016) use data from a survey
of Mid-Atlantic leafy greens and tomato growers to assess the likely cost burden of
adopting on-farm food safety measures like those required under FSMA. They find
substantial economies of scale but a fairly modest cost burden on farms of all sizes.
Concerns about the burdens imposed on small producers—and the resulting
potential for increased industry concentration—have been recurring issues in discus-
sions of food safety regulation. In the late 1990s, compliance with Hazard Analysis
and Critical Control Points (HACCP) regulations in meat packing and processing
was believed to exhibit economies of scale because large firms had in-house testing
facilities, technicians, and other investments in physical and human capital that
small firms did not have, as well as being able to benefit from bulk discounts for
106
outside testing services (MacDonald and Crutchfield 1996; Loader and Hobbs 1999;
Unnevehr and Jensen 1999; Antle 1999). Similarly, the cost burden of adopting
global ISO 9000 food safety certification standards was thought to involve propor-
tionally greater resources for small firms than large ones (Holleran et al. 1999).
The empirical evidence on this score is decidedly mixed. An econometric
analysis by Antle (2000) found no evidence of differences in HACCP compliance costs
in U.S. beef, pork, and poultry processing. An econometric study by Hooker et al.
(2002), by contrast, found that small meat processors incurred higher compliance
costs, while Muth et al. (2003) found that meat and poultry plants classified as
small or very small under HACCP regulations were more likely to exit during the
period of HACCP implementation, although the differential effect of regulation on
exit by plant size was quantitatively quite small.
3.2.2 The Food Safety Modernization Act and the Produce Rule
FSMA was signed into law in January of 2011. While a growing number of
supermarket chains, commodity group organizations and others had been instituting
private food standards for food quality and safety over the preceding decade, a series
of bacterial outbreaks during the mid-2000s indicated that such voluntary efforts
would be insufficient to provide adequate levels of safety (Henson and Reardon
2005; Paggi et al. 2013). The FDA published the original proposed Produce Rule,
officially known as Standards for the Growing, Harvesting, Packing, and Holding of
Produce for Human Consumption, in January of 2013. The Rule was finalized in
107
November of 20153 and became effective in January of 2016. It establishes standards
across various aspects of agricultural production, most notably with regards to:
(1) agricultural water; (2) biological soil amendments of animal origin; (3) health
and hygiene; (4) intrusion of domesticated and wild animals; and (5) sanitation of
equipment, tools, and buildings.
For agricultural water that contacts produce or food-contact surfaces, the Rule
establishes quality standards, periodic inspection and testing provisions, and treat-
ment requirements for water not meeting sanitary standards. For soil amendments
of animal origin, the Rule establishes treatment standards and application require-
ments for treated and untreated soil amendments. For health and hygiene, the rule
establishes hygienic practices and training requirements for all farm personnel who
handle produce covered by the Rule. For intrusion of domesticated and wild animals,
the Rule establishes waiting periods between grazing and crop harvest for domes-
ticated animals and monitoring requirements for wild animal intrusion. Lastly, the
Rule establishes sanitary standards for equipment and tools that come in contact
with produce, as well as requirements for pest control, hand washing and toilet fa-
cilities, and sewage and trash disposal. In addition to these measures, the Rule also
requires recordkeeping and documentation to show compliance with each standard.
The Produce Rule applies to farms that grow and sell produce usually con-
sumed raw and not intended for commercial processing (e.g., canning, etc.). Farms
whose annual produce sales averaged less than $25,000 during the preceding three
3The FDA revised the Rule twice in the interim period, each time based on feedback from a
public comment period.
108
years are exempt. Farms that have three-year annual produce sales of less than
$500,000 and sell a majority of food directly to a qualified end-user—a consumer,
restaurant, or retail food establishment (e.g., a supermarket, etc.) located in-state
or within 275 miles of the farm—are not subject to the food safety standards in the
Rule.4 Additionally, the compliance dates in the Rule allow more time for smaller
farms to adopt the established safety provisions. Farms with annual produce sales
between $25,000 and $250,000—classified as “very small” farms in the Rule—have
four years after the Rule’s effective date to comply with most provisions. Farms with
annual produce sales between $250,000 and $500,000—classified as “small” farms in
the Rule—have three years. Farms with annual produce sales over $500,0005 have
two years. Furthermore, the compliance dates for water quality standards (including
testing and recordkeeping) are an additional two years after the compliance dates
for the rest of the Rule.
3.3 Data
3.3.1 Survey Design
We use data from an original national survey of fruit and vegetable growers
to analyze the likely cost burden of produce safety measures required under the
Product Rule.6 The survey includes background questions on farm economics, farm
4This provision is an amendment to FSMA introduced by Senators Jon Tester and Kay Hagan
and is commonly referred to as the Tester-Hagan Exemption or, more formally, direct marketing
modified requirements.
5The Produce Rule does not designate a name for farms that fall into this category, so we refer
to them as “medium/large” farms throughout the text and tables.
6Appendix C.2 includes a full version of the survey instrument used.
109
characteristics, and use of marketing channels in addition to questions regarding use
and treatment of soil amendments, microbial testing, field monitoring, remedial food
safety actions, preventive food safety actions, and recordkeeping. Soil amendment
questions covered whether animal-based soil amendments were used; whether they
were treated, and if so, at what cost; and the time interval between application
of soil amendments and crop harvest. Microbial testing questions covered whether
the farm collected water, soil amendment, and/or crop samples for testing, and
if so, at what cost (including employee wages, materials, etc.). Field monitoring
questions covered whether the fields were monitored for animal intrusion, flooding,
and/or other contamination; how often these events were observed; and the costs
associated with monitoring the fields. Remedial action food safety questions covered
whether any remedial actions (e.g., sanitation, product disposal, water treatment,
etc.) were taken following test results, flooding, and/or animal intrusion, and if so,
at what cost. Preventive food safety questions covered whether harvest containers
were sanitized prior to harvest or if new containers were used, whether crops were
washed prior to sale, whether the farm used third party food safety audits, and
whether precautions were taken with regards to employee sanitation and hygiene
(e.g., training, tool sanitation, toilet and hand washing facilities, etc.). Lastly,
recordkeeping questions covered whether the farm kept written records for food
safety practices, and if so, how many hours each week were spent doing so.
110
3.3.2 Survey Administration
The survey was designed and administered electronically using Qualtrics sur-
vey software. It included skip logic so respondents only answered questions relevant
to their farm operation. Data was collected in person at eight major fruit and
vegetable grower conferences across the U.S. and through online grower listservs
provided by several state fruit and vegetable growers’ associations, university ex-
tension services, and other grower organizations via a password-protected Internet
survey. To address concerns related to the Rule’s impact on sustainable growers, we
also surveyed members of listservs for several sustainable grower organizations and
attendees at a major sustainable grower conference.7 At the conferences, a booth
was set up alongside other exhibitors in the trade show or a similar high traffic area,
and attendees passing by the booth were asked to participate in the survey, after
which they could enter a drawing for a chance to win an Apple iPad. After con-
senting to participate, respondents completed the fifteen-minute survey on tablet
computers, providing information about the 2014 growing season. Upon completing
the survey, the software automatically redirected each respondent to a separate form
in which she could choose to enter the iPad drawing.
The survey sent to grower listservs was identical to the version administered at
the grower conferences, except that respondents completed the survey independently
on their own web-enabled device, typically either a personal computer or tablet.
Members of each listserv were sent an email soliciting their participation in the
7Appendix C.1 includes detailed lists of grower conferences and online listservs from which
responses were collected.
111
survey. Each email included a description of the survey and research goal, informed
consent agreement, our contact information, and a web link and password to take the
survey. We used the password to identify the listserv through which the respondent
was contacted. The survey software uses browser cookies to prevent respondents
from taking the survey more than once. As an incentive to participate, online
respondents were given the same offer to enter a drawing for an iPad after completing
the survey. All the survey collectors for the grower listservs remained open until
May 2, 2015.
3.3.3 Summary Statistics
In total, 394 growers completed the survey. A large majority (277, about 70%)
came from listservs while the remainder (117, about 30%) completed the survey
at a grower convention (Table 3.1). Almost 80% of the respondents (311) grow
vegetables, 193 grow berries, and 194 grow fruit and tree nuts (many growers raise
produce from more than one category).
Our sample is weighted towards commercial-size farms compared to the 2012
USDA Agricultural Census as measured by revenue and acreage (Figure 3.1). In
terms of regional distribution, our sample consists of relatively more farms in the
Northeast and South and fewer farms in the West than the U.S. as a whole (Ta-
ble 3.2). The West consists of states that represent a significant portion of large
fruit and vegetable agribusiness growers in the U.S. (e.g., California, Arizona, etc.)
that account for a majority of the produce consumed nationwide. Many of these
112
Western growers were already obligated to meet food safety standards under the
LGMA, which are quite similar to those required under the Produce Rule. The
geographic distribution of respondents suggests that, while our sample may be less
representative of total U.S. fruit and vegetable production, it is more representative
of the population of U.S. produce growers likely to be affected by the Produce Rule,
which is ultimately of greater relevance to the issues we wish to address.
Overall, one-third of our sample comes from members of sustainable grower
organizations,8 with the balance comprised of members of conventional grower orga-
nizations. In terms of farm size, sustainable growers in our sample tend to operate
smaller farms than conventional growers, with about half of sustainable growers
falling into the exempt farm classification compared to just over a quarter of con-
ventional growers (Table 3.3). On average, growers in our sample sell more than
half their produce directly to consumers (Table 3.1).
Respondents from grower conventions and listservs are quite similar in most
respects, the major difference being that almost all respondents identified as sus-
tainable came from listservs. Likely for that reason, respondents from conventions
are slightly larger, sell less of their output direct to consumers, do more sampling
and testing, and are a little less likely to be diversified. All of those differences are
small, however.
Based on current sales and practices, over three-fourths of our sample are
exempt from the Produce Rule, which is almost evenly split based on the farm size
8To be concise, we refer to members of sustainable grower organizations as “sustainable growers”
and members of conventional grower organizations as “conventional growers” throughout the text.
113
and the Tester-Hagan exemptions (Table 3.3). Almost all very small growers and
three-fourths of small growers in our sample qualify for the Tester-Hagan exemption.
In terms of farming practices, nearly all sustainable growers in our sample qualify
for an exemption, with over half exempt based on size, and about two-thirds of
conventional growers are exempt.
3.4 Profit Maximizing Choice of Food Safety Practices: Theory and
Econometric Specification of Expenditures
We are interested in how the costs of implementing food safety practices re-
quired under the Produce Rule vary with respect to farm size. To fix ideas, consider
a grower using productive inputs X1, . . . , XJ and food safety practices Z1, . . . , ZK
to produce a vector of marketable outputs Y on a farm of fixed acreage A using
a technology represented by a transformation function f(Y,X,Z;A) ≤ 0. Here
marketable output represents a combination of quantity and perceived quality (i.e.,
marketable outputs can differ in terms of quality (locally grown, sustainably grown,
heirloom variety, etc.) and thus price received).
The grower wants to maximize profit:
pi = pY −wX− vZ− F s.t. f(Y,X,Z;A) ≤ 0 (3.1)
where p, w, v, and F are vectors of output prices, input prices, food safety prac-
tice costs, and fixed costs, respectively. The necessary conditions imply optimal
114
use of productive inputs X∗j (p,w,v;A) and food safety practices Z
∗
k(p,w,v;A).
Expenditure on food safety practice k is
ek = vkZ
∗
k .
The change in expenditure on food safety practice k with respect to farm size A is
∂ek
∂A
= vk
∂Zk
∂A
=
vkZk
A
∂Zk
∂A
A
Zk
The elasticity of expenditure on food safety practice k with respect to farm
size A is thus
ηk ≡ A
ek
∂ek
∂A
=
∂Zk
∂A
Zk
A
(3.2)
If the marginal change in use of food safety practice k as size (acreage) increases
is less than the average, then expenditure on food safety practice k is inelastic.
Inelastic expenditure on a food safety practice implies that larger operations spend
less per acre than smaller ones; elastic expenditure implies that larger operations
spend more per acre than smaller ones; and unit elastic expenditure implies that
expenditure per acre is constant (invariant with respect to acreage).
Since we are interested in estimating the elasticities of food safety practices re-
quired under the Produce Rule, we assume a log-linear specification of expenditures.
115
For grower j, we have
log(ejk) = βk +
∑
i
γiTji + ηk log(Aj) + ujk (3.3)
Here Tji represents variables like crop type and an indicator for sustainable farming
practices assumed to shift expenditures on average and ujk is a normally distributed
white noise error representing all unobservables that influence expenditure on food
safety practice k. We are especially interested in the coefficient of the “sustain-
able” indicator, i.e., average differences in compliance cost between conventional
and sustainable growers, since the latter have been among the most vocal in ex-
pressing concerns about the impacts of compliance with the Produce Rule on their
operations.
For food safety measures that are part of a larger group of practices (e.g., em-
ployee training is one action in employee sanitation and hygiene, etc.), we stream-
lined the survey by asking respondents to provide total cost estimates for the overall
group of food safety practices: all forms of testing, all field inspections, all harvest
container sanitation measures, all employee sanitation and hygiene measures, and
treatment of all soil amendments. For these food safety practices, we control for
each specific action by including indicator variables for each type of action used by
the grower. For these groups of practices, we specify expenditures as
log(ejk) = βk +
∑
m
(βkmPjkm) +
∑
i
γiTji + log(Aj)
∑
m
(Pjkmηkm) + ujk (3.4)
116
where Pjkm is an indicator variable taking on a value of 1 if grower j uses individual
action m within the group of practices k. To assess robustness, we also estimate a
model where the elasticities of all practices within a group are assumed equal, as
specified in Equation 3.3.
3.5 Estimation Method
Estimating the parameters of the models specified in the preceding section
is complicated by two factors: (1) we observe cost only for growers who actually
use a food safety practice and (2) some growers who used a food safety practice
did not report cost. For example, between 6% and 22% of respondents in our
sample reported not using one or more of these practices, with as many as 40-60%
reporting not doing any sampling, inspection, or third party audits. Of those who
reported using these food safety practices, between 5% and 30% failed to report
cost. Not surprisingly, non-response rates for cost were substantially higher among
respondents who completed the survey at grower conventions (20-50% of users) than
those responding from listserv solicitations (3-17% of users).
It is likely that unobserved factors affect the probability of a grower using
a specific food safety practice, the cost of implementing that practice, and the
probability that a grower fails to report the cost of that practice. We therefore
estimate a double hurdle model to control for potential selection bias from both
non-use and non-reporting among users. We specify the probability that grower j
uses food safety practice (or group of food safety practices) k as
117
Pr(vjk ≤ ξk + µkOj +
∑
i
ζiTji + θk log(Aj)) = Φ(ξk + µkOj +
∑
i
ζiTji + θk log(Aj)) (3.5)
where vjk is a normally distributed white noise error representing all unobservables
that influence the choice of whether to use food safety practice k, and Φ(·) denotes
a standard normal cumulative distribution.
The selection equation contains two variable not included in the expenditure
equation: Oj, an indicator taking on a value of 1 if grower j has a contractual
obligation to use one or more food safety practices, and the share of output sold
to wholesalers/repackers, mass merchandisers, exporters, brokers, or other outlets
(included in the set of grower characteristics Tji). These two variables thus serve as
instruments to identify selection.
A substantial theoretical literature suggests that marketing channels may be
important in creating incentives for growers to adopt food safety practices (Henson
and Caswell 1999; Segerson 1999; Fares and Rouviere 2010; Hennessy et al. 2001;
Henson and Reardon 2005; Fulponi 2006; Carriquiry and Babcock 2007; Rouvie`re
and Caswell 2012). A foodborne illness outbreak can damage the reputation of
a downstream agent that sells directly to consumers (such as a grocery store or
restaurant), and, if a seller is found liable, can result in direct financial losses as
well. Therefore, in some marketing channels downstream buyers may require pro-
duce growers to use certain food safety practices, motivating our use of Oj as an
instrument for selection. Farmers that sell all of their output direct to consumers
118
have no contractual obligations; about 27% of farmers with some direct sales fall
into this category and are thus recorded as having no contractual obligations to use
any food safety practices. (Oj = 0 if the share of direct sales = 100%). To avoid
collinearity, we omit the share of output sold to grocery retailers and foodservice
operations, which are primarily local and thus have a high degree of traceability
and large reputation effects. The same considerations lead us to drop the share of
output sold direct to consumers from the selection equation as well.
We include the indicator of sustainable production practices in the selection
equation to examine how sustainable and conventional growers differ in their use of
food safety practices. There is considerable evidence that consumers view sustain-
ably grown food as safer than conventionally grown foods (see Bourn and Prescott
2002; Yiridoe et al. 2005; Hughner et al. 2007, for example). Additionally, sustain-
able growers tend to utilize more direct-to-consumer marketing arrangements such
as CSAs and farmers’ markets to sell their produce (Martinez 2010) where trace-
ability is likely high. These considerations suggest that sustainable growers may be
less likely than conventional growers to invest in food safety practices, other than
washing product to ensure its attractiveness to potential buyers.
