THE EFFECTS OF INFORMATION DISCLOSURE AND LABELING ON DEMAND FOR PLANT-BASED MEAT ALTERNATIVES: EVIDENCE FROM RETAIL SCANNER DATA A Thesis Presented to the Faculty of the Graduate School of Cornell University in Partial Fulfillment of the Requirements for the Degree of Masters of Science by Haoran Wang August 2024 © 2024 Haoran Wang ALL RIGHTS RESERVED ABSTRACT This study examines the impact of different labeling claims and information provision on the price, total sales, and quantity sold of plant-based meat alternatives (PBMA) prod- ucts. Specifically, we analyze how specific labeling terms and certifications interact with information about the products’ environmental, health, and ethical claims to affect con- sumer purchase decisions. The findings indicate that “plant-based” and “protein” labels are associated with higher demand and sales, while “meat substitute” labels correlate with higher prices but lower sales volume. Using a structural demand model, I estimate the different effects of attributes on demand and find that low-calorie claims and meat substitutes have the strongest positive impact on product choice. In contrast, protein content has a significant negative impact on choice. The random coefficients logit model refined our previous demand estimations. Additionally, we estimate the cross-price elasticities and own-price elasticities of the plant- based products, as well as related metrics such as marginal cost, markup ratio, and diversion ratios. These results have important implications for current and proposed PBMA market- ing policies and labeling legislation. Specifically, restricting the use of the word “meat” in plant-based product labeling could significantly impact the demand for these prod- ucts. By understanding how various labeling claims and information provision influence consumer behavior, policymakers and marketers can make more informed decisions that support the growth and acceptance of plant-based meat alternatives. BIOGRAPHICAL SKETCH Haoran Wang was born in Xinjiang, China. He obtained a Bachelor’s degree in Elec- tronic Information Engineering and Economics, as well as a Master’s degree in Computer Science. To further his research in economics, Haoran entered the Master of Science pro- gram in Applied Economics and Management at Cornell University. Thanks to all the help and support from his cohort, friends, and faculty members, Haoran will continue his Ph.D. journey. iii This document is dedicated to all Cornell graduate students. iv ACKNOWLEDGEMENTS The analysis, findings, and conclusions expressed in this report should not be at- tributed to Circana. First, I want to express my sincere gratitude to Dr. Aaron Adalja for his kind help and strong support. His intelligence, creativity, and dedication to his work set a high standard not only for an outstanding scholar but also as a mentor and supervisor. His integrity, approachability, and many other positive qualities make him a role model for me. There are always new things I can learn from him. His diverse and remarkable experiences also illustrate how inspiring a person’s journey can be. Next, I would like to thank my research assistant supervisor, Dr. Sylvia Hristakeva, for her strong support and for always considering my best interests. She opened new doors for me during my academic journey. I am also grateful to Dr. Heather Schofield for her unending inspiration and passion for research and her care for students. I am deeply grateful to Dr. Anne Byrne and Dr. Bradley Rickard for their invaluable guidance, support, and advice during my pursuit of a master’s degree. I also thank Dr. C.-Y. Cynthia Lin Lawell, who was always concerned about my application status and provided detailed guidance and encouragement during office hours. My thanks also go to Dr. David Just for his amazing research assistant opportunities and the food pantry program, which allowed me to engage with and support the local community in upstate New York. Words cannot fully express my gratitude to Cornell and its faculty. Lastly, I am thankful to my family, the ones I love and who love me, my friends, and my cohort for their endless support. v TABLE OF CONTENTS Biographical Sketch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii 1 Introduction 1 2 Literature Review 5 2.1 Information Interventions and Consumer Behavior . . . . . . . . . . . . . 5 2.2 Market Demand and Consumption Behavior . . . . . . . . . . . . . . . . 5 2.3 Labelling Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 3 Data 7 3.1 Data Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.2 Selection of Plant Based Meat Categories . . . . . . . . . . . . . . . . . . 8 3.3 Key Variables and Summary Tables . . . . . . . . . . . . . . . . . . . . . 8 3.4 Market Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4 Methodology 12 4.1 Hedonic Regression Models . . . . . . . . . . . . . . . . . . . . . . . . . . 12 4.2 Structural Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.2.1 Logit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.2.2 Random Coefficients Logit Model . . . . . . . . . . . . . . . . . . 15 5 Results 18 5.1 Hedonic Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . 18 5.1.1 Price Per Ounce . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 5.1.2 Total Sales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 5.1.3 Quantity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 5.1.4 The Role of Nutrition Content . . . . . . . . . . . . . . . . . . . . 22 5.2 Logit Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 5.3 Random Coefficients Logit Results . . . . . . . . . . . . . . . . . . . . . 25 5.3.1 Market Structure Assumptions . . . . . . . . . . . . . . . . . . . 25 5.3.2 Demand Parameters . . . . . . . . . . . . . . . . . . . . . . . . . 26 5.3.3 Marginal Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 5.3.4 Price Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 5.3.5 Markups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 5.3.6 Market Concentration . . . . . . . . . . . . . . . . . . . . . . . . 31 6 Conclusions 33 A Appendix 35 vi LIST OF TABLES 3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 5.1 Linear Estimates of the Impact of Label Claims . . . . . . . . . . . . . . 19 5.2 Log-Linear Estimates of the Impact of Label Claims . . . . . . . . . . . 20 5.3 Log-Log Estimates of the Impact of Label Claims and Nutrition Content 23 5.4 Logit Results Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 5.5 Beta Estimates (Robust SEs in Parentheses) . . . . . . . . . . . . . . . . 26 5.6 Nonlinear Coefficient Estimates (Robust SEs in Parentheses) . . . . . . 27 A.1 The Impact of Label Claims and Nutrition Content . . . . . . . . . . . . 35 A.2 Linear Estimates of the Impact of Nutrition Content . . . . . . . . . . . 36 A.3 Log-Linear Estimates of the Impact of Nutrition Content . . . . . . . . . 37 vii LIST OF FIGURES 5.1 Marginal Cost (dollars) . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 5.2 Mean Own Elasticities and Aggregate Elasticities . . . . . . . . . . . . . 30 5.3 Markups, % . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 5.4 Diversion Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 viii CHAPTER 1 INTRODUCTION Vegan diets have a long history which could be traced back thousands of years (Ruby, 2012). In recent years, Plant-Based Meat Alternatives (PBMA) have emerged as a new category of food, attracting increasing global attention. These products are esteemed for their minimal environmental impact, health benefits, and considerations for animal welfare (He et al., 2020). In Western countries, although meat is considered a source of high-quality protein, it is often deemed unsustainable. Processed meats in traditional diets, such as hot dogs, bacon, and sausages, have been classified as carcinogenic by the International Agency for Research on Cancer of the World Health Organization, while unprocessed red meats, such as beef and pork, are categorized as ”probably carcinogenic.” Additionally, there is growing concern about severe water pollution caused by industrial meat production (Hu et al., 2019). Keoleian and Heller (2018) find that plant-based meat alternatives have a much smaller environmental impact. Consequently, people are gradually adopting diets that reduce meat consumption and therefore seeking out delicious and healthy meat alternatives. In line with this trend, the market is shifting towards plant proteins, such as those derived from beans, wheat gluten, and soy protein, to produce plant-based meats. These products closely resemble real meat in terms of texture, flavor, color, and nutritional characteristics (Kyriakopoulou et al., 2019). However, the rapid growth of the plant-based food industry has sparked concerns within the traditional meat and dairy sectors about potential consumer confusion over plant-based products being labeled as familiar meat or dairy items (Gleckel, 2020). This has led to a surge of proposed and enacted legislation at both state and federal levels aimed at regulating the labeling of plant-based foods. The legal landscape is evolving quickly, with courts striking down some laws while others remain pending (Taylor, 2020). The U.S. Food and Drug Administration (FDA) oversees the production and labeling of plant- 1 based foods but has yet to issue specific regulations or formal guidance for plant-based alternatives to animal-derived ingredients. The FDA has generally exercised enforcement discretion in this area. In 2018, the U.S. Department of Agriculture (USDA) received a petition from the U.S. Cattlemen’s Association (2018) to limit the definitions of “beef” and “meat” to products derived from animals. This petition was denied by the USDA in 2021, citing jurisdictional limitations under the Federal Meat Inspection Act and the Poultry Products Inspection Act. In the absence of clear federal regulations, various states have implemented their own labeling laws to address plant-based food products. These laws vary significantly; some states allow qualified labels with terms like “plant-based” or “vegan” prominently displayed, while others, like Arkansas, prohibit any language associated with meat-based products. Missouri has even criminalized labeling plant-based foods as meat, making violations a Class A misdemeanor (Turtle Island Foods, 2024). These state laws have faced legal challenges, primarily on First Amendment grounds. For example, in Arkansas, a preliminary injunction was granted to Turtle Island Foods, preventing enforcement of the state’s labeling law against the company. In contrast, a court in Oklahoma upheld labeling requirements for plant-based foods, asserting that these requirements were reasonably related to protecting consumers from confusion. Mis- souri’s law was upheld on appeal, as the court found it did not apply to Tofurky’s labels. In California, the Department of Food and Agriculture challenged Miyoko’s Kitchen’s use of the term “butter” for its vegan product, leading to a legal dispute over free speech rights, where Miyoko’s partially won on summary judgment (Fund, 2023). Our study significantly extends the existing literature by simultaneously considering the effects of labeling and information disclosure on consumer choices regarding plant- based products. Unlike previous studies that have primarily focused on either aspect in isolation, our research integrates these two interdependent factors. We first analyze how the labeling of plant-based products, including the use of specific terms