Now combine the selection and expenditure models for each practice. Let Ijk
be an indicator taking on a value of 1 if grower j uses practice k and a value of 0
if grower j does not use practice k. Normalize the variance of vjk to 1 and let σ
denote the variance of ujk and ρ denote the correlation between ujk and vjk. The log
likelihood function for a single food safety practice taking into account both non-use
and non-reporting is:
119
log(Lk) =∑
j|Ijk=0
log Φ
(
−
(
ξk + µkOj +
∑
i
ζiTji + θk log(Aj)
))
+
∑
j|Ijk=1,ejk=0
log Φ
(
ξk + µkOj +
∑
i
ζiTji + θk log(Aj)
)
+
∑
j|Ijk=1,ejk>0
log Φ
(
ξk + µkOj +
∑
i ζiTji + θk log(Aj) +
ρ
σ (log(ejk)− βk −
∑
i γiTji − ηk log(Aj))√
1− ρ2
)
+
∑
j|Ijk=1,ejk>0
log
(
1
σ
φ
(
log(ejk)− βk −
∑
i γiTji − ηk log(Aj)
σ
))
(3.6)
The first term in the log likelihood function is the probability that grower j does
not use practice k, summed up over all growers not using practice k. The second
term is the probability that grower j uses practice k, summed up over all growers
using practice k but not reporting expenditures on practice k. The third term is the
probability that grower j uses practice k conditional on grower j reporting positive
expenditures on practice k, summed up over all growers reporting expenditures on
practice k. The fourth term is the probability of observing reported expenditures
log(ejk), summed up over all growers reporting expenditures on practice k. The third
and fourth terms combined thus represent the unconditional probability of observing
reported expenditures log(ejk) summed up over all growers reporting expenditures
on practice k.
The model for groups of practices is analogous. We estimate two versions of
the model for cases where total expenditures are reported for groups of practices.
As presented in Equation 3.4, the main specification adds individual practices in-
tercepts,
∑
m(Pjkmβkm), to the expenditure portion of the model; and it allows the
120
acreage elasticities of each individual practice to vary by replacing ηk log(Aj) with
log(Aj)
∑
m(Pjkmηkm). A second specification uses only the group-level intercept βk
and assumes that the acreage elasticities of all practices within each group are the
same, thus using ηk log(Aj), as presented in Equation 3.3. The specification of the
selection equation is the same in both cases.
We estimate the parameters of these double hurdle models simultaneously for
each practice or group of practices, i.e., our estimates are full information maximum
likelihood estimates for each practice or group of practices.
3.6 Estimation Results
The elasticities of food safety practice expenditures, the impacts of sustain-
able production practices on food safety expenditures, and the marginal effects of
factors influencing the probability of using a food safety practice, as estimated by
maximum likelihood using the double hurdle model specified above, are reported in
Tables 3.4, 3.5, and 3.6.9 The number of observations used in each model varies
because of differences in response rates for questions about the use of food safety
practices (Table 3.4); missing observations for fruit and vegetable acreage Aj and
the indicator for contractual obligations to use food safety practices Oj reduced
the size of the usable sample by an additional 24-25 observations. The errors for
sampling and testing, field inspection, container sanitation, and written records are
positively correlated, underscoring the need for the double hurdle specification (Ap-
9Full sets of coefficient estimates for the double hurdle models without and with individual
practice intercepts and interaction terms are presented in Appendix Tables C.2 and C.3, respec-
tively.
121
pendix Table C.2). Because the probit selection model is nonlinear, we calculate
average marginal effects for each of the regressors to help quantify the effect farm
size, grower characteristics, and marketing channel have on the probability that
growers use each safety measure (Table 3.6). The estimated marginal effects re-
ported in Table 3.6 indicate that the contractual obligations indicator is a strong
instrument for sampling and testing, field inspections, harvest container sanitation,
written record keeping, and third party audits. Similarly, the share of output sold
through wholesalers/repackers, mass merchandisers, exporters, brokers, or other
outlets is a strong instrument for employee sanitation and hygiene practices. Only
soil amendment treatment lacks a strong instrument for selection, likely because
relatively few growers in our sample use soil amendments.
3.6.1 Effect of Farm Size on Expenditures on Food Safety Practices
There is concern that the Rule may adversely affect small produce growers
and that compliance costs may be sufficiently burdensome to force many out of
business, thereby increasing concentration in the industry. To better gauge this
concern, we investigate how the expenditures and use of food safety practices vary
with farm size. For each food safety practice, we use the estimated coefficient on the
log of acreage (the elasticity of food safety expenditure for practice k with respect
to acreage) to analyze how the expenditures for each food safety practice varies with
farm size (Table 3.4). As noted earlier, inelastic expenditure on a food safety practice
implies that larger operations spend less per acre than smaller ones, consistent with
122
increasing returns to farm size; elastic expenditure implies that larger operations
spend more per acre than smaller ones, consistent with decreasing returns to farm
size; and unit elastic expenditure implies that expenditure per acre is constant,
consistent with constant returns to farm size.
Consistent with previous studies of specific crops and geographic areas (Hard-
esty and Kusunose 2009; Ribera et al. 2012; Paggi et al. 2013; Driven to Discover
2012; Lichtenberg and Page 2016), the estimated coefficients for all but two of the
food safety practices included in our survey data indicate that expenditures increase
with operation size but less than proportionally.
In most cases, the estimated elasticities are not significantly different from
zero, consistent with expenditures being fixed and thus decreasing in farm size. Ad-
ditionally, likelihood ratio tests indicate that the estimated acreage elasticities of
expenditure on soil samples (χ2 = 28.72, p < 0.01),10 inspection for other causes
(χ2 = 38.41, p < 0.01), new harvest containers (χ2 = 101.4, p < 0.01), employee
education and training (χ2 = 48.86, p < 0.01), equipment and tool sanitation (χ2 =
73.53, p < 0.01), building sanitation (χ2 = 79.90, p < 0.01), sewage and trash dis-
posal (χ2 = 36.24, p < 0.01), and other preventive measures (χ2 = 40.61, p < 0.01)
are all significantly different from one, ruling out expenditures per acre proportional
to farm size.
In several other cases, the estimated elasticities are less than one but signifi-
cantly different from zero, ruling out expenditures independent of farm size. Like-
10Likelihood ratio tests for inspection for flooding and wildlife intrusion are each conducted on
the sum of the individual flooding/wildlife intrusion and the simultaneous inspection for flooding
and wildlife intrusion simultaneously.
123
lihood ratio tests indicate that the estimated elasticities of water samples (χ2 =
35.42, p < 0.01), washing harvest containers (χ2 = 23.82, p < 0.01), washing prod-
uct (χ2 = 19.60, p < 0.01), keeping written records (χ2 = 319.39, p < 0.01), and
third party audits (χ2 = 39.58, p < 0.01) are all significantly different from one as
well as zero, consistent with expenditures per acre being variable but decreasing in
farm size.
The estimated coefficients of the remaining food safety practices are somewhat
mixed. Likelihood ratio tests indicate that the estimated elasticity of treating mul-
tiple soil amendments is not significantly different from one (χ2 = 0.38, p = 0.54)
even though it is significantly different from zero, consistent with constant ex-
penditures per acre. The estimated elasticities of inspection for wildlife intrusion
(χ2 = 0.16, p = 0.69) and treating a single soil amendment (χ2 = 0.31, p = 0.58)
are not significantly different from either zero or one. Both are close in magnitude
to one, though (recall that the elasticity of inspection for wildlife intrusion is the
sum of the wildlife intrusion and wildlife intrusion and flooding elasticities, 0.75),
suggesting that expenditures on these two practices are also likely constant per acre.
3.6.2 Effect of Sustainable Farming Practices on Expenditures on
Food Safety Practices
Sustainably grown food products, which are often sold locally through direct-
to-consumer channels such as farmers’ markets and community supported agricul-
ture arrangements, account for a growing share of total U.S. agricultural sales (Mar-
124
tinez 2010). Between 2006 and 2014, the number of farmers’ markets in the U.S.
grew 180 percent, and local food sales totaled an estimated $6.1 billion in 2012.
Business survival also appears to be greater for farms selling into these markets
(Low 2015).
Sustainable grower organizations have expressed concern that some provisions
of the Produce Rule may be unduly burdensome to sustainable growers, particularly
for the treatment of animal-based soil amendments. We use the estimated coefficient
of the sustainable farming practices indicator to examine this possibility. The esti-
mated coefficient of the sustainable indicator is positive for all food safety practices
and significantly different from zero in the regressions for expenditures on harvest
container sanitation, washing product, and keeping written records. All three indi-
cate that sustainable growers spend as much as twice as much on these practices as
conventional growers. However, it is by no means clear that greater expenditures on
washing product and harvest container sanitation can be attributed to compliance
with the Produce Rule, since they could easily be due to marketing considerations
that predate the Produce Rule (e.g., delivering clean products in clean containers
direct to consumers).
The estimated coefficients of the sustainable farming practices indicator in
the equations for sampling and testing, field inspection, and written recordkeeping
all suggest that sustainable grower spend 40-60% more on these practices than
conventional growers. All of these coefficients except for that of written records
are estimated imprecisely, however. Overall, then, the estimated coefficients of
our econometric models provide limited evidence supporting the contention that
125
compliance with the Produce Rule be more burdensome for sustainable growers
than conventional ones.11
3.6.3 Effect of Farm Size and Sustainable Farming Practices on Use
of Food Safety Practices
Lower costs should make larger farms more likely to invest in food safety. We
find evidence supporting that hypothesis for some food safety practices. We find
positive, statistically significant marginal effects of acreage on the probability of
using sampling and testing, conducting field inspections, keeping written records,
and having third-party audits. The effects of acreage on sampling and testing,
field inspections, and third party audits are substantial: a 1 percent increase in
acreage is associated with a 5-6 percentage point increase in the probability that each
practice is used. The effect of acreage on written records is somewhat smaller but
still substantial—a 1% increase in acreage is associated with a 2.5 percentage point
increase in the probability of keeping written records. The estimated marginal effects
for the remaining food safety practices are not statistically significantly different
from zero and quite small in magnitude, implying that larger farms are no more
or less likely than smaller operations to sanitize harvest containers, wash product,
utilize employee sanitation and hygiene measures, or treat soil amendments.
We argued earlier that growers using sustainable farming practices might be
11For completeness, we investigate whether the elasticities of food safety practice expenditures
with respect to acreage differed for conventional and sustainable grower. None of the coefficients of
the interaction between the log of acreage and the sustainable grower indicator were significantly
different from zero. All were small in magnitude as well.
126
less likely to invest in food safety measures since such investments would have low
returns. Our econometric results provide another reason: higher costs of food safety
measures. We find that sustainable growers are neither more nor less likely than
conventional growers to employ most food safety measures. The exception is washing
product: sustainable growers are 21 percentage points more likely to wash product,
a result that is consistent with their heavier reliance on selling direct to consumers
with whom they are likely to have personal ties. Overall, these estimates suggest
that the Produce Rule will require changes in growing practices for similar shares
of sustainable and conventional growers.
3.7 Policy Implications
Our estimates of the elasticities of food safety expenditures with respect to
acreage indicate that per acre expenditures on most food safety practices decrease
with farm size, suggesting that compliance with the Produce Rule will be more dif-
ficult for smaller operators than for larger ones. We examine the potential impacts
of compliance with the Rule more closely by examining the cost burden of compli-
ance. We define the cost burden of food safety practice k as expenditures on food
safety practice k as a share of total production expenditures ek/
(∑
j wjXj +
∑
n en
)
. We
investigate how that cost burden varies with (a) farm size and (b) the exemption
thresholds using simulations based on our econometric estimates.
127
3.7.1 Farm Size and the Cost Burden
The elasticity of this cost burden with respect to acreage is ψk = ηk − λ,
where λ denotes the elasticity of total expenditure with respect to acreage. We use
our survey data to obtain an estimate of λ, which we obtain by regressing the log
of total expenditures for fruit and vegetable production reported by each grower
on the log of fruit and vegetable acreage. The estimate of λ obtained from this
regression is 1.0049 (with a standard error of 0.0459, R2 = 0.66 and N = 244) and
the constant term is 8.3201 (with a standard error of 0.1450). Since the estimated
elasticity is indistinguishable from one, we set λ = 1 in the simulations that follow
in this section.
Overall, our results support the contention that the cost of compliance with
the Produce Rule requirements will be more burdensome to small farms than large
ones. The cost burden of food safety practices for which expenditures are fixed
(invariant with respect to farm size) is inversely proportional to acreage. As we have
seen, those practices include inspection for flooding, inspection for other causes,
use of new harvest containers, employee education and training, equipment and
tool sanitation, building sanitation, sewage and trash disposal, and other preventive
sanitation measures. For each of these, a farm with ten times the acreage of a smaller
farm will have a cost burden only a tenth the size that of the smaller farm. Similarly,
the cost burden of food safety practices with acreage elasticities between zero and
one will decline with farm size, albeit at a lower rate. The acreage elasticities of
testing water samples and washing harvest containers, for instance, are about 0.55,
128
implying an elasticity of the cost burden of -0.45 for each of these practices. The cost
burden of these two practices for a farm with ten times the acreage of a smaller farm
will be about 35% (10−0.45) that of the smaller farm. Only in the cases of inspection
for wildlife intrusion and testing soil amendments (either single or multiple) does
the cost burden appear to be invariant with respect to farm size. Overall, then, our
estimated coefficients indicate that that the Produce Rule will impose a much larger
food safety cost burden on smaller operations than larger ones, consistent with the
concerns raised by many small farm advocacy organizations.
3.7.2 Effect of Changes to FSMA Exemption Thresholds on Food
Safety Cost Burden
The farm size and Tester-Hagan exemptions to FSMA are intended to protect
very small and local food producers for whom the costs of complying with FSMA
would be unduly burdensome. To better understand how growers at these exemption
thresholds are affected by FSMA, we simulate how the magnitude of the food safety
cost burden varies by farm revenue and share of direct sales. The former determines
the size threshold for exemption while the two combined determine the Tester-Hagan
exemption.
The cost burden of food safety practice k with our log-linear specification
is exp(αk − α1)A(ηk−λ); here α1 is the constant term in the regression of log of
total expenditures on log of acreage and αk is the constant term in the expen-
diture equation for food safety practice k, which is different for each combina-
129
tion of crop type and farming practice. To conduct this simulation, we need to
know how acreage varies with revenue and the share of direct sales. We use our
survey data to obtain estimates of these two parameters by regressing the log
of fruit and vegetable acreage on the log of fruit and vegetable revenue and the
share of fruits and vegetables sold direct to consumers. Our estimate of the cost
burden for a threshold combination of revenue R and share of direct sales s is:
exp(αk − α1)ψ(ηk−λ)0 Rψ1(ηk−λ) exp (ψ2(ηk − λ)s).12 The coefficient of the log of rev-
enue is ψ1 = 0.64 (with a standard error of 0.0336). The coefficient of the share
of direct sales is ψ2 = −0.0115 (with a standard error of 0.002). The constant
term is ψ0 = −3.9904 (with a standard error of 0.4456). The regression uses 240
observations and has an R2 = 0.71.
Table 3.7 presents our estimates of the cost burdens of each group of food safety
practices for each combination of crop type and sustainable/conventional farming
practices. To examine the impact of altering the thresholds for the Tester-Hagan
exemption, we estimate cost burdens at the current levels ($500,000 in annual sales
and 50% direct sales) and at combinations representing increases of 50% in each
($750,000 in annual sales and 75% direct sales). For the size exemption, we estimate
cost burdens at the current level ($25,000 in annual sales) and at double that level
($50,000 in annual sales) assuming 49% direct sales in both cases. All estimates are
presented as percentages (i.e., on a scale of 0-100).
Our simulation results suggest that even with significant economies of scale,
12For brevity, we limit our analysis to groups of practices (rather than individual practices within
each group) using the acreage elasticities estimated with no interaction terms.
130
compliance with the Produce Rule would impose modest burdens on growers cur-
rently exempt from the Rule due to Tester-Hagan limits on size and share of direct
sales. Expenditures on sampling and testing, field inspections, and keeping written
records are each under 0.2% of total expenditures. Harvest container sanitation
expenditures are larger but still under 0.75% of total expenditures. Washing veg-
etables requires the greatest expenditures but even this practice amounts to only 3.5
percent of total expenditures at the current exemption limits. Adding up the cost
burden estimates of individual food safety practices for growers with the highest cost
burdens (sustainable vegetable growers with annual sales of $500,000 selling 50% of
output direct to consumers who treat soil amendments) gives an upper bound esti-
mate of the impact of raising Tester-Hagan exemption thresholds. Doing so gives a
total estimated cost burden of 10.5%. This figure is probably an overestimate, since
most growers qualifying for the Tester-Hagan exemption wash product for marketing
reasons already; deducting expenditures on washing product reduces the estimated
cost burden to 7.0%. Such an increase in expenditures is quite modest. It could,
however, correspond to a significant reduction in profit if margins are low.
While a large portion of small growers qualify for exemption from FSMA based
on size or Tester-Hagan rules, small growers with revenue over $25,000 that do not
sell at least 50% through direct marketing channels are not exempt from FSMA. The
impact of raising the exemption threshold on those growers would be considerable.
Consistent with the findings of Ribera et al. (2016), that impact is largely due to the
high cost of third party audits, which amount to about 8.5% of total expenditures
for sustainable berry growers with $25,000 in annual sales and almost 7.5% of total
131
expenditures for conventional berry growers with $25,000 in annual sales. At the
same time, however, the cost burdens of sampling and testing, field inspections, and
keeping written records are quite low, all well under 0.5%, indicating that some of
the food safety practices required under the Produce Rule could be extended to very
small growers at acceptably low cost.
3.8 Conclusion
The passage of the Food Safety Modernization Act in 2011 authorized the FDA
to regulate growing, harvesting, packing, and holding of fresh fruits and vegetables.
The rule for produce safety, which became effective in January of 2016, sets stan-
dards for agricultural water, soil amendments of animal origin, domesticated and
wild animals, employee health and hygiene, and equipment and building sanitation
that could be costly for growers to implement. In particular, small farms worry that
the costs of implementing the required food safety practices could put them out of
business, and sustainable growers are afraid that the new standards could make it
prohibitively expensive for them to maintain their current farming practices.