and certifications, interacts with the disclosure of information about these products’ environmental, health, 2 and ethical benefits. We then estimate structural demand models to recover own-price and cross-price elasticities for the plant-based products. We outline several key hypotheses that guide our analysis of the plant-based product market. • H1 (Functional Label Appeal): Labels with functional orientation, such as “meat substitute,” are more attractive to consumers than descriptive labels (e.g., ”plant-based” or ”veggie”) because they directly convey the product’s use, reduc- ing cognitive load for consumers. This hypothesis is based on the theory of cogni- tive load (Zimmerman and Shimoga, 2014), which suggests that consumers prefer straightforward, functional information that simplifies decision-making. Addition- ally, functional labels can provide reassurance about the product’s role as a direct substitute for traditional options, which can further enhance its appeal. (Eden, 2011) • H2 (Sensory Expectation): Labels like “plant-based” and “protein” might trig- ger negative sensory expectations, with consumers associating these terms with poor taste or over-processing, thus reducing product appeal. This hypothesis is based on what Piqueras-Fiszman and Spence (2015) researched: names and images can shape consumer expectations and reactions to food. If the name or description sets very high expectations, the food must meet them; otherwise, the consumer may experience negative disconfirmation. • H3 (Processing Perception): Emphasizing certain attributes (e.g., high protein content) might lead consumers to perceive the product as unnatural or overly pro- cessed, especially under the “clean label” trend. Inguglia et al. (2023) show that consumers react negatively to food labels that suggest heavy processing. An arti- cle in MDPI highlights that the concept of a “clean label” is driven by consumer demand for transparency and natural ingredients. • H4 (Health Component): Consumers prefer products with high protein content but dislike those with high calorie and fat content. Consumer trends indicate a 3 growing demand for high-protein products, reflecting a preference for healthier food choices (FMCG Gurus, 2020). • H5 (Low Marginal Cost): The marginal cost in the plant-based product market is relatively low, potentially due to lower raw material costs, economies of scale, or advancements in production technology. • H6 (Differentiated Price Sensitivity): While demand for individual plant- based meat products is highly sensitive to price changes, the overall demand elas- ticity for the plant-based product market is low, indicating that consumers are very sensitive to price differences between specific products but less influenced by price when deciding to choose plant-based products as a category. The rest of the paper proceeds as follows. We provide brief review of relevant literature in chapter 2. We then describe the data and key variables in chapter 3. We present our empirical methodologies in chapter 4, and present the results in chapter 5. At last, we close with a discussion of implications and directions for future research in chapter 6. 4 CHAPTER 2 LITERATURE REVIEW 2.1 Information Interventions and Consumer Behavior Katare et al. (2023) conducted a randomized controlled trial and found that information on environmental and health impacts significantly influenced demand for sustainably produced beef and plant-based meat substitutes. Present-biased females paid less for sustainable beef, while future-biased individuals were more willing to pay a premium for plant-based meat. Segovia et al. (2023) found that health information nudges motivated meat eaters to buy plant-based alternatives, whereas environmental information did not. Meat eaters viewed meat as healthier but less environmentally friendly compared to plant-based options. 2.2 Market Demand and Consumption Behavior Van Loo et al. (2020) found that, with constant prices, traditional beef burgers domi- nate the market share compared to lab-grown meat and two types of plant-based meats, indicating limited consumer acceptance of non-traditional meats. Ortega et al. (2022) demonstrated that identity labels reduce demand for traditional meat and increase de- mand for plant-based and cultured meat alternatives, indicating the importance of labels and identity in consumer choice in the Chinese market. Zhao et al. (2023) analyzed mar- ket data and found PBMA complements beef and pork but substitutes chicken, turkey, and fish. Despite lower overall demand, PBMA sales and market share are growing sig- nificantly, with education, income, and location being key factors. Neuhofer and Lusk (2022) observed that households buying PBMA products did so multiple times without reducing meat purchases. DeMuth et al. (2023) found that over 30% of consumers could not distinguish between traditional and non-traditional meats. Labeling restrictions did 5 not reduce confusion or affect substitution between products. Zhao et al. (2023) noted sig- nificant growth in PBMA sales, appealing beyond traditional vegetarians. Their findings offer insights into future marketing strategies and policies. 