Data on the likely cost of the actions required under the new Rule is limited.
We use data from a national survey of 394 fruit and vegetable growers to examine
how expenditures on (and thus the cost burden of) food safety practices varies by
farm size and farming practices. We estimate food safety practice expenditures as
a function of acreage, whether sustainable growing practices were used, and other
farm characteristics using a double hurdle model to control for selection in the use
132
of each food safety practice and reporting expenditures on that practice. We find
that expenditures on most food safety practices rise less than proportionally as
farm size increases, implying that the cost burden of using them falls with farm
size. Expenditures on many of these practices appear to be invariant with respect
to farm size, implying that they fall in proportion to the inverse of acreage.
We also find evidence that sustainable growers spend substantially more on av-
erage on many of these practices than conventional growers, although our estimates
are precise only for harvest container sanitation, washing product, and keeping writ-
ten records. Recent years have seen growing consumer interest in local food markets,
of which sustainable farming practices are an integral component (Thompson et al.
2008). While most sustainable growers in our sample are presently exempt from
FSMA, our results suggest that the Produce Rule may pose barriers to expansion of
sustainable farming. Our finding that sustainable growers spend more on average on
food safety practices compared to conventional growers suggest these growers could
face a considerable cost disadvantage upon surpassing the FSMA exemption thresh-
old. While our simulations suggest that the magnitude of this burden is modest for
most growers, the Produce Rule could limit the growth of some sustainable farming
operations, making it desirable for them to stay below the exemption threshold. As
the Produce Rule is phased in over the next three years, it will be important to keep
an eye on its effect on sustainable growers and on the agricultural markets in which
they participate.
We use our econometric results to conduct two simulation exercises examining
how the cost burden of using these food safety practices varies with farm size. The
133
first simulation exercise examines the ratio of food safety expenditures to total
production expenditures, which we define as the cost burden. We find that this cost
burden falls substantially with farm size: For most food safety practices, increasing
farm size by a factor of ten decreases this cost burden by 45-90%. The second
simulation exercise estimates the cost burden for farms that are currently exempt
due to size in terms annual sales or due to the Tester-Hagan amendment, which
exempts farms using a combination of annual sales and share of sales direct to
consumers. This exercise shows that raising exemption thresholds would impose
a relatively small cost burden on farms that are currently exempt by the Tester-
Hagan amendment. For farms that are exempt by size class, raising the threshold for
exemption would impose modest increases in cost for most practices, the exception
being third party audit. We thus find little or no evidence that compliance with the
Produce Rule will threaten the competitiveness of small operations to any significant
extent.
A caveat to the inferences drawn here comes from the nature of our data. Our
sample is broadly representative of commercial produce growers in most of the U.S.
but underrepresents commercial produce growers in California, Arizona, Texas, and
Florida, the states that account for the largest shares of fruit and vegetable produc-
tion. Growers in these underrepresented states operate largely under contracts with
packing firms, marketing firms, and grocery chains; those contracts often mandate
the use of many of the food safety practices considered here. Food safety practice
usage rates and expenditures by growers in these states may thus differ from those
of growers elsewhere in the U.S. Our sample is broadly representative of the popula-
134
tion of fruit and vegetable growers in the U.S., so that our analysis yields reasonable
inferences about the likely effects of the Produce Rule on growers. But our sample
underrepresents fruit and vegetable production and thus cannot be used to draw
inferences about effects of the Produce Rule on fruit and vegetable markets (i.e.,
output and price levels).
135
F
ig
u
re
3.
1:
E
m
p
ir
ic
al
D
is
tr
ib
u
ti
on
s
of
F
ru
it
an
d
V
eg
et
ab
le
F
ar
m
R
ev
en
u
e
an
d
F
ar
m
A
cr
ea
ge
S
ou
rc
e:
2
0
1
2
U
S
D
A
C
en
su
s
o
f
A
g
ri
cu
lt
u
re
n
a
ti
o
n
a
l
v
eg
et
a
b
le
to
ta
ls
N
ot
e:
14
0
re
sp
on
d
en
ts
(3
6%
)
ch
o
se
n
o
t
to
re
p
o
rt
re
ve
n
u
e,
a
n
d
2
9
re
sp
o
n
d
en
ts
(7
%
)
ch
o
se
n
o
t
to
re
p
o
rt
a
cr
ea
g
e.
136
Table 3.1: Survey Descriptive Statistics
Variable Listserv
Average
Conference
Average
Total Ob-
servations
Number of Observations 277 117 394
Use of Food Safety Practices (Share of Subsample)
Sampling and Testing 0.38 0.5 328
Field Inspections 0.36 0.44 318
Harvest Container Sanitation 0.67 0.76 321
Wash Product 0.54 0.59 317
Employee Sanitation and Hygiene 0.66 0.84 313
Written Records 0.56 0.69 309
Soil Amendment Treatment 0.3 0.2 169
Third Party Audits 0.14 0.24 303
Crop Type (Share of Subsample)
Berries Only 0.03 0.07 394
Fruit and Tree Nuts Only 0.17 0.09 394
Vegetables Only 0.25 0.37 394
Fruit/Tree Nuts and Berries 0.03 0.01 394
Vegetables and Berries 0.16 0.23 394
Vegetables and Fruit/Tree Nuts 0.09 0.06 394
Berries, Fruit/Tree Nuts, and
Vegetables
0.27 0.18 394
Farming Practices (Share of Subsample)
Sustainable 0.44 0.07 394
Conventional 0.56 0.93 394
Share of Output Sold through Marketing Channel (%)
Direct Sale 62.18 53.85 338
Wholesale/Mass Merchandiser /
Exporter / Broker / Shipper /
Other
28.14 30.28 338
Grocery Retailers 6.69 13.3 338
Foodservice Operations 3.63 3.05 338
Other Explanatory Variables
Contractual Obligation (Share of
Subsample)
0.23 0.37 326
Log Fruit and Vegetable Acreage 2.14 3.85 365
Log Fruit and Vegetable Expen-
ditures
10.42 11.47 253
Log Fruit and Vegetable Revenue 10.84 11.85 254
137
Table 3.2: Regional Distributions of Farm Operations
U.S. Region U.S. Census of Agriculture (%) Sample (%)
Vegetables 72045 287
Midwest 23.9 25.8
Northeast 19.6 27.9
South 29.8 37.3
West 26.8 9.1
Berries 30538 181
Midwest 22.1 19.9
Northeast 25.5 27.6
South 31.6 42
West 20.8 10.5
Fruit and Tree Nuts 31126 181
Midwest 3.2 21
Northeast 0.7 25.4
South 44.6 28.7
West 51.5 24.9
Source: 2012 USDA Census of Agriculture - Vegetable Operations
Note: 5 respondents (1%) chose not to report state
138
Table 3.3: Exemption Status by Farm Size and Grower Organization
Classification Size Exempt Tester-Hagan
Exempt
Grand Total
Exempt
Economic Class
Exempt 91 (100%) - 91
($25,000 or less)
Very Small - 80 (93%) 80
($25,001 to $250,000)
Small - 24 (75%) 24
($250,001 to $500,000)
Medium/Large - - -
(More than $500,000)
Grower Organization
Conventional 42 (27%) 60 (38%) 102 (65%)
Sustainable 49 (51%) 44 (45%) 93 (96%)
TOTAL 91 (36%) 104 (41%) 195 (77%)
Note: 140 respondents (36%) chose not to report revenue, so exemption status
cannot be determined.
139
Table 3.4: Estimated Farm Size Elasticities of Food Safety Practice Expenditures
Dependent variable = log expenditures With In-
teractions
No
Interaction
Panel A. Sampling and Testing (N = 303)
Log Fruit and Veg. Acreage x Water Samples 0.5510∗∗∗ -
(0.0754)
Log Fruit and Veg. Acreage x Soil Amendment
Samples
0.2094 -
(0.1475)
Log Fruit and Veg. Acreage x Product Samples 0.0604 -
(0.1126)
Log Fruit and Veg. Acreage - 0.6955∗∗∗
(0.0764)
Panel B. Field Inspection (N = 294)
Log Fruit and Veg. Acreage x Flooding -0.2032 -
(0.6088)
Log Fruit and Veg. Acreage x Wildlife Intrusion 0.5097∗∗∗ -
(0.1049)
Log Fruit and Veg. Acreage x Other Causes 0.1554 -
(0.1363)
Log Fruit and Veg. Acreage x Flooding x Wildlife
Intrusion
0.2448 -
(0.6166)
Log Fruit and Veg. Acreage - 0.5608∗∗∗
(0.0795)
Panel C. Harvest Container Sanitation (N = 297)
Log Fruit and Veg. Acreage x Wash Containers -0.0291 -
(0.1022)
Log Fruit and Veg. Acreage x New Containers 0.5607∗∗∗ -
(0.0900)
Log Fruit and Veg. Acreage - 0.5475∗∗∗
(0.0602)
Panel D. Washing Product (N = 294)
Log Fruit and Veg. Acreage - 0.6816∗∗∗
(0.0719)
Panel E. Employee Sanitation and Hygiene (N = 290)
Log Fruit and Veg. Acreage x Employee Training 0.0877 -
(0.1305)
Continued. . .
140
Table 3.4 – continued from previous page
Dependent variable = log expenditures With In-
teractions
No
Interaction
Log Fruit and Veg. Acreage x Toilet/Handwash
Facilities
-0.1345 -
(0.1323)
Log Fruit and Veg. Acreage x Equipment Sanita-
tion
0.0337 -
-0.1081)
Log Fruit and Veg. Acreage x Building Sanitation 0.6863∗∗∗ -
-0.2062)
Log Fruit and Veg. Acreage x Sewage/Trash Dis-
posal
-0.1469 -
(0.1905)
Log Fruit and Veg. Acreage x Other Preventive
Actions
0.117 -
(0.1386)
Log Fruit and Veg. Acreage - 0.6319∗∗∗
(0.0542)
Panel F. Written Records (N = 296)
Log Fruit and Veg. Acreage - 0.2560∗∗∗
(0.0416)
Panel G. Soil Amendment Treatment (N = 169)
Log Fruit and Veg. Acreage x Single Soil Amend-
ment
1.5959 -
(1.078)
Log Fruit and Veg. Acreage x Multiple Soil
Amendments
0.8995∗∗∗ -
(0.1626)
Log Fruit and Veg. Acreage - 0.9180∗∗∗
(0.1587)
Panel H. Third Party Audit (N = 283)
Log Fruit and Veg. Acreage - 0.3798∗∗∗
(0.0986)
Note: Each regression is estimated by maximum likelihood using a double hurdle
specification. All regressions contain crop type (vegetables, berries, fruit and tree
nuts) and an indicator for sustainable farming practices as controls. Interaction
models for groups of practices also contain indicators for each type of practice within
the group as controls. Standard errors are reported in parentheses. Asterisk (∗),
double asterisk (∗∗), and triple asterisk (∗∗∗) indicate significance at the 10, 5 and 1
percent level, respectively.
141
Table 3.5: Estimated Impact of Sustainable Farming Practices on Food Safety Prac-
tice Expenditures
Practice Type With Interactions No Interaction
Sampling and Testing 0.4900 0.4638
(0.3559) (0.3922)
Field Inspection 0.4125 0.4685
(0.3652) (0.3754)
Harvest Container Sanitation 0.6788∗∗∗ 0.6253∗∗
(0.2477) (0.2448)
Washing Product - 0.7333∗∗
(0.3231)
Employee Sanitation 0.1592 0.2675
(0.2416) (0.2482)
Written Records - 0.4459∗∗
(0.1954)
Soil Amendment Treatment 0.2349 0.2322
(0.3965) (0.3848)
Third Party Audit - 0.1249
(0.5112)
Note: Each regression is estimated by maximum likelihood using a double hurdle
specification. All regressions contain crop type (vegetables, berries, fruit and
tree nuts) as controls. Interaction models for groups of practices also contain
indicators for each type of practice within the group as controls. Standard errors
are reported in parentheses. Asterisk (∗), double asterisk (∗∗), and triple asterisk
(∗∗∗) indicate significance at the 10, 5 and 1 percent level, respectively.
142
Table 3.6: Marginal Effects of Acreage, Marketing Channel, Farming Practices, and Crop Type on the Probability of Safety
Measure Use
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Washing
Product
Employee
Sanita-
tion &
Hygiene
Written
Records
Soil
Amend-
ment
Treat-
ment
Third-
Party
Audits
Log Fruit and Vegetable Acreage 0.0649∗∗∗ 0.0520∗∗∗ -0.0119 -0.0102 0.0107 0.0254∗ -0.0238 0.0455∗∗∗
(0.0146) (0.0148) (0.0088) (0.0126) (0.0096) (0.0135) (0.0234) (0.0094)
Contractual Obligation 0.1543∗∗ 0.2517∗∗∗ 0.0779∗ -0.0288 0.0276 0.3197∗∗∗ 0.1339 0.2141∗∗∗
(0.0749) (0.0674) (0.0444) (0.0607) (0.0493) (0.1077) (0.1318) (0.0287)
Wholesale / Other Sale Share 0.0007 -0.0007 -0.0003 0.001 -0.0011∗∗ 0.0000 -0.0021 0.0001
(0.0008) (0.0008) (0.0004) (0.0007) (0.0006) (0.0007) (0.0014) (0.0005)
Sustainable 0.0304 -0.0678 0.0071 0.2129∗∗∗ 0.0263 0.03 0.0217 -0.0479
(0.0601) (0.0638) (0.0393) (0.0603) (0.0401) (0.0471) (0.0824) (0.0503)
Berries -0.0234 -0.0433 0.0993∗∗∗ 0.0347 0.0711∗ -0.0162 -0.0353 -0.0341
(0.0518) (0.0556) (0.0368) (0.0473) (0.0365) (0.0414) (0.0798) (0.0354)
Fruit and Tree Nut 0.0462 -0.0098 -0.0181 0.0651 -0.0566 0.0531 0.0234 -0.0134
(0.0541) (0.0585) (0.037) (0.051) (0.0375) (0.0452) (0.0804) (0.0387)
Vegetables -0.1097 0.1421∗ -0.0205 0.3123∗∗∗ -0.1508∗∗ 0.0495 -0.0532 -0.0557
(0.0752) (0.0807) (0.0478) (0.0554) (0.0619) (0.062) (0.2069) (0.0458)
Note: Each regression is estimated by maximum likelihood using a double hurdle specification. All regressions contain crop type
(vegetables, berries, fruit and tree nuts) and an indicator for sustainable farming practices as controls. Models for groups of
practices also contain indicators for each type of practice within the group as controls. Standard errors (reported in parentheses)
were estimated using the delta method. Asterisk (∗), double asterisk (∗∗), and triple asterisk (∗∗∗) indicate significance at the 10,
5 and 1 percent level, respectively.
143
Table 3.7: Effects of Changes to Exemption Thresholds on Food Safety Cost Burden
Percentage for Each Food Safety Practice
A. Change to Tester-Hagan Exemption Revenue Threshold (Direct Sales Share =
50%)
Farming Practices Sustainable Conventional
Revenue $500,000 $750,000 $500,000 $750,000
Sampling and
Testing
Berries 0.1709 0.1577 0.1075 0.0992
Fruit and Tree Nuts 0.1149 0.1061 0.0723 0.0667
Vegetables 0.0981 0.0905 0.0617 0.0569
Field
Inspection
Berries 0.1205 0.1074 0.0754 0.0672
Fruit and Tree Nuts 0.1997 0.1779 0.125 0.1114
Vegetables 0.1868 0.1664 0.1169 0.1042
Harvest
Container
Sanitation
Berries 0.7155 0.6354 0.3829 0.34
Fruit and Tree Nuts 0.5367 0.4766 0.2872 0.255
Vegetables 0.7848 0.697 0.4199 0.3729
Washing
Product
Berries 1.5824 1.455 0.76 0.6988
Fruit and Tree Nuts 2.6088 2.3988 1.253 1.1522
Vegetables 3.5089 3.2265 1.6853 1.5497
Employee
Sanitation
and Hygiene
Berries 1.023 0.9286 0.7829 0.7106
Fruit and Tree Nuts 1.1086 1.0063 0.8484 0.7701
Vegetables 1.1754 1.0669 0.8995 0.8165
Written
Records
Berries 0.0048 0.0039 0.003 0.0025
Fruit and Tree Nuts 0.0032 0.0026 0.0021 0.0017
Vegetables 0.0033 0.0028 0.0021 0.0018
Soil
Amendment
Treatment
Berries 1.4652 1.4325 1.1615 1.1356
Fruit and Tree Nuts 1.956 1.9123 1.5506 1.516
Vegetables 2.7608 2.6992 2.1886 2.1398
Third Party
Audit
Berries 2.572 2.1869 2.2701 1.9302
Fruit and Tree Nuts 0.8542 0.7263 0.7539 0.641
Vegetables 1.0672 0.9074 0.9419 0.8009
B. Change to Tester-Hagan Exemption Direct Sales Share Threshold (Revenue =
$500,000)
Farming Practices Sustainable Conventional
Direct Sales Share 50% 75% 50% 75%
Sampling and
Testing
Berries 0.1709 0.1868 0.1075 0.1175
Fruit and Tree Nuts 0.1149 0.1257 0.0723 0.079
Vegetables 0.0981 0.1072 0.0617 0.0674
Continued. . .
144
Table 3.7 – continued from previous page
Field
Inspection
Berries 0.1205 0.137 0.0754 0.0857
Fruit and Tree Nuts 0.1997 0.2269 0.125 0.142
Vegetables 0.1868 0.2122 0.1169 0.1328
Harvest
Container
Sanitation
Berries 0.7155 0.8162 0.3829 0.4367
Fruit and Tree Nuts 0.5367 0.6122 0.2872 0.3276
Vegetables 0.7848 0.8952 0.4199 0.479
Washing
Product
Berries 1.5824 1.7367 0.76 0.8341
Fruit and Tree Nuts 2.6088 2.8632 1.253 1.3752
Vegetables 3.5089 3.8511 1.6853 1.8497
Employee
Sanitation
and Hygiene
Berries 1.023 1.1389 0.7829 0.8716
Fruit and Tree Nuts 1.1086 1.2342 0.8484 0.9445
Vegetables 1.1754 1.3086 0.8995 1.0014
Written
Records
Berries 0.0048 0.0059 0.003 0.0038
Fruit and Tree Nuts 0.0032 0.004 0.0021 0.0025
Vegetables 0.0033 0.0042 0.0021 0.0027
Soil
Amendment
Treatment
Berries 1.4652 1.5023 1.1615 1.1909
Fruit and Tree Nuts 1.956 2.0055 1.5506 1.5899
Vegetables 2.7608 2.8307 2.1886 2.2441
Third Party
Audit
Berries 2.572 3.0789 2.2701 2.7175
Fruit and Tree Nuts 0.8542 1.0225 0.7539 0.9025
Vegetables 1.0672 1.2775 0.9419 1.1275
C. Change to FSMA Revenue Exemption Threshold (Direct Sales Share = 49%)
Farming Practices Sustainable Conventional
Revenue $25,000 $50,000 $25,000 $50,000
Sampling and
Testing
Berries 0.3082 0.2687 0.1938 0.169
Fruit and Tree Nuts 0.2073 0.1807 0.1304 0.1136
Vegetables 0.1769 0.1542 0.1113 0.097
Field
Inspection
Berries 0.281 0.2307 0.1759 0.1444
Fruit and Tree Nuts 0.4655 0.3822 0.2914 0.2393
Vegetables 0.4354 0.3575 0.2725 0.2238
Harvest
Container
Sanitation
Berries 1.7108 1.3966 0.9155 0.7473
Fruit and Tree Nuts 1.2833 1.0476 0.6867 0.5606
Vegetables 1.8764 1.5318 1.0041 0.8197
Washing
Product
Berries 2.9304 2.5388 1.4075 1.2194
Fruit and Tree Nuts 4.8313 4.1857 2.3205 2.0104
Vegetables 6.4982 5.6298 3.1211 2.704
Continued. . .
145
Table 3.7 – continued from previous page
Employee
Sanitation
and Hygiene
Berries 2.0825 1.7649 1.5938 1.3507
Fruit and Tree Nuts 2.2568 1.9126 1.7271 1.4637
Vegetables 2.3928 2.0279 1.8312 1.5519
Written
Records
Berries 0.0198 0.0142 0.0127 0.0091
Fruit and Tree Nuts 0.0134 0.0096 0.0086 0.0061
Vegetables 0.0139 0.01 0.0089 0.0064
Soil
Amendment
Treatment
Berries 1.7292 1.6637 1.3708 1.3189
Fruit and Tree Nuts 2.3084 2.2211 1.83 1.7608
Vegetables 3.2583 3.135 2.583 2.4853
Third Party
Audit
Berries 8.4646 6.4148 7.4711 5.6619
Fruit and Tree Nuts 2.8112 2.1304 2.4812 1.8804
Vegetables 3.5121 2.6616 3.0999 2.3492
Note: Food safety cost burdens are expressed in percentage terms.
146
Appendix A: Appendix to Chapter 1
A.1 Solution for the Baseline Equilibrium
In the baseline equilibrium, Firm A solves the profit maximization problem
Π∗A = max
qiA, piA
NHxH
(
pHA − γq
2
HA
2
)
+NLxL
(
pLA − γq
2
LA
2
)
,
and Firm B solves the profit maximization problem
Π∗B = max
qiB , piB
NH(1− xH)
(
pHB − γq
2
HB
2
)
+NL(1− xL)
(
pLB − γq
2
LB
2
)
,
where xi takes the form specified in Equation 1.1. By solving the first order con-
ditions for the above problems, we obtain the equilibrium prices and qualities in
Equation 1.2 and firm profits are
Π∗A =
1
18
(
NHkH(2 + aH + bH)
2 +NLkL(2 + aL + bL)
2
)
Π∗B =
1
18
(
NHkH(−4 + aH + bH)2 +NLkL(−4 + aL + bL)2
)
147
A.2 Solution When Only Firm A Labels Products
If only Firm A chooses to obtain the non-GMO label, the new location xi of
the consumer in segment i indifferent between buying products of quality qiA and
qiB offered by each firm is
xi =
θi(qiA − qiB) + ki(ai + bi)− (piA − piB) + δi
2ki
if ai ≤ xi ≤ bi, (A.1)
and the respective profit maximization problems that each firm solves are
Π∗A = max
piA
NHxH
(
pHA − γq
2
HA
2
− α
)
+NLxL
(
pLA − γq
2
LA
2
− α
)
,
Π∗B = max
piB
NH(1− xH)
(
pHB − γq
2
HB
2
)
+NL(1− xL)
(
pLB − γq
2
LB
2
)
.
By solving the first order conditions for the new prices and substituting in the
equilibrium qualities from the baseline case, the new optimal prices are
p∗iA =
1
6
(2ki(2 + ai + bi) + 4α + 2δi) +
θ2i
2γ
,
p∗iB =
1
6
(2ki(4− ai − bi) + 2α− 2δi) + θ
2
i
2γ
.
From this result, it follows that Firm A captures additional market share and com-
mands higher price-cost margins than Firm B, provided that δi > α. Furthermore,
Firm B’s profit decreases relative to the baseline case prior to labeling.
148
A.3 Equilibrium When Both Firms Labels Products
If both firms chooses to obtain the non-GMO label, the indifference location
xi once again takes the form specified in Equation 1.1, and the respective profit
maximization problems that each firm solves are
Π∗A = max
piA
NHxH
(
pHA − γq
2
HA
2
− α
)
+NLxL
(
pLA − γq
2
LA
2
− α
)
,
Π∗B = max
piB
NH(1− xH)
(
pHB − γq
2
HB
2
− α
)
+NL(1− xL)
(
pLB − γq
2
LB
2
− α
)
.
By solving the first order conditions for the new prices and substituting in the
equilibrium qualities from the baseline case, the new optimal prices are
p∗iA =
1
6
(2ki(2 + ai + bi)) +
θ2i
2γ
+ α,
p∗iB =
1
6
(2ki(4− ai − bi)) + θ
2
i
2γ
+ α.
It is now trivial to show that neither firm gains any market share nor increases their
profits above the baseline equilibrium.
A.4 Equilibrium When Firms Develop New Products
In this scenario, when both market segments are fully served, the location xH
of the consumer in segment H indifferent between buying products of quality qHA
149
and qHB offered by each firm is
xH =
θH(1 + δH)(qHA − qHB) + kH(1 + δH)(aH + bH)− (pHA − pHB)
2kH(1 + δH)
.
The location xL is given by Equation 1.1. Firm A solves the profit maximization
problem
Π∗A = max
qHA, piA
NHxH
(
pHA − γq
2
HA
2
− α
)
+NLxL
(
pLA − γq
2
LA
2
− α
)
,
and Firm B solves the profit maximization problem
Π∗B = max
qHB , piB
NH(1− xH)
(
pHB − γq
2
HB
2
− α
)
+NL(1− xL)
(
pLB − γq
2
LB
2
− α
)
,
By solving the first order conditions for the above problems, we obtain the equilib-
rium prices and qualities in Equation 1.3 and firm profits are
Π∗A =
1
18
(
NHkH(1 + δH)(2 + aH + bH)
2 +NLkL(2 + aL + bL)
2
)
Π∗B =
1
18
(
NHkH(1 + δH)(−4 + aH + bH)2 +NLkL(−4 + aL + bL)2
)
.
A.5 Price Premium Estimation with Full Sample
As a robustness check, the regression specification in Equation 1.5 was es-
timated using an unrestricted sample of Nielsen data spanning 2009 to 2014. In
addition to the observations included in the restricted sample, this sample includes
150
products that were non-GMO certified with less than 6 months of sales data prior
to being certified and/or 12 months of sales after certification, and products that
never obtained non-GMO certification. Table A.1 presents results from this sample
with a progression of fixed effects identical to those presented in the main paper.
Across all three specifications, the coefficient estimates for the post-certification
treatment indicators are very small and not statistically significantly different from
zero, consistent with the results presented in the main paper using the restricted
sample.
151
Table A.1: Price Premium Regressions - Unrestricted Sample
I II III
Pre-Cert. 6-12 Mos. −0.007∗ −0.005 −0.006
(0.003) (0.004) (0.005)
Pre-Cert. 0-6 Mos. −0.002 0.001 0.000
(0.003) (0.005) (0.006)
Post-Cert. 0-6 Mos −0.002 −0.001 0.001
(0.004) (0.007) (0.007)
Post-Cert. 6-12 Mos. −0.001 −0.007 −0.003
(0.004) (0.007) (0.008)
Post-Cert. 12-24 Mos. −0.005 −0.009 −0.002
(0.004) (0.008) (0.009)
Post-Cert. 24+ Mos. 0.007 0.000 0.012
(0.005) (0.011) (0.012)
UPC FEs Yes Yes Yes
Week FEs Yes Yes Yes
× Category Yes No Yes
× Manufacturer No Yes Yes
Adj. R2 0.969 0.973 0.973
Num. obs. 10366743 10366743 10366743
Note: Each column represents a separate regression. Standard
errors are clustered at the product level in parentheses: ∗∗∗p <
0.001, ∗∗p < 0.01, ∗p < 0.05, ·p < 0.1.
152
Appendix B: Appendix to Chapter 2
B.1 Additional Tables
Tables B.1, B.2, B.3, B.4, and B.5 present median own- and cross-price elas-
ticities for each of the 50 RTE cereal brands, across all 5,988 markets.
153
Table B.1: Median Own- and Cross-Price Elasticities - Brands 1-10
Brand 1 2 3 4 5 6 7 8 9 10
1 4.131 -0.002 -0.000 -0.000 -0.000 -0.001 -0.000 -0.001 -0.019 -0.000
2 -0.000 -3.341 0.159 0.035 0.015 0.268 0.017 0.118 0.022 0.056
3 -0.000 0.189 -3.792 0.043 0.017 0.326 0.020 0.142 0.029 0.066
4 -0.000 0.166 0.165 -3.573 0.015 0.279 0.017 0.123 0.023 0.057
5 -0.000 0.144 0.138 0.031 -3.153 0.239 0.015 0.103 0.017 0.048
6 -0.000 0.185 0.186 0.042 0.017 -3.583 0.019 0.140 0.028 0.064
7 -0.000 0.154 0.152 0.034 0.014 0.256 -3.403 0.112 0.018 0.052
8 -0.000 0.165 0.164 0.037 0.015 0.282 0.017 -3.450 0.024 0.057
9 -0.000 0.052 0.057 0.012 0.004 0.097 0.005 0.040 -1.173 0.020
10 -0.000 0.190 0.189 0.043 0.017 0.319 0.020 0.140 0.028 -3.911
11 -0.000 0.106 0.105 0.024 0.010 0.175 0.011 0.076 0.005 0.036
12 0.000 0.003 0.003 0.001 0.000 0.005 0.000 0.002 0.000 0.001
13 0.000 0.007 0.008 0.002 0.001 0.013 0.001 0.006 0.001 0.003
14 -0.000 0.119 0.114 0.026 0.011 0.194 0.012 0.084 0.008 0.041
15 -0.000 0.109 0.110 0.024 0.010 0.186 0.009 0.077 0.003 0.039
16 -0.000 0.117 0.118 0.025 0.010 0.198 0.010 0.081 0.004 0.040
17 -0.000 0.080 0.076 0.018 0.007 0.131 0.007 0.057 -0.005 0.028
18 0.000 0.014 0.019 0.003 0.001 0.030 0.001 0.011 0.001 0.006
19 -0.000 0.077 0.104 0.018 0.006 0.155 0.005 0.056 0.002 0.029
20 -0.000 0.128 0.135 0.026 0.014 0.210 0.006 0.088 0.005 0.040
21 -0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
22 -0.000 0.148 0.150 0.033 0.014 0.254 0.016 0.112 0.018 0.053
23 -0.000 0.201 0.203 0.045 0.018 0.342 0.021 0.147 0.029 0.071
24 -0.000 0.155 0.154 0.035 0.014 0.261 0.016 0.115 0.019 0.054
25 0.000 0.221 0.231 0.050 0.020 0.381 0.023 0.163 0.032 0.080
26 0.000 0.237 0.249 0.054 0.021 0.415 0.024 0.175 0.034 0.087
27 -0.000 0.194 0.195 0.044 0.018 0.326 0.020 0.143 0.029 0.069
28 0.000 0.250 0.277 0.059 0.022 0.456 0.025 0.189 0.032 0.096
29 0.000 0.231 0.238 0.052 0.020 0.398 0.024 0.169 0.033 0.084
30 -0.000 0.139 0.134 0.030 0.013 0.226 0.014 0.100 0.015 0.047
31 -0.000 0.120 0.113 0.026 0.011 0.193 0.012 0.085 0.009 0.041
32 -0.000 0.100 0.099 0.022 0.009 0.165 0.010 0.070 0.002 0.035
33 -0.000 0.142 0.137 0.031 0.013 0.230 0.014 0.101 0.015 0.048
34 0.000 0.252 0.287 0.059 0.021 0.468 0.025 0.193 0.030 0.099
35 0.000 0.281 0.278 0.062 0.021 0.486 0.030 0.208 0.033 0.114
36 -0.000 0.001 0.001 0.000 0.000 0.001 0.000 0.001 0.000 0.000
37 -0.000 0.001 0.001 0.000 0.000 0.001 0.000 0.001 0.000 0.000
38 0.000 0.001 0.001 0.000 0.000 0.002 0.000 0.001 0.000 0.000
39 0.000 0.212 0.242 0.047 0.018 0.395 0.019 0.154 0.024 0.080
40 -0.000 0.215 0.217 0.048 0.019 0.365 0.022 0.158 0.031 0.077
41 0.000 0.255 0.288 0.060 0.022 0.471 0.025 0.193 0.031 0.100
42 -0.000 0.181 0.178 0.040 0.017 0.300 0.019 0.130 0.026 0.062
43 -0.000 0.183 0.182 0.039 0.016 0.306 0.018 0.127 0.022 0.063
44 -0.000 0.189 0.195 0.040 0.017 0.321 0.018 0.132 0.023 0.066
45 0.000 0.216 0.223 0.049 0.019 0.369 0.022 0.160 0.031 0.077
46 0.000 0.220 0.227 0.050 0.020 0.374 0.023 0.161 0.032 0.079
47 -0.000 0.180 0.179 0.040 0.016 0.304 0.018 0.130 0.025 0.063
48 -0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
49 0.000 0.002 0.002 0.000 0.000 0.003 0.000 0.001 0.000 0.001
50 0.000 0.003 0.003 0.001 0.000 0.005 0.000 0.002 0.000 0.001
Note: Each cell entry in row i, column j represents the median elasticity of the market share of
brand i with respect to the price of brand j, over all 5,988 markets.
154
Table B.2: Median Own- and Cross-Price Elasticities - Brands 11-20
Brand 11 12 13 14 15 16 17 18 19 20
1 -0.002 0.000 0.000 -0.001 -0.003 -0.005 -0.004 0.000 -0.000 -0.004
2 0.017 0.000 0.000 0.010 0.010 0.023 0.010 0.000 0.003 0.007
3 0.019 0.000 0.001 0.011 0.014 0.029 0.013 0.000 0.005 0.011
4 0.017 0.000 0.000 0.010 0.011 0.024 0.011 0.000 0.003 0.008
5 0.015 0.000 0.000 0.008 0.009 0.019 0.009 0.000 0.001 0.006
6 0.019 0.000 0.001 0.011 0.013 0.027 0.012 0.000 0.004 0.009
7 0.016 0.000 0.000 0.009 0.010 0.022 0.010 0.000 0.003 0.008
8 0.017 0.000 0.000 0.010 0.011 0.023 0.010 0.000 0.003 0.008
9 0.002 0.000 0.000 0.002 0.001 0.002 -0.002 0.000 0.000 0.001
10 0.019 0.000 0.001 0.011 0.014 0.028 0.013 0.000 0.004 0.009
11 -2.446 0.000 0.000 0.006 0.006 0.012 0.005 0.000 0.001 0.004
12 0.000 -3.424 0.061 0.000 0.001 0.002 0.000 0.018 0.001 0.005
13 0.001 0.030 -3.442 0.000 0.004 0.006 0.000 0.010 0.007 0.005
14 0.011 0.000 0.000 -2.718 0.008 0.016 0.007 0.000 0.001 0.005
15 0.009 0.000 0.002 0.006 -3.170 0.032 0.004 0.001 0.002 0.006
16 0.009 0.000 0.002 0.007 0.017 -3.255 0.005 0.001 0.002 0.007
17 0.006 0.000 0.000 0.004 0.003 0.007 -1.773 0.000 0.000 0.002
18 0.001 0.033 0.036 0.001 0.004 0.009 0.000 -3.272 0.007 0.009
19 0.005 0.025 0.022 0.003 0.005 0.015 0.002 0.005 -3.404 0.010
20 0.009 0.031 0.042 0.006 0.007 0.028 0.005 0.002 0.007 -3.084
21 0.000 0.007 0.003 0.000 0.000 0.000 0.000 0.000 -0.000 -0.004
22 0.016 0.000 0.000 0.009 0.010 0.021 0.009 0.000 0.003 0.008
23 0.021 0.000 0.001 0.012 0.015 0.030 0.013 0.000 0.005 0.011
24 0.016 0.000 0.000 0.009 0.010 0.022 0.009 0.000 0.003 0.008
25 0.022 0.000 0.001 0.013 0.017 0.034 0.014 0.001 0.006 0.013
26 0.022 0.000 0.001 0.014 0.018 0.037 0.015 0.001 0.006 0.013
27 0.020 0.000 0.001 0.012 0.014 0.029 0.013 0.000 0.004 0.010
28 0.022 0.000 0.001 0.014 0.019 0.039 0.015 0.001 0.007 0.015
29 0.022 0.000 0.001 0.014 0.017 0.035 0.015 0.001 0.006 0.013
30 0.015 0.000 0.000 0.008 0.009 0.019 0.008 0.000 0.002 0.007
31 0.012 0.000 0.000 0.007 0.007 0.015 0.006 0.000 0.001 0.005
32 0.009 0.000 0.000 0.005 0.005 0.012 0.004 0.000 0.001 0.005
33 0.015 0.000 0.000 0.008 0.009 0.020 0.009 0.000 0.003 0.008
34 0.022 0.000 0.001 0.014 0.019 0.039 0.014 0.001 0.007 0.015
35 0.028 0.000 0.001 0.016 0.023 0.037 0.016 0.000 0.001 0.019
36 0.000 0.015 0.009 0.000 0.000 0.000 0.000 0.001 0.000 0.000
37 0.000 0.016 0.009 0.000 0.000 0.000 0.000 0.001 0.000 0.000
38 0.000 0.019 0.012 0.000 0.000 0.000 0.000 0.001 0.000 0.000
39 0.017 0.001 0.005 0.012 0.026 0.050 0.012 0.002 0.007 0.015
40 0.021 0.000 0.001 0.013 0.016 0.033 0.014 0.000 0.005 0.012
41 0.022 0.000 0.001 0.014 0.019 0.039 0.014 0.001 0.007 0.015
42 0.018 0.000 0.001 0.011 0.013 0.026 0.012 0.000 0.004 0.010
43 0.017 0.000 0.001 0.011 0.019 0.037 0.012 0.001 0.005 0.012
44 0.017 0.000 0.002 0.011 0.020 0.040 0.012 0.001 0.005 0.013
45 0.021 0.000 0.001 0.013 0.016 0.034 0.014 0.001 0.006 0.014
46 0.021 0.000 0.001 0.013 0.016 0.034 0.014 0.001 0.006 0.014
47 0.018 0.000 0.001 0.011 0.013 0.027 0.012 0.000 0.003 0.011
48 0.000 0.003 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
49 0.000 0.028 0.035 0.000 0.001 0.001 0.000 0.009 0.000 0.000
50 0.000 0.032 0.060 0.000 0.002 0.003 0.000 0.016 0.000 0.004
Note: Each cell entry in row i, column j represents the median elasticity of the market share of
brand i with respect to the price of brand j, over all 5,988 markets.
155
Table B.3: Median Own- and Cross-Price Elasticities - Brands 21-30
Brand 21 22 23 24 25 26 27 28 29 30
1 -0.000 -0.001 -0.000 -0.001 0.000 0.000 -0.000 0.000 0.000 -0.002
2 0.000 0.049 0.062 0.084 0.216 0.205 0.027 0.113 0.064 0.062
3 0.000 0.061 0.075 0.102 0.272 0.259 0.033 0.152 0.079 0.073
4 0.000 0.052 0.064 0.088 0.222 0.214 0.028 0.122 0.065 0.064
5 0.000 0.042 0.053 0.073 0.183 0.175 0.024 0.095 0.055 0.054
6 0.000 0.059 0.073 0.100 0.263 0.251 0.033 0.145 0.076 0.071
7 0.000 0.048 0.058 0.081 0.204 0.191 0.025 0.106 0.060 0.060
8 0.000 0.052 0.063 0.089 0.225 0.214 0.028 0.120 0.065 0.063
9 0.000 0.014 0.021 0.025 0.072 0.070 0.009 0.035 0.022 0.016
10 0.000 0.060 0.075 0.101 0.267 0.258 0.033 0.147 0.079 0.073
11 0.000 0.033 0.040 0.055 0.137 0.127 0.017 0.066 0.040 0.042
12 0.002 0.001 0.001 0.002 0.004 0.004 0.001 0.002 0.001 0.001
13 0.000 0.002 0.003 0.004 0.012 0.011 0.002 0.006 0.003 0.003
14 0.000 0.036 0.044 0.060 0.151 0.145 0.020 0.077 0.045 0.046
15 0.000 0.033 0.044 0.056 0.154 0.152 0.018 0.083 0.045 0.039
16 0.000 0.035 0.047 0.059 0.162 0.161 0.019 0.088 0.049 0.042
17 0.000 0.023 0.031 0.040 0.101 0.097 0.014 0.050 0.031 0.028
18 0.000 0.005 0.009 0.008 0.036 0.034 0.003 0.020 0.010 0.005
19 -0.000 0.027 0.040 0.044 0.147 0.146 0.012 0.081 0.043 0.030
20 -0.001 0.040 0.054 0.064 0.191 0.192 0.017 0.100 0.053 0.054
21 2.063 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
22 0.000 -3.315 0.060 0.087 0.212 0.195 0.026 0.109 0.060 0.061
23 0.000 0.065 -4.085 0.111 0.306 0.285 0.035 0.167 0.087 0.080
24 0.000 0.053 0.062 -3.326 0.218 0.199 0.027 0.113 0.061 0.063
25 0.000 0.072 0.095 0.123 -4.011 0.342 0.041 0.210 0.101 0.087
26 0.000 0.076 0.104 0.130 0.394 -4.132 0.045 0.241 0.114 0.090
27 0.000 0.062 0.077 0.105 0.278 0.271 -4.023 0.158 0.081 0.075
28 0.000 0.081 0.117 0.139 0.464 0.461 0.050 -4.210 0.133 0.094
29 0.000 0.074 0.099 0.125 0.371 0.362 0.043 0.221 -4.402 0.089
30 0.000 0.045 0.054 0.076 0.187 0.171 0.023 0.094 0.053 -3.040
31 0.000 0.036 0.045 0.062 0.150 0.142 0.020 0.076 0.044 0.045
32 0.000 0.030 0.037 0.051 0.130 0.121 0.017 0.064 0.038 0.037
33 0.000 0.044 0.053 0.074 0.186 0.174 0.024 0.095 0.054 0.055
34 0.000 0.081 0.123 0.138 0.488 0.490 0.052 0.319 0.140 0.093
35 0.000 0.089 0.129 0.148 0.485 0.498 0.056 0.314 0.149 0.104
36 0.000 0.000 0.000 0.000 0.001 0.001 0.000 0.001 0.000 0.000
37 0.000 0.000 0.000 0.000 0.001 0.001 0.000 0.001 0.000 0.000
38 0.000 0.000 0.000 0.001 0.002 0.002 0.000 0.001 0.001 0.000
39 0.000 0.068 0.102 0.116 0.407 0.416 0.038 0.262 0.117 0.079
40 0.000 0.068 0.090 0.116 0.329 0.318 0.039 0.189 0.094 0.083
41 0.000 0.082 0.123 0.139 0.486 0.492 0.052 0.322 0.141 0.094
42 0.000 0.056 0.071 0.096 0.249 0.236 0.030 0.134 0.073 0.070
43 0.000 0.057 0.074 0.096 0.266 0.267 0.031 0.154 0.080 0.070
44 0.000 0.059 0.080 0.100 0.293 0.291 0.032 0.171 0.086 0.072
45 0.000 0.071 0.092 0.118 0.337 0.335 0.040 0.200 0.098 0.084
46 0.000 0.070 0.093 0.118 0.344 0.342 0.040 0.203 0.101 0.085
47 0.000 0.056 0.071 0.096 0.252 0.239 0.031 0.138 0.074 0.069
48 -0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
49 0.000 0.001 0.001 0.001 0.002 0.002 0.000 0.001 0.001 0.001
50 0.001 0.001 0.001 0.001 0.004 0.003 0.001 0.002 0.001 0.001
Note: Each cell entry in row i, column j represents the median elasticity of the market share of
brand i with respect to the price of brand j, over all 5,988 markets.
156
Table B.4: Median Own- and Cross-Price Elasticities - Brands 31-40
Brand 31 32 33 34 35 36 37 38 39 40
1 -0.004 -0.007 -0.001 0.000 0.000 -0.000 -0.000 0.000 0.000 -0.000
2 0.037 0.041 0.023 0.017 0.002 0.000 0.000 0.000 0.026 0.214
3 0.042 0.049 0.027 0.026 0.002 0.000 0.000 0.000 0.036 0.261
4 0.037 0.042 0.024 0.020 0.002 0.000 0.000 0.000 0.027 0.217
5 0.032 0.036 0.020 0.014 0.001 0.000 0.000 0.000 0.020 0.179
6 0.041 0.047 0.026 0.025 0.002 0.000 0.000 0.000 0.033 0.252
7 0.034 0.039 0.022 0.015 0.001 0.000 0.000 0.000 0.023 0.195
8 0.036 0.041 0.023 0.020 0.002 0.000 0.000 0.000 0.027 0.219
9 0.007 0.003 0.006 0.004 0.000 0.000 0.000 0.000 0.007 0.074
10 0.042 0.048 0.027 0.025 0.002 0.000 0.000 0.000 0.034 0.258
11 0.022 0.024 0.015 0.009 0.001 0.000 0.000 0.000 0.013 0.133
12 0.001 0.001 0.000 0.000 0.000 0.006 0.005 0.003 0.002 0.004
13 0.001 0.001 0.001 0.001 0.000 0.001 0.001 0.001 0.010 0.011
14 0.026 0.027 0.017 0.011 0.001 0.000 0.000 0.000 0.017 0.152
15 0.022 0.024 0.015 0.009 0.001 0.000 0.000 0.000 0.029 0.155
16 0.024 0.026 0.016 0.010 0.001 0.000 0.000 0.000 0.030 0.165
17 0.015 0.013 0.011 0.007 0.001 0.000 0.000 0.000 0.010 0.102
18 0.002 0.002 0.002 0.002 0.000 0.000 0.000 0.000 0.016 0.032
19 0.012 0.017 0.013 0.006 0.000 0.000 0.000 0.000 0.022 0.150
20 0.029 0.040 0.020 0.011 0.001 0.000 0.000 0.000 0.029 0.193
21 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
22 0.033 0.037 0.021 0.018 0.002 0.000 0.000 0.000 0.025 0.201
23 0.044 0.052 0.029 0.030 0.002 0.000 0.000 0.000 0.040 0.289
24 0.034 0.039 0.022 0.019 0.002 0.000 0.000 0.000 0.025 0.205
25 0.048 0.056 0.031 0.037 0.003 0.000 0.000 0.000 0.051 0.333
26 0.051 0.058 0.033 0.044 0.003 0.000 0.000 0.000 0.057 0.367
27 0.044 0.051 0.028 0.027 0.002 0.000 0.000 0.000 0.035 0.265
28 0.052 0.059 0.034 0.055 0.004 0.000 0.000 0.000 0.072 0.418
29 0.050 0.057 0.032 0.040 0.003 0.000 0.000 0.000 0.052 0.345
30 0.030 0.034 0.020 0.014 0.001 0.000 0.000 0.000 0.021 0.182
31 -2.649 0.030 0.017 0.011 0.001 0.000 0.000 0.000 0.016 0.152
32 0.022 -2.272 0.015 0.008 0.001 0.000 0.000 0.000 0.014 0.129
33 0.033 0.038 -3.140 0.014 0.001 0.000 0.000 0.000 0.022 0.181
34 0.051 0.057 0.034 -4.251 0.005 0.000 0.000 0.000 0.078 0.444
35 0.051 0.051 0.032 0.085 -4.251 0.000 0.000 0.000 0.078 0.450
36 0.000 0.000 0.000 0.000 0.000 -1.434 0.001 0.001 0.000 0.001
37 0.000 0.000 0.000 0.000 0.000 0.002 -1.521 0.001 0.000 0.001
38 0.000 0.000 0.000 0.000 0.000 0.003 0.002 -1.877 0.001 0.002
39 0.044 0.051 0.029 0.038 0.003 0.000 0.000 0.000 -4.247 0.372
40 0.047 0.054 0.030 0.034 0.003 0.000 0.000 0.000 0.045 -4.007
41 0.052 0.057 0.034 0.061 0.005 0.000 0.000 0.000 0.077 0.443
42 0.041 0.048 0.026 0.022 0.002 0.000 0.000 0.000 0.031 0.245
43 0.041 0.048 0.027 0.023 0.002 0.000 0.000 0.000 0.046 0.265
44 0.041 0.050 0.027 0.025 0.002 0.000 0.000 0.000 0.054 0.285
45 0.047 0.055 0.031 0.036 0.003 0.000 0.000 0.000 0.048 0.322
46 0.049 0.056 0.031 0.035 0.003 0.000 0.000 0.000 0.049 0.330
47 0.040 0.047 0.025 0.023 0.002 0.000 0.000 0.000 0.033 0.249
48 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
49 0.000 0.000 0.000 0.000 0.000 0.007 0.006 0.003 0.001 0.002
50 0.001 0.000 0.000 0.000 0.000 0.006 0.005 0.003 0.004 0.003
Note: Each cell entry in row i, column j represents the median elasticity of the market share of
brand i with respect to the price of brand j, over all 5,988 markets.
157
Table B.5: Median Own- and Cross-Price Elasticities - Brands 41-50
Brand 41 42 43 44 45 46 47 48 49 50
1 0.000 -0.000 -0.000 -0.000 0.000 0.000 -0.000 -0.000 0.000 0.000
2 0.026 0.017 0.010 0.011 0.044 0.063 0.037 0.000 0.000 0.000
3 0.036 0.020 0.012 0.014 0.056 0.081 0.043 0.000 0.000 0.000
4 0.028 0.017 0.010 0.011 0.046 0.066 0.037 0.000 0.000 0.000
5 0.020 0.014 0.008 0.009 0.037 0.052 0.031 0.000 0.000 0.000
6 0.034 0.019 0.011 0.013 0.054 0.077 0.043 0.000 0.000 0.000
7 0.023 0.016 0.009 0.010 0.041 0.058 0.032 0.000 0.000 0.000
8 0.028 0.017 0.010 0.011 0.046 0.065 0.037 0.000 0.000 0.000
9 0.007 0.005 0.002 0.002 0.014 0.020 0.009 0.000 0.000 0.000
10 0.035 0.020 0.012 0.013 0.055 0.079 0.044 0.000 0.000 0.000
11 0.014 0.010 0.006 0.006 0.026 0.037 0.021 0.000 0.000 0.000
12 0.000 0.000 0.000 0.000 0.001 0.001 0.001 0.000 0.005 0.027
13 0.001 0.001 0.001 0.001 0.002 0.003 0.002 0.000 0.002 0.026
14 0.017 0.012 0.007 0.007 0.030 0.043 0.026 0.000 0.000 0.000
15 0.018 0.011 0.009 0.011 0.030 0.043 0.020 0.000 0.000 0.000
16 0.019 0.012 0.009 0.011 0.032 0.047 0.021 0.000 0.000 0.000
17 0.010 0.008 0.004 0.004 0.020 0.029 0.016 0.000 0.000 0.000
18 0.004 0.001 0.002 0.003 0.007 0.010 0.002 0.000 0.002 0.028
19 0.021 0.008 0.006 0.010 0.029 0.042 0.006 0.000 0.000 0.015
20 0.030 0.024 0.009 0.021 0.045 0.064 0.002 0.000 0.000 0.018
21 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -0.000 0.000 0.007
22 0.026 0.015 0.009 0.010 0.043 0.060 0.033 0.000 0.000 0.000
23 0.042 0.021 0.013 0.015 0.062 0.089 0.047 0.000 0.000 0.000
24 0.026 0.016 0.010 0.011 0.044 0.061 0.034 0.000 0.000 0.000
25 0.051 0.024 0.015 0.018 0.074 0.105 0.052 0.000 0.000 0.000
26 0.060 0.025 0.016 0.020 0.082 0.116 0.057 0.000 0.000 0.000
27 0.037 0.020 0.012 0.014 0.057 0.080 0.045 0.000 0.000 0.000
28 0.075 0.027 0.018 0.023 0.095 0.136 0.062 0.000 0.000 0.000
29 0.055 0.024 0.015 0.019 0.078 0.112 0.055 0.000 0.000 0.000
30 0.022 0.014 0.008 0.009 0.039 0.054 0.029 0.000 0.000 0.000
31 0.016 0.012 0.006 0.007 0.031 0.043 0.025 0.000 0.000 0.000
32 0.013 0.010 0.006 0.006 0.026 0.037 0.020 0.000 0.000 0.000
33 0.021 0.014 0.008 0.009 0.038 0.054 0.029 0.000 0.000 0.000
34 0.082 0.028 0.019 0.024 0.100 0.143 0.063 0.000 0.000 0.000
35 0.086 0.027 0.019 0.016 0.107 0.145 0.073 0.000 0.000 0.000
36 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.012
37 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.012
38 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.002 0.015
39 0.064 0.023 0.020 0.026 0.078 0.117 0.049 0.000 0.000 0.001
40 0.046 0.023 0.014 0.017 0.068 0.098 0.051 0.000 0.000 0.000
41 -4.320 0.028 0.019 0.024 0.100 0.143 0.063 0.000 0.000 0.000
42 0.031 -3.812 0.011 0.013 0.051 0.074 0.041 0.000 0.000 0.000
43 0.035 0.020 -4.083 0.018 0.055 0.079 0.041 0.000 0.000 0.000
44 0.040 0.021 0.016 -4.187 0.059 0.086 0.041 0.000 0.000 0.000
45 0.050 0.023 0.014 0.018 -4.281 0.102 0.050 0.000 0.000 0.000
46 0.050 0.024 0.015 0.018 0.073 -4.273 0.051 0.000 0.000 0.000
47 0.032 0.019 0.011 0.013 0.052 0.076 -3.802 0.000 0.000 0.000
48 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.582 0.000 0.002
49 0.000 0.000 0.000 0.000 0.001 0.001 0.000 0.000 -2.778 0.023
50 0.000 0.000 0.000 0.000 0.001 0.001 0.001 0.000 0.005 -3.484
Note: Each cell entry in row i, column j represents the median elasticity of the market share of
brand i with respect to the price of brand j, over all 5,988 markets.
158
B.2 Computational Details
My computation approach for estimating the random-coefficients logit demand
model follows Nevo (2000b) very closely; however, I incorporate several modest
improvements that significantly speed up model convergence without negatively
impacting the quality or robustness of the approximation. I start by porting all of
the original MATLAB code to the R programming language (R Core Team 2016). In
light of the findings in Dube´ et al. (2012), I use an inner-loop tolerance of εin = 10
−14
and outer-loop tolerance of εout = 10
−8, significantly increasing the CPU time for
the BLP contraction mapping, since it converges linearly. To speed up convergence
of the contraction mapping without sacrificing numerical accuracy, I instead use the
squared extrapolation method (SQUAREM) algorithm for accelerating fixed-point
iterations, which produces faster and more robust convergence than the traditional
BLP contraction mapping (Reynaerts et al. 2012; Varadhan 2010). To ensure that
the model converges to a global minimum, I estimated the model using ten different
sets of starting values for the θ2 parameters, each set randomly drawn from the
standard normal distribution.
Additionally, because the market-level calculations are independent of one
another (each t represents a separate DMA-Month), an opportunity exists to drasti-
cally improve computational performance by parallelizing computation of the mean
utility (δjt) as well as the Jacobian of the implicit function that defines the mean
utility. Computing these values in parallel also requires parallelization of the func-
tions that compute the individual probabilities of choosing each brand (sijt), the
159
market shares for each brand (sjt), the heteroskedastic nonlinear component of util-
ity (µijt), and the BLP contraction mapping. In my case, I set up a socket cluster
with 64 nodes on a Windows 7 Server, and the work is distributed to each node on
a market-by-market basis using the doParallel package in R (Revolution Analytics
and Weston 2015). Section B.2.1 provides R code for initialization of the parallel
cluster, Section B.2.2 provides modified functions to accommodate parallelization
of the mean utility calculation, and Section B.2.3 provides modified functions for
parallelization of the Jacobian.
B.2.1 Parallel Cluster Inititalization
1 s e t u p c l u s t e r <− f unc t i on ( nc=32) {
2 c l <<− makeCluster ( nc )
3 r e g i s t e r D o P a r a l l e l ( c l )
4 ## Find out the names o f the loaded packages
5 loaded . package . names <− c ( s e s s i o n I n f o ( ) $basePkgs , names ( s e s s i o n I n f o
( ) $ otherPkgs ) )
6
7 t h i s . env <− environment ( )
8 whi le ( i d e n t i c a l ( t h i s . env , g loba l env ( ) ) == FALSE ) {
9 c lu s t e rExpor t ( c l ,
10 l s ( a l l . names=TRUE, env=t h i s . env ) ,
11 env i r=t h i s . env )
12 t h i s . env <− parent . env ( environment ( ) )
13 }
14 ## repeat f o r the g l o b a l environment
15 c lu s t e rExpor t ( c l ,
16 l s ( a l l . names=TRUE, env=globa l env ( ) ) ,
17 env i r=g loba l env ( ) )
18
19 parLapply ( c l , 1 : l ength ( c l ) , f unc t i on ( xx ) {
20 l app ly ( loaded . package . names , f unc t i on ( yy ) {
21 r e q u i r e ( yy , cha rac t e r . only=TRUE) })
22 })
23 }
B.2.2 Parallelized Mean Utility
1 parmeanval <− f unc t i on ( theta2 , newvars=c ( ” mvalold ” ) ,n . mkt=nmkt) {
160
2 ## parmeanval s e r v e s as a he lpe r func t i on f o r p a r a l l e l i z i n g the
meanval func t i on below , which now i n c l u d e s a index f o r market
3 c lu s t e rExpor t ( c l , v a r l i s t=newvars )
4 f<−f o r each ( i =1:n . mkt , . combine=c ) %dopar% {
5 meanval ( theta2 , id=(cdid==i ) )
6 }
7 mvalold<<−f
8 re turn ( f )
9 }
10
11 meanval<−f unc t i on ( theta2 , id , maxiter =10000 , t o l=1e−14){
12 theta2w<−matrix (0 , r theta , c theta )
13 theta2w [ which ( theta2w st !=0) ]<−theta2
14 expmu<−exp ( mufunc ( x2 , theta2w , id ) )
15 ## Don ’ t s t a r t with bad va lue s o f d e l t a vec to r ##
16 i f ( any ( i s . na ( mvalold [ id ] ) | i s . i n f i n i t e ( mvalold [ id ] ) ) ) {
17 d e l t a i d <− m v a l l o g i t [ id ]
18 } e l s e {
19 d e l t a i d <− mvalold [ id ]
20 }
21 fp<−squarem ( par=d e l t a i d , expmu=expmu , s=s j t [ id ] , f i x p t f n=blp . inner ,
c o n t r o l=l i s t ( t o l=to l , maxiter=maxiter ) )
22 re turn ( fp $par )
23 }
24
25 mufunc<−f unc t i on ( x2 , theta2w , id ) {
26 #This func t i on computes the non−l i n e a r part o f the u t i l i t y
27 x2<−x2 [ id , ]
28 n<−nrow ( x2 )
29 k<−nco l ( x2 )
30 j<−nco l ( theta2w )−1
31 mu<−matrix (0 , n , ns )
32 f o r ( i in 1 : ns ) {
33 v i<−v f u l l [ id , seq ( i , k∗ns , ns ) ]
34 d i<−d f u l l [ id , seq ( i , j ∗ns , ns ) ] # 2256 by 4 take (1 ,21 ,41 , 61 )
(2 , 22 ,42 , 62 ) . . ( income , income ˆ2 , age , c h i l d )
35 mu[ , i ]<−( ( x2∗ v i ) %∗% ( theta2w [ , 1 ] ) )+(x2 ∗( d i %∗% t ( theta2w [ , ( 2 : ( j
+1) ) ] ) ) ) %∗% rep (1 , k ) #theta2w [ ,1 ]= sigma and theta2w [ , ( 2 : ( j +1) )
]= pi
36 }
37 re turn (mu)
38 }
39
40 blp . inner<−f unc t i on ( de l ta , expmu , s ) {
41 d e l t a <− d e l t a + log ( s ) − l og ( sha r e f cn ( de l ta , expmu) [ [ 2 ] ] )
42 re turn ( de l t a )
43 }
44
45 sha r e f cn<−f unc t i on ( mval , expmu) {
46 # This func t i on computes the ” i n d i v i d u a l ” p r o b a b i l i t i e s o f choos ing
each brand
47 # AND the market share s f o r each product . The output i s a l i s t with
both va lues
48 eg<−expmu∗ kronecker ( t ( rep (1 , ns ) ) , exp ( mval ) )
161
49 sum1<−rbind ( colSums ( eg ) )
50 denom<−1/(1+sum1)
51 denom<−denom [ rep (1 , nrow ( eg ) ) , ]
52 ind . cho i c e<−eg∗denom
53 cho i c e<−rowSums( ind . cho i c e ) /ns
54 re turn ( l i s t ( ind . cho ice , cho i c e ) )
55 }
B.2.3 Parallelized Jacobian
1 par jacob <− f unc t i on ( mval , theta2 , n . mkt=nmkt) {
2 ## parjacob s e r v e s as a he lpe r func t i on f o r p a r a l l e l i z i n g the
jacob ian func t i on below , which now i n c l u d e s a index f o r market
3 f<−f o r each ( i =1:n . mkt , . combine=rbind ) %dopar% {
4 jacob ( mval , theta2 , id=(cdid==i ) )
5 }
6 re turn ( f )
7 }
8
9 jacob<−f unc t i on ( mval , theta2 , id ) {
10 # This func t i on computes the Jacobian o f the i m p l i c i t f unc t i on that
d e f i n e s the mean u t i l i t y
11 theta2w<−matrix (0 , r theta , c theta )
12 theta2w [ which ( theta2w st !=0) ]<−theta2
13 expmu<−exp ( mufunc ( x2 , theta2w , id ) )
14 share s<−sha r e f cn ( mval [ id ] , expmu) [ [ 1 ] ] # n x ns
15 x2<−x2 [ id , ]
16 n<−nrow ( x2 )
17 K<−nco l ( x2 )
18 J<−nco l ( theta2w )−1
19
20 t h e t i<−which ( theta2w !=0 , a r r . ind=TRUE) [ , 1 ]
21 t h e t j<−which ( theta2w !=0 , a r r . ind=TRUE) [ , 2 ]
22 r e l<−t h e t i +( the t j −1)∗max( t h e t i )
23
24 f 1<−matrix (0 , n ,K∗( J+1) )
25 f<−matrix (0 , n , l ength ( t h e t i ) )
26
27 # Computing ( p a r t i a l share ) /( p a r t i a l sigma )
28 f o r ( i in 1 :K) {
29 xv<−( x2 [ , i ] %∗% t ( rep (1 , ns ) ) ) ∗ v f u l l [ id , ( ( ns ∗( i −1)+1) : ( ns∗ i ) ) ] # n
x ns
30 sum1<−rbind ( colSums ( xv∗ share s ) ) # 1 x ns
31 f 1 [ , i ]<−rowMeans ( share s ∗( xv−sum1 [ rep (1 , n) , ] ) ) # n x 1
32 }
33 # Computing ( p a r t i a l share ) /( p a r t i a l p i )
34 f o r ( j in 1 : J ) {
35 d<−d f u l l [ id , ( ( ns ∗( j−1)+1) : ( ns∗ j ) ) ]
36 temp1<−matrix (0 , n ,K)
37 f o r ( i in 1 :K) {
38 xd<−( x2 [ , i ] %∗% t ( rep (1 , ns ) ) ) ∗d # n x ns
162
39 sum1<−rbind ( colSums ( xd∗ share s ) ) # 1 x ns
40 temp1 [ , i ]<−rowMeans ( share s ∗(xd−sum1 [ rep (1 , n) , ] ) ) #n x 1
41 }
42 f 1 [ , ( (K∗ j +1) : (K∗( j +1) ) ) ]<−temp1
43 }
44 # Computing ( p a r t i a l d e l t a ) /( p a r t i a l theta2 )
45 temp <− share s
46 H1<−temp %∗% t ( temp )
47 H<−( d iag ( rowSums( temp ) )−H1) /ns
48 f <− −s o l v e (H, t o l=1e−20) %∗% f1 [ , r e l ]
49 re turn ( f )
50 }
163
Appendix C: Appendix to Chapter 3
C.1 Additional Tables
(Continued on next page...)
164
Table C.1: Grower Conferences and Online Grower Listservs Surveyed
Conference / Organization Dates
Pacific Northwest Vegetable Conference & Trade Show,
Kennewick, WA
November 12-13, 2014
29th Annual Southeast Vegetable & Fruit Expo, Myrtle
Beach, SC
December 2-3, 2014
Great Lakes Fruit and Vegetable Expo, Grand Rapids,
MI
December 9-11, 2014
Southeast Regional Fruit & Vegetable Conference, Sa-
vannah, GA
January 8-11, 2015
Future Harvest Chesapeake Alliance for Sustainable
Agriculture Conference, College Park, MD
January 15-17, 2015
Ohio Produce Growers & Marketers Association
Congress, Sandusky, OH
January 19-21, 2015
Mid-Atlantic Fruit and Vegetable Convention, Hershey,
PA
January 27-29, 2015
New Jersey Agricultural Convention and Trade Show,
Atlantic City, NJ
February 2-5, 2015
Georgia Fruit & Vegetable Growers Association January 9, 2014
Michigan State University Extension December 11, 2014
Future Harvest Chesapeake Alliance for Sustainable
Agriculture
January 16, 2015
Center for Produce Safety February 10, 2015
North Carolina Farm Bureau January 23, 2015
Ohio State University Extension January 18, 2015
Oregon State University Extension January 30, 2015
Pennsylvania Association for Sustainable Agriculture January 27, 2015
Pennsylvania Vegetable Growers Association January 26, 2015
University of Florida Extension January 22, 2015
Virginia Association for Biological Farming February 8, 2015
Vegetable Growers Association of New Jersey February 2, 2015
Carolina Farm Stewardship Association December 8, 2014
Cornell Produce Safety Alliance December 9, 2014
Michigan Food & Farming Systems December 9, 2014
Note: The survey closed on May 2, 2015, for all online grower Listservs
165
Table C.2: Estimated Coefficients for Double Hurdle Specification With No Interactions
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Washing
Product
Employee
Sanita-
tion &
Hygiene
Written
Records
Soil
Amend-
ment
Treat-
ment
Third-
Party
Audits
Use Variables
Intercept -0.4001 -0.6526∗∗ 1.4989∗∗∗ -0.5749* 2.0709∗∗∗ 0.2122 0.7985 -1.8187∗∗∗
(0.2756) (0.2851) (0.4299) (0.3057) (0.214) (0.3289) (0.6579) (0.4246)
Log Fruit and Vegetable Acreage 0.2276∗∗∗ 0.1552∗∗∗ -0.0872 -0.0391 0.0679 0.1127∗ -0.0691 0.3062∗∗∗
(0.0477) (0.0466) (0.0648) (0.0487) (0.0617) (0.0605) (0.0684) (0.0661)
Contractual Obligation 0.4133∗∗ 0.6936∗∗∗ 0.5795∗ -0.1107 0.1216 1.4203∗∗∗ 0.3883 1.4397∗∗∗
(0.2084) (0.2153) (0.326) (0.2335) (0.3106) (0.4935) (0.3829) (0.2464)
Wholesale / Other Sale Share 0.0013 -0.0015 -0.0026 0.004 -0.0065∗∗ 0.0002 -0.0061 0.0006
(0.0022) (0.0023) (0.0033) (0.0028) (0.0032) (0.0031) (0.0042) (0.0034)
Sustainable 0.0797 -0.181 0.078 0.8183∗∗∗ 0.1819 0.1331 0.0626 -0.3221
(0.1835) (0.1824) (0.2933) (0.2413) (0.2511) (0.2099) (0.2386) (0.3378)
Berries -0.0067 -0.1248 0.7460∗∗∗ 0.1332 0.4751∗∗ -0.0721 -0.1027 -0.2294
(0.1576) (0.1598) (0.2655) (0.1824) (0.1975) (0.1837) (0.2313) (0.2402)
Fruit and Tree Nut 0.116 -0.0539 -0.1435 0.2501 -0.3784∗ 0.2361 0.0677 -0.0899
(0.1628) (0.1658) (0.273) (0.1971) (0.1941) (0.2017) (0.2329) (0.2609)
Vegetables -0.3321 0.4017∗ -0.164 1.2003∗∗∗ -0.9954∗∗∗ 0.2201 -0.1544 -0.3746
(0.2312) (0.233) (0.3523) (0.2441) (0.191) (0.2766) (0.6034) (0.3096)
Expenditure Variables
Intercept 2.6668∗∗∗ 3.0660∗∗∗ 4.3212∗∗∗ 4.8510∗∗∗ 5.0245∗∗∗ 0.7847∗∗∗ 4.2457∗∗∗ 5.9150∗∗∗
(0.4903) (0.5985) (0.3617) (0.8305) (0.3325) (0.2859) (1.2161) (0.675)
Log Fruit and Vegetable Acreage 0.6955∗∗∗ 0.5608∗∗∗ 0.5475∗∗∗ 0.6816∗∗∗ 0.6319∗∗∗ 0.2560∗∗∗ 0.9180∗∗∗ 0.3798∗∗∗
Continued. . .
166
Table C.2 – continued from previous page
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Washing
Product
Employee
Sanita-
tion &
Hygiene
Written
Records
Soil
Amend-
ment
Treat-
ment
Third-
Party
Audits
(0.0764) (0.0795) (0.0602) (0.0719) (0.0542) (0.0416) (0.1587) (0.0986)
Sustainable 0.4638 0.4685 0.6253∗∗ 0.7333∗∗ 0.2675 0.4459∗∗ 0.2322 0.1249
(0.3922) (0.3754) (0.2448) (0.3231) (0.2482) (0.1954) (0.3848) (0.5112)
Berries 0.0034 -0.2336 0.1864 -0.1714 -0.125 0.005 -0.0479 1.0149∗∗∗
(0.3021) (0.3223) (0.226) (0.2709) (0.2114) (0.1675) (0.4127) (0.3179)
Fruit and Tree Nut -0.3933 0.2712 -0.1011 0.3285 -0.0446 -0.3887∗∗ 0.241 -0.0873
(0.3205) (0.3233) (0.226) (0.2798) (0.2199) (0.1773) (0.4118) (0.3829)
Vegetables -0.5517 0.2044 0.2788 0.6249 0.0139 -0.3462 0.5856 0.1353
(0.3793) (0.4108) (0.3021) (0.6044) (0.2774) (0.2244) (1.1349) (0.3976)
Sigma 1.8964∗∗∗ 1.6828∗∗∗ 1.4946∗∗∗ 1.5790∗∗∗ 1.5092∗∗∗ 1.2357∗∗∗ 1.3839∗∗∗ 1.0722∗∗∗
(0.1611) (0.2371) (0.0961) (0.1014) (0.0716) (0.0769) (0.1635) (0.1031)
Rho 0.9090∗∗∗ 0.8283∗∗∗ 0.6672∗∗∗ 0.2111 0.0001 0.8256∗∗∗ 0.325 -0.2603
(0.0423) (0.1378) (0.2286) (0.4141) ) (0.0881) (0.4009) (0.2813)
No. of Observations 303 294 297 294 290 286 157 283
Log Likelihood -389.6464 -347.5561 -428.3126 -442.7636 -496.4305 -453.1219 -198.3643 -164.5507
Note: Standard errors (reported in parentheses) were estimated using the delta method. Asterisk (∗), double asterisk (∗∗), and
triple asterisk (∗∗∗) indicate significance at the 10, 5 and 1 percent level, respectively.
167
Table C.3: Estimated Coefficients for Double Hurdle Specification With Interactions
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Employee
Sanita-
tion &
Hygiene
Soil
Amend-
ment
Treat-
ment
Use Variables
Intercept -0.3409 -0.6411∗∗ 1.4962∗∗∗ 2.0511∗∗∗ 0.7974
(0.28) (0.2836) (0.4332) (0.4536) (0.6542)
Log Fruit and Vegetable Acreage 0.2011∗∗∗ 0.1473∗∗∗ -0.0886 0.0685 -0.069
(0.0493) (0.0443) (0.0652) (0.0614) (0.0684)
Contractual Obligation 0.4782∗∗ 0.7127∗∗∗ 0.5780∗ 0.1761 0.3877
(0.2373) (0.2022) (0.3273) (0.3141) (0.3852)
Wholesale / Other Sale Share 0.0021 -0.0018 -0.0025 -0.0071∗∗ -0.0061
(0.0024) (0.0022) (0.0033) (0.0035) (0.0042)
Sustainable 0.0943 -0.1921 0.0525 0.1678 0.0629
(0.1863) (0.1818) (0.2916) (0.2562) (0.2387)
Berries -0.0724 -0.1226 0.7366∗∗∗ 0.4541∗ -0.1022
(0.1607) (0.1579) (0.2641) (0.2317) (0.2315)
Fruit and Tree Nut 0.1431 -0.0279 -0.1342 -0.3612 0.0678
(0.1683) (0.1655) (0.274) (0.2387) (0.2328)
Vegetables -0.34 0.4024∗ -0.1519 -0.9626∗∗ -0.154
(0.2357) (0.2309) (0.3542) (0.3936) (0.5992)
Expenditure Variables
Intercept 2.8170∗∗∗ 3.5581∗∗∗ 4.1740∗∗∗ 3.6239∗∗∗ 3.6368∗∗
(0.6047) (1.0693) (0.3958) (0.5814) (1.5395)
Sustainable 0.49 0.4125 0.6788∗∗∗ 0.1592 0.2349
Continued. . .
168
Table C.3 – continued from previous page
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Employee
Sanita-
tion &
Hygiene
Soil
Amend-
ment
Treat-
ment
(0.3559) (0.3652) (0.2477) (0.2416) (0.3965)
Berries -0.0768 -0.2777 0.123 -0.278 -0.0504
(0.2571) (0.3149) (0.2318) (0.2235) (0.4114)
Fruit and Tree Nut -0.1766 0.2693 -0.1104 -0.1386 0.2684
(0.283) (0.3231) (0.2267) (0.2274) (0.412)
Vegetables -0.275 0.4475 0.2136 0.1309 0.5797
(0.3298) (0.4135) (0.3072) (0.298) (1.1347)
Log Fruit and Veg. Acreage x Water Samples 0.5510∗∗∗
(0.0754)
Log Fruit and Veg. Acreage x Soil Amendment Samples 0.2094
(0.1475)
Log Fruit and Veg. Acreage x Product Samples 0.0604
(0.1126)
Water Samples 0.0128
(0.506)
Soil Amendment Samples -0.0747
(0.4907)
Product Samples 1.4928∗∗∗
(0.5315)
Log Fruit and Veg. Acreage x Flooding -0.2032
(0.6088)
Log Fruit and Veg. Acreage x Wildlife Intrusion 0.5097∗∗∗
Continued. . .
169
Table C.3 – continued from previous page
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Employee
Sanita-
tion &
Hygiene
Soil
Amend-
ment
Treat-
ment
(0.1049)
Log Fruit and Veg. Acreage x Other Causes 0.1554
(0.1363)
Log Fruit and Veg. Acreage x Flooding x Wildlife Intrusion 0.2448
(0.6166)
Flooding Inspections 0.1746
(1.7905)
Animal Intrusion Inspections -0.8544
(1.029)
Other Inspections -0.0412
(0.5903)
Flooding x Wildlife Intrusion Inspections 0.0268
(1.8359)
Log Fruit and Veg. Acreage x Wash Containers -0.0291
(0.1022)
Log Fruit and Veg. Acreage x New Containers 0.5607∗∗∗
(0.09)
New Harvest Containers 0.4063
(0.3128)
Log Fruit and Veg. Acreage x Employee Training 0.0877
(0.1305)
Log Fruit and Veg. Acreage x Toilet/Handwash Facilities -0.1345
Continued. . .
170
Table C.3 – continued from previous page
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Employee
Sanita-
tion &
Hygiene
Soil
Amend-
ment
Treat-
ment
(0.1323)
Log Fruit and Veg. Acreage x Equipment Sanitation 0.0337
(0.1081)
Log Fruit and Veg. Acreage x Building Sanitation 0.6863∗∗∗
(0.2062)
Log Fruit and Veg. Acreage x Sewage/Trash Disposal -0.1469
(0.1905)
Log Fruit and Veg. Acreage x Other Preventive Actions 0.117
(0.1386)
Employee Education/Training 0.7699∗
(0.4175)
Equipment & Tool Sanitation 0.3901
(0.3985)
Building Sanitation 0.1024
(0.3516)
Toilet & Handwashing Facilities 0.1537
(0.5569)
Proper Disposal of Sewage/Trash 0.7019
(0.479)
Other Employee Actions -0.2481
(0.4841)
Total Number of Employees 0.0012∗
Continued. . .
171
Table C.3 – continued from previous page
Variables Sampling
&
Testing
Field
Inspec-
tions
Harvest
Con-
tainer
Sanita-
tion
Employee
Sanita-
tion &
Hygiene
Soil
Amend-
ment
Treat-
ment
(0.0007)
Log Fruit and Veg. Acreage x Single Soil Amendment 1.5959
(1.078)
Log Fruit and Veg. Acreage x Multiple Soil Amendments 0.8995∗∗∗
(0.1626)
Multiple Soil Amendments 0.6437
(0.9754)
Sigma 1.4970∗∗∗ 1.6996∗∗∗ 1.4895∗∗∗ 1.4037∗∗∗ 1.3782∗∗∗
(0.1733) (0.2066) (0.0983) (0.0905) (0.1636)
Rho 0.7779∗∗∗ 0.8982∗∗∗ 0.6436∗∗∗ -0.3263 0.325
(0.1451) (0.0801) (0.2563) (0.5064) (0.4048)
No. of Observations 303 294 297 290 157
Log Likelihood -374.9738 -341.5166 -423.2073 -468.5369 -198.1192
Note: Standard errors (reported in parentheses) were estimated using the delta method. Asterisk (∗), double asterisk (∗∗),
and triple asterisk (∗∗∗) indicate significance at the 10, 5 and 1 percent level, respectively.
172
C.2 Survey Instrument
Own
Manage
Own and Manage
Neither
No
Yes
 
Please answer the following questions to the best of your abilities with respect to the 2014 growing season (unless otherwise noted).
Reasonably accurate estimates are acceptable.
Background Information
This is a research project being conducted by Professor Erik Lichtenberg at the University of
Maryland, College Park. The purpose of this research is to identify the current prevalence and burden
of different food safety risk-reduction strategies for vegetable and fruit growers nationwide.
Your participation in this research study is voluntary. You may choose not to participate. If you
decide to participate in this research survey, you may withdraw at any time. The procedure involves
filling an online survey that will take approximately 15 to 20 minutes.  At the end of this survey, you
will be asked if you are interested in providing contact information in order to participate in a follow-
up survey. The decision to provide this information is completely voluntary. Your responses will be
kept confidential. All data will be stored in a password protected electronic database. The results of
this study will be used for scholarly purposes only.
If you have any questions about the research study, please contact Professor Erik Lichtenberg,
Department of Agricultural and Resource Economics, University of Maryland, College Park at (301)
405-1279 or elichten@umd.edu. This research has been reviewed according to the University of
Maryland IRB procedures for research involving human subjects. If you have questions about your
rights as a research participant or wish to report a research-related injury, please contact: 
University of Maryland College Park 
Institutional Review Board Office 
IRB Protocol 11-0513 
1204 Marie Mount 
College Park, Maryland, 20742 
E-mail: irb@umd.edu 
Telephone: (301) 405-0678 
By clicking the “NEXT” button below you are indicating that you are at least 18 years of age;
you have read this consent form or have had it read to you; and you voluntarily agree to
participate in this research study.
Do you own or manage a farm?
Is ALL the produce you grow intended for canning or a similar type of commercial processing that
kills pathogens?
What vegetables and/or fruit were produced in 2014?
173
Yes
No
Artichokes Eggplant Potatoes and/or Sweet Potatoes
Asparagus Fresh Herbs Radishes and/or Turnips
Beans (any type) Grains, Oilseeds, and/or Hay Squash (any type) and/or Pumpkins
Beets Leafy Greens Sweet Corn
Berries (any type) Melons (any type) Tomatoes
Broccoli and/or Cauliflower Okra Tree Fruits
Brussel Sprouts Onions (any type) Tree Nuts
Carrots Peas (any type) Other Vegetables
Celery and/or Rhubarb Peppers (any type) Other Fruit
Cucumbers     
Were livestock or other domesticated animals raised on the farm, as well?
In what county and state are your farm fields located?
State
County
How many full-time and seasonal employees were employed?
Full-time employees
Seasonal employees
In total, how many acres of land were in production?
Vegetables and Fruit  acres
All Farm Production  acres
What was the total annual revenue and total annual expenditures for all farm production?
Estimates are acceptable.
Total Annual Revenue $ 
Total Annual Expenditures $ 
What share of total annual revenue and total annual expenditures were attributable to vegetable and
fruit production?  
Share of Revenue  %
Share of Expenditures  %
174
Yes
No
Yes
No
Please identify the percentage share of all vegetables and fruit sold directly to the listed entities. 
The column must sum to 100.
  All Vegetables and Fruit
Direct Sales   0  %
Grocery Retailers   0  %
Foodservice Operations   0  %
Produce Wholesalers/Repackers   0  %
Mass Merchandisers   0  %
Exporters   0  %
Brokers   0  %
Other   0  %
Total   0  %
Did you have any contractual obligation to adhere to any specific safety standards and testing
procedures?
Please identify the entities with which you had contractual obligations regarding food safety and any
corresponding safety standards (e.g., guidance documents, certification programs, USDA GAP,
Harmonized GAP, etc.).
Vegetables
and Fruit Vegetables and Fruit  
Contractual
Safety
Obligation?
Safety Standard (e.g., GAPs,
Certification, etc.)
Grocery Retailers  
Foodservice Operations  
Produce
Wholesalers/Repackers  
Mass Merchandisers  
Exporters  
Brokers  
Shippers  
Other:  
For vegetable and fruit production, did operations include the use of a soil amendment or soil
treatment that contained animal manures or animal products (e.g., raw manure, compost, fish
emulsions, fish meal, blood meal, etc.)?
175
Yes
No
Yes: All soil amendments were treated.
Yes: Some soil amendments were treated, while some were left untreated.
No: All soil amendments were untreated.
Yes
No
Yes
No
Was more than one soil amendment and/or soil treatment of animal origin used for vegetable and
fruit production?
Were any biological soil amendments of animal origin treated using scientifically-valid physical,
chemical or composting processes before application?
Was the biological soil amendment of animal origin treated using a scientifically-valid physical,
chemical or composting process before application?
What was the approximate total annual cost of treating biological soil amendments of animal origin
before application?
Estimates are acceptable.
$ 
What was the shortest time interval in days between the application of soil amendments of animal
origin and harvesting of crops for any growing area on which they were applied?
Treated Soil Amendments  days
Untreated Soil Amendments  days
End of Survey
Thank you for participating in our survey!
If you are interested in learning the survey results, please contact Erik Lichtenberg at elichten@umd.edu.
Please click NEXT to end the survey.
Sampling & Testing, Wildlife, & Flooding
Was more than one water source used for growing, harvesting, packing, or holding vegetables and
fruit?
What water source(s) was used for growing, harvesting, packing, or holding vegetables and fruit?
Please check all that apply. 
176
Pond or Lake
River
Stream or Spring
Shallow Well (less than 30 feet)
Deep Well (greater than 30 feet)
Municipal / City Water
Other
Water Samples
Soil Amendment and/or Soil Treatment Samples
Crop/Product Samples
No samples were taken
Grazing only
Working Animals only
Grazing and Working Animals
Neither
Please indicate whether the following samples were collected for microbial testing (e.g., pathogens,
generic E. coli, coliforms, etc.).
If no samples were taken, please check the last box.
How frequently were samples collected?
   Weekly Montlhy Once a Season Never Other
Water   
Soil Amendments and/or Soil
Treatments   
Crop/Product   
How frequently were the following samples collected?
Water Samples
Soil Amendments and/or Soil Treatment
Samples
Crop/Product Samples
What were the total annual expenditures associated with sampling and testing (including employee
wages, materials, etc.)?
Estimates are acceptable.
$ 
Were animals allowed to graze and/or used as working animals in fields where vegetables and fruit is
grown?
177
Yes
No
Flooding
Animal Intrusion
Other Contamination Sources
No field inspections were conducted
Sanitary Surveys/Sanitation
Additional Testing
Water Treatments
Leave enough time between last irrigation and harvest or between harvest and end of storage for microbes to die off
Processing/Treatment of Soil Amendments
Use of Substitutes for Contaminated Materials
Material Disposal
Product Disposal
What was the shortest time interval in days between grazing and harvesting of crops for any growing
area that was grazed?
 days
Were any measures taken to prevent the introduction of hazards onto covered produce from working
animals (e.g., segregated horse paths, etc.)?
Were field inspections for flooding, animal intrusion, and/or other contamination sources conducted
prior to harvest?
If no field inspections were conducted, please check the last box.
What was the total annual cost of conducting field inspections (including employee wages, etc.)?
Estimates are acceptable.
$ 
For each of the following, did test results, flooding, animal intrusion, and/or other contamination
sources lead to remedial actions (e.g., sanitation, product disposal, water treatment, etc.)?
   Yes No
Test Results   
Flooding   
Animal Intrusion   
Other Contamination Sources   
Please identify all remedial actions taken following testing, flooding, and/or animal intrusion. 
178
Delayed Future Production on Site
Other:
Other:
Yes
No
Yes
No
What was the approximate total annual cost associated with these remedial actions (including the
value of any disposed materials/products, value of lost future production on site, etc.)?
Estimates are acceptable.
$ 
Preventive Actions
Were harvest containers washed and/or sanitized prior to harvest for any vegetable or fruit crops?
Were new harvest containers used for any vegetable or fruit crops?
For which vegetables and fruit crops were harvest containers washed and/or sanitized prior to
harvest?
Artichokes Eggplant Potatoes and/or Sweet Potatoes
Asparagus Fresh Herbs Radishes and/or Turnips
Beans (any type) Grains, Oilseeds, and/or Hay Squash (any type) and/or Pumpkins
Beets Leafy Greens Sweet Corn
Berries (any type) Melons (any type) Tomatoes
Broccoli and/or Cauliflower Okra Tree Fruits
Brussel Sprouts Onions (any type) Tree Nuts
Carrots Peas (any type) Other Vegetables
Celery and/or Rhubarb Peppers (any type) Other Fruit
Cucumbers     
For which vegetables and fruit crops were new harvest containers used?
Artichokes Eggplant Potatoes and/or Sweet Potatoes
Asparagus Fresh Herbs Radishes and/or Turnips
Beans (any type) Grains, Oilseeds, and/or Hay Squash (any type) and/or Pumpkins
Beets Leafy Greens Sweet Corn
Berries (any type) Melons (any type) Tomatoes
Broccoli and/or Cauliflower Okra Tree Fruits
179
Yes
No
Employee Education/Training
Clean and Accessible Toilet and Handwashing Facilities
Brussel Sprouts Onions (any type) Tree Nuts
Carrots Peas (any type) Other Vegetables
Celery and/or Rhubarb Peppers (any type) Other Fruit
Cucumbers     
What was the total approximate annual cost of washing and/or sanitizing harvest containers
(including the cost of disinfectants, employee wages, etc.)?
Estimates are acceptable.
$ 
What was the total approximate annual cost for the new harvest containers?
Estimates are acceptable.
$ 
Were any harvested crops/products washed prior to storage or sale?
Which vegetable and fruit crops were washed prior to storage or sale?
Artichokes Eggplant Potatoes and/or Sweet Potatoes
Asparagus Fresh Herbs Radishes and/or Turnips
Beans (any type) Grains, Oilseeds, and/or Hay Squash (any type) and/or Pumpkins
Beets Leafy Greens Sweet Corn
Berries (any type) Melons (any type) Tomatoes
Broccoli and/or Cauliflower Okra Tree Fruits
Brussel Sprouts Onions (any type) Tree Nuts
Carrots Peas (any type) Other Vegetables
Celery and/or Rhubarb Peppers (any type) Other Fruit
Cucumbers     
What was the approximate total annual cost of washing the crops/products (including employee
wages, etc.)?
Estimates are acceptable.
$ 
With regards to employee hygiene and general sanitation, please identify all preventive actions taken.
If no action was taken, please check the last box.
180
Equipment and Tool Sanitation
Building Sanitation
Proper Disposal of Sewage and Trash
Other:
Other:
No preventive action was taken.
Yes
No
Yes
No
Food Hygiene and Food Safety Policies and Procedures
Water Treatment Methods
What was the approximate total annual cost associated with these preventive actions?
Estimates are acceptable.
$ 
Besides field inspections, employee hygiene precautions, general sanitation, washing, sampling,
and/or testing, were any other preventive actions taken to directly reduce the risk of pathogen
contamination?
What other preventive actions were taken to directly reduce the risk of pathogen contamination?
What was the total annual cost of implementing these additional preventive actions (including
employee wages, cost of materials and equipment, etc.)?
Estimates are acceptable.
$ 
Do you have a third-party food safety audit program in place? 
What is the total annual cost of these food safety audits?
Estimates are acceptable.
$ 
Do you keep written records or documentation for any of the following?
If not, please check the last box.
181
Water Treatment Monitoring Results
Water Testing Results
Soil Amendment and/or Soil Treatement Application Dates
Produce/Crop Harvest Dates
Soil Amendment and/or Soil Treatement Testing Results
Produce/Crop Testing Results
Flooding
Animal Intrusion
Other Contamination
No written records or documentation were kept
On a weekly basis, how much time do you spend on record keeping (in hours)?
 hours
182
References
Ackerberg, D. A. (2001). Empirically distinguishing informative and prestige effects
of advertising. The RAND Journal of Economics, 32(2):316–333.
Ackerberg, D. A. (2003). Advertising, learning, and consumer choice in experi-
ence good markets: an empirical examination. International Economic Review,
44(3):1007–1040.
Adalja, A. A. (2016). Who pays for voluntary quality certification? evidence from
the non-gmo project verified label. Working Paper.
Akerlof, G. (1970). The market for lemons: qualitative uncertainlyand market mech-
anism. Quarterly Journal of Economics, 89.
Andreoni, J. (1990). Impure altruism and donations to public goods: A theory of
warm-glow giving. The Economic Journal, 100(401):464–477.
Antle, J. M. (1999). Benefits and costs of food safety regulation. Food Policy,
24(6):605–623.
Antle, J. M. (2000). No such thing as a free safe lunch: the cost of food safety
regulation in the meat industry. American Journal of Agricultural Economics,
82(2):310–322.
Berry, S., Levinsohn, J., and Pakes, A. (1995). Automobile prices in market equi-
librium. Econometrica, 63(4):841–890.
Berry, S., Levinsohn, J., and Pakes, A. (2004). Differentiated products demand
systems from a combination of micro and macro data: The new car market.
Journal of Political Economy, 112(1):68–105.
Berry, S. T. (1994). Estimating Discrete-Choice Models of Product Differentiation.
The RAND Journal of Economics, 25(2):242–262.
Besanko, D., Dube´, J.-P., and Gupta, S. (2005). Own-Brand and Cross-Brand Retail
Pass-Through. Marketing Science, 24(1):123–137.
Bonroy, O. and Constantatos, C. (2014). On the economics of labels: How their
introduction affects the functioning of markets and the welfare of all participants.
American Journal of Agricultural Economics, 97(1):239–259.
183
Bourn, D. and Prescott, J. (2002). A comparison of the nutritional value, sen-
sory qualities, and food safety of organically and conventionally produced foods.
Critical reviews in food science and nutrition, 42(1):1–34.
Bren, L. (2003). Genetic engineering: The future of foods. http:
//permanent.access.gpo.gov/lps1609/www.fda.gov/fdac/features/2003/
603_food.html. Accessed: 30-October-2016.
Carriquiry, M. and Babcock, B. A. (2007). Reputations, market structure, and the
choice of quality assurance systems in the food industry. American Journal of
Agricultural Economics, 89(1):12–23.
Carter, C. A., Grure, G. P., McLaughlin, P., and Maclachlan, M. (2012). California’s
proposition 37: effects of mandatory labeling of GM food. ARE Update, 15(6):3–8.
Chamberlain, G. (1982). Multivariate regression models for panel data. Journal of
Econometrics, 18(1):5–46.
Chidmi, B. (2012). Vertical relationships in the ready-to-eat breakfast cereal indus-
try in boston. Agribusiness, 28(3):241–259.
Costanigro, M. and Lusk, J. L. (2014). The signaling effect of mandatory labels on
genetically engineered food. Food Policy, 49, Part 1:259–267.
Dannenberg, A., Scatasta, S., and Sturm, B. (2011). Mandatory versus voluntary
labelling of genetically modified food: evidence from an economic experiment.
Agricultural Economics, 42(3):373–386.
Darby, M. R. and Karni, E. (1973). Free competition and the optimal amount of
fraud. Journal of law and economics, 16(1):67–88.
Desai, P. (2001). Quality Segmentation in Spatial Markets: When Does Cannibal-
ization Affect Product Line Design? Marketing Science, 20(3):265–283.
Dranove, D. and Jin, G. Z. (2010). Quality disclosure and certification: Theory and
practice. Journal of Economic Literature, 48(4):935–963.
Driven to Discover (2012). Cost of food safety: Leafy greens food safety cost study
final report. http://safety.cfans.umn.edu/links/costs. Accessed 2015-05-
20.
Dube´, J.-P., Fox, J. T., and Su, C.-L. (2012). Improving the numerical performance
of static and dynamic aggregate discrete choice random coefficients demand esti-
mation. Econometrica, 80(5):2231–2267.
Fares, M. and Rouviere, E. (2010). The implementation mechanisms of voluntary
food safety systems. Food Policy, 35(5):412–418.
FDA (2015). Guidance for industry: Voluntary labeling indicating whether foods
have or have not been developed using bioengineering.
184
Fulponi, L. (2006). Private voluntary standards in the food system: The perspective
of major food retailers in oecd countries. Food Policy, 31(1):1–13.
Goldberg, P. K. (1995). Product differentiation and oligopoly in international mar-
kets: The case of the U.S. automobile industry. Econometrica, 63(4):891–951.
Greene, C., Wechsler, S., Adalja, A., and Hanson, J. (2016). Economic Issues in
the Coexistence of Organic, Genetically Engineered (GE), and Non-GE Crops.
Technical Report EIB-149, USDA ERS.
Grossman, S. J. (1981). The informational role of warranties and private disclosure
about product quality. Journal of Law and Economics, 24(3):461–483.
Hall, P. K. (2016). Congress finalizes mandatory
gmo labeling law. http://nationalaglawcenter.org/
congress-finalizes-mandatory-gmo-labeling-law-2/. Accessed: 2016-
09-21.
Hardesty, S. D. and Kusunose, Y. (2009). Growers’ compliance costs for the leafy
greens marketing agreement and other food safety programs. UC Small Farm
Program Brief, 20.
Hassanein, N. (2011). Matters of scale and the politics of the food safety modern-
ization act. Agriculture and human values, 28(4):577–581.
Hennessy, D. A., Roosen, J., and Miranowski, J. A. (2001). Leadership and the
provision of safe food. American Journal of Agricultural Economics, 83(4):862–
874.
Henson, S. and Caswell, J. (1999). Food safety regulation: an overview of contem-
porary issues. Food Policy, 24(6):589–603.
Henson, S. and Reardon, T. (2005). Private agri-food standards: Implications for
food policy and the agri-food system. Food Policy, 30(3):241–253.
Holleran, E., Bredahl, M. E., and Zaibet, L. (1999). Private incentives for adopting
food safety and quality assurance. Food Policy, 24(6):669–683.
Hooker, N. H., Nayga, R. M., and Siebert, J. W. (2002). The impact of haccp on costs
and product exit. Journal of Agricultural and Applied Economics, 34(01):165–174.
Hotelling, H. (1929). Stability in competition. The Economic Journal, 39(153):41–
57.
Houde, S. (2016). Consumers’ Response to Quality Disclosure and Certification: An
Application to Energy Labels. Working Paper.
Hughner, R. S., McDonagh, P., Prothero, A., Shultz, C. J., and Stanton, J. (2007).
Who are organic food consumers? a compilation and review of why people pur-
chase organic food. Journal of consumer behaviour, 6(2-3):94–110.
185
Jalonick, M. C. (2015). Usda develops new government label for gmo-free
products. http://www.washingtonpost.com/national/health-science/
usda-develops-certification-labeling-for-gmo-free-foods/2015/05/
14/61822a1c-fa54-11e4-9ef4-1bb7ce3b3fb7_story.html. Accessed: 2014-
07-03.
Knittel, C. R. and Metaxoglou, K. (2014). Estimation of random-coefficient de-
mand models: Two empiricists’ perspective. Review of Economics and Statistics,
96(1):34–59.
Knutson, R. D., Palma, M. A., Paggi, M., Seale, J., Ribera, L. A., and Bessler, D.
(2014). Role of trade in satisfying us fresh fruit and vegetable demand. Journal
of International Food & Agribusiness Marketing, 26(4):326–343.
Leggett, C. G. (2002). Environmental Valuation with Imperfect Information: The
Case of the Random Utility Model. Environmental and Resource Economics,
23(3):343–355.
Lichtenberg, E. and Page, E. T. (2016). Prevalence and cost of on-farm produce
safety measures in the mid-atlantic. Food Control, 69:315–323.
Loader, R. and Hobbs, J. E. (1999). Strategic responses to food safety legislation.
Food Policy, 24(6):685–706.
Low, Sarah A, e. a. (2015). Trends in us local and regional food systems: A report
to congress. USDA Economic Research Service, AP-068.
Lowe, P. (2016). Senators reach deal on national gmo labeling bill.
http://www.npr.org/sections/thesalt/2016/06/23/483290269/
senate-unveils-a-national-gmo-labeling-bill. Accessed: 23-June-2016.
Lusk, J. L., Jamal, M., Kurlander, L., Roucan, M., and Taulman, L. (2005). A meta-
analysis of genetically modified food valuation studies. Journal of Agricultural and
Resource Economics, 30(1):28–44.
MacDonald, J. M. and Crutchfield, S. (1996). Modeling the costs of food safety
regulation. American Journal of Agricultural Economics, 78(5):1285–1290.
Martinez, Steve, e. a. (2010). Local food systems: Concepts, impacts, and issues.
USDA Economic Research Service, ERR 97.
McCluskey, J. (2000). A game theoretic approach to organic foods: An analysis of
asymmetric information and policy. Agricultural and Resource Economics Review,
29(1):1–9.
McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior.
Frontiers in Econometrics, pages 105–142.
186
Milgrom, P. and Roberts, J. (1986). Relying on the information of interested parties.
RAND Journal of Economics, 17(1):18–32.
Moorthy, K. S. (1984). Market segmentation, self-selection, and product line design.
Marketing Science, 3(4):288–307.
Mussa, M. and Rosen, S. (1978). Monopoly and product quality. Journal of Eco-
nomic Theory, 18(2):301–317.
Muth, M. K., Wohlgenant, M. K., Karns, S. A., and Anderson, D. W. (2003). Ex-
plaining plant exit in the us meat and poultry industries. Journal of Agricultural
& Food Industrial Organization, 1(1):7.
Nelson, P. (1970). Information and consumer behavior. Journal of Political Econ-
omy, 78(2):311–329.
Nevo, A. (2000a). Mergers with differentiated products: The case of the ready-to-eat
cereal industry. The RAND Journal of Economics, pages 395–421.
Nevo, A. (2000b). A practitioner’s guide to estimation of random-coefficients logit
models of demand. Journal of Economics and Management Strategy, 9(4):513–
548.
Nevo, A. (2001). Measuring market power in the ready-to-eat cereal industry. Econo-
metrica, 69(2):307–342.
Nevo, A. and Wolfram, C. (2002). Why do manufacturers issue coupons? an empir-
ical analysis of breakfast cereals. RAND Journal of Economics, pages 319–339.
Non-GMO Project (2014a). Faqs-product verification. http://www.
nongmoproject.org/product-verification/faqs/. Accessed: 2014-08-09.
Non-GMO Project (2014b). Gmo facts. http://www.nongmoproject.org/
learn-more/. Accessed: 2014-08-09.
Non-GMO Project (2014c). Non-GMO Project Standard. Non-GMO Project,
Bellingham, Washington. Available at: http://www.nongmoproject.org/
wp-content/uploads/2014/06/Non-GMO-Project-Standard-v11.pdf.
Non-GMO Project (2014d). The non-gmo project verified seal. http://www.
nongmoproject.org/learn-more/understanding-our-seal/. Accessed: 2014-
08-09.
Non-GMO Project (2014e). What is GMO? http://www.nongmoproject.org/
learn-more/what-is-gmo/. Accessed: 2014-08-09.
Non-GMO Project (2017). History. https://www.nongmoproject.org/about/
history/. Accessed: 2017-05-04.
187
Onyango, B., Nayga, R. M., and Govindasamy, R. (2006). U.s. consumers’ will-
ingness to pay for food labeled ‘genetically modified’. Agricultural and Resource
Economics Review, 35(2):299.
Paggi, M. S., Yamazaki, F., Ribera, L., Palma, M., and Knutson, R. (2013). Do-
mestic and trade implications of leafy green marketing agreement type policies
and the food safety modernization act for the southern produce industry. Journal
of Agricultural and Applied Economics, 45(3):453–64.
Painter, John A, e. a. (2013). Attribution of foodborne illnesses, hospitalizations,
and deaths to food commodities by using outbreak data, united states, 1998–2008.
Emerging Infectious Disease, 19(3):407–415.
Petrin, A. (2002). Quantifying the benefits of new products: The case of the minivan.
Journal of Political Economy, 110(4):705–729.
R Core Team (2016). R: A Language and Environment for Statistical Computing.
R Foundation for Statistical Computing, Vienna, Austria.
Revolution Analytics and Weston, S. (2015). doParallel: Foreach Parallel Adaptor
for the ’parallel’ Package. R package version 1.0.10.
Reynaerts, J., Varadhan, R., and Nash, J. C. (2012). The Convergence Properties of
the BLP (1995) Contraction Mapping and Alternative Algorithms in R. VIVES
Discussion Paper, (35).
Ribera, L., Knutson, R., Paggi, M., Yamazaki, F., Palma, M., and Anciso, J.
(2014). Impacts of evolving food safety standards on the sustainability of small-
and medium-size produce farms: Implications for implementation of the fda food
safety modernization act: Final report. Technical Report USDA-NIFA Award No.
2012-68006-30187.
Ribera, L. A. and Knutson, R. D. (2011). The fda’s food safety modernization act
and its economic implications. Choices, 26(4):16–21.
Ribera, L. A., Palma, M. A., Paggi, M., Knutson, R., Masabni, J. G., and Anciso, J.
(2012). Economic analysis of food safety compliance costs and foodborne illness
outbreaks in the united states. HortTechnology, 22(2):150–156.
Ribera, L. A., Yamazaki, F., Paggi, M., Seale Jr, J. L., et al. (2016). The food safety
modernization act and production of specialty crops. Choices, 31(1).
Richards, T. J. and Hamilton, S. F. (2015). Variety pass-through: An examination
of the ready-to-eat breakfast cereal market. Review of Economics and Statistics,
97(1):166–180.
Richards, T. J. and Patterson, P. M. (2006). Firm-level competition in price and
variety. Journal of Agricultural and Applied Economics, 38:49l–5l2.
188
Roe, B. and Sheldon, I. (2007). Credence good labeling: The efficiency and distribu-
tional implications of several policy approaches. American Journal of Agricultural
Economics, 89(4):1020–1033.
Roe, B. and Teisl, M. F. (2007). Genetically modified food labeling: The impacts
of message and messenger on consumer perceptions of labels and products. Food
Policy, 32(1):49–66.
Rouvie`re, E. and Caswell, J. A. (2012). From punishment to prevention: A french
case study of the introduction of co-regulation in enforcing food safety. Food
Policy, 37(3):246–254.
Scherer, F. M. (1979). The welfare economics of product variety: an application
to the ready-to-eat cereals industry. The Journal of Industrial Economics, pages
113–134.
Schmalensee, R. (1978). Entry Deterrence in the Ready-to-Eat Breakfast Cereal
Industry. The Bell Journal of Economics, 9(2):305–327.
Segerson, K. (1999). Mandatory vs. voluntary approaches to food safety. Agribusi-
ness, 15(1):53–70.
Stone, R. (1954). Linear expenditure systems and demand analysis: an application
to the pattern of british demand. The Economic Journal, pages 511–527.
Thomas, L. A. (1999). Incumbent firms’ response to entry: Price, advertising,
and new product introduction. International Journal of Industrial Organization,
17(4):527–555.
Thompson, E., Harper, A. M., and Kraus, S. (2008). Think globally–eat locally:
San francisco foodshed assessment. American Farmland Trust, 23.
Unnevehr, L. J. and Jensen, H. H. (1999). The economic implications of using haccp
as a food safety regulatory standard. Food Policy, 24(6):625–635.
Varadhan, R. (2010). Squarem: Squared extrapolation methods for acceler-
ating fixed-point iterations. http://CRAN.R-project.org/package=SQUAREM,
Rpackageversion. Accessed: 2017-04-28.
Viscusi, W. K. (1978). A Note on ”Lemons” Markets with Quality Certification.
The Bell Journal of Economics, 9(1):277–279.
Yiridoe, E. K., Bonti-Ankomah, S., and Martin, R. C. (2005). Comparison of con-
sumer perceptions and preference toward organic versus conventionally produced
foods: a review and update of the literature. Renewable Agriculture and Food
Systems, 20(04):193–205.
189