2.3 Labelling Effect Ericsson and Kintsch (1995) suggests that cognitive processes are seen as a sequence of stages dealing with the final product of stable states. In skilled activities, acquired memory skills allow these final products to be stored in long-term memory and accessed directly through retrieval cues in short-term memory. Spronk et al. (2014) notes that nutrition knowledge is a fundamental component of health literacy. DeMuth et al. (2023) found that changing labels does not significantly reduce misconceptions about the ingre- dients and nutritional content of non-traditional meat substitutes. Nelson (1970) found that consumers often make choices with limited information about prices and product quality, which significantly impacts market structure. This lack of information can lead to suboptimal decisions and market inefficiencies. Campos et al. (2011) indicates that nutrition labels on pre-packaged foods are widely regarded as reliable and are frequently used by consumers to make healthier food choices. However, the effectiveness of these labels can vary across different demographic groups. Adalja (2022) utilized a structural demand model for the ready-to-eat (RTE) cereal industry, revealing that non-GMO labels positively impact demand despite varied consumer tastes. This suggests that voluntary quality certifications can serve as an effective non-price marketing strategy. These studies collectively highlight the significant role that labeling and information disclosure play in consumer decision-making and market dynamics. Labels not only serve as important tools for conveying product quality and nutritional information but also influence consumer preferences and behaviors, thereby shaping market outcomes. 6 CHAPTER 3 DATA 3.1 Data Source InfoScan collects weekly UPC retail sales data (sales and quantity) from various retail establishments across the U.S. and Puerto Rico. Some retailers provide data at the store level, capturing sales data for a specific location, while others provide data at the Retail Market Area (RMA) level, which aggregates geographic areas defined by each retailer. InfoScan maintains comprehensive product dictionaries that offer detailed information about all products in both the Consumer Network and InfoScan, which is also the basis we relied on to filter plant-based products. These dictionaries include UPC-coded product descriptions, such as flavor and style, nutritional information from the back of the package, and health and wellness claims from the front of the package. InfoScan data encompasses retail establishments across the U.S. and Puerto Rico, providing weekly UPC retail sales data, including sales and quantity, from various store types such as grocery stores and supermarkets, at both store and RMA levels. The data we used includes both store-level and Retail Market Area (RMA) level information, encompassing different types and numbers of plant-based meat alternatives’ UPCs due to the varied types of retailers. The same product with different labels across years would also have different UPCs. For example, a plant-based product that changes its keywords on the label or modifies its nutrition claims would have a different UPC in subsequent years. Product descriptions are detailed in product dictionaries across InfoScan, covering aspects such as flavor (e.g., BBQ), style, nutritional information (e.g., different vitamins, proteins), and claims (e.g., whether it is high or low in fat, whether it is organic). Store information, including physical addresses, has been provided by InfoScan. This information is crucial for us, especially since most of the plant-based meat alternatives we study come from big retailers. The addresses are used to calculate market size by 7 aggregating the data by zip code, with geocoded addresses enabling linkage to InfoScan data. Although IRI does not specify store locations visited by households, mappings to retailer IDs allow us to link transaction data based on proximity to store locations. 3.2 Selection of Plant Based Meat Categories During the data processing phase, we implemented a series of filtering operations to ex- clude products that do not align with our research focus on plant-based meat alternatives. Specifically, we used the following criteria to drop irrelevant categories and products: dropped categories unrelated to plant-based substitutes, such as “COFFEE,” “CORN ON THE COB - FZ,” “ENGLISH MUFFINS,” “FRESH BREAD ROLLS,” and “NAT- URAL CHEESE;” excluded brands that do not belong to the top 16 plant-based brands identified above; removed product categories that include complex plant-based prod- ucts which might confound our analysis, such as “BAKING MIXES,” “BREAKFAST FOOD - FZ,” “APPETIZERS/SNACK ROLLS - FZ,” and “PASTRY/DOUGHNUTS;” and dropped additional specific products like “burrito,” “noodle,” and “pad thai” to maintain the focus on primary plant-based meat alternatives. By refining our dataset in this manner, we aimed to mitigate the influence of extraneous variables and concentrate on the core plant-based products relevant to our study, ensuring that the findings are attributable to the specific plant-based brands under investigation. 3.3 Key Variables and Summary Tables We utilized InfoScan retail scanner data to obtain detailed claims information and de- scriptions for each product. To precisely capture the characteristics of PBMA products, we extracted relevant keywords from the product descriptions, including “plant-based” or “veggie,” “meat substitute,” and “protein.” We then conducted a correlation analysis to ensure that these variables did not exhibit severe multicollinearity. 8 In the original datasets, we extracted “veggie” and “plant-based” as keywords. How- ever, we found a high correlation between “veggie” and “plant-based,” with overlapping meanings in product descriptions. Therefore, we combined them into a new variable. In- terestingly, many plant-based meat alternatives describe themselves as “chicken,” while very few describe themselves as “beef.” This may be influenced by related state legisla- tion, so we extracted “chicken” as a separate dummy variable. Additionally, we selected “low calorie,” “low fat,” and “organic” as claim variables from the InfoScan data. For “organic,” we cleaned the data to only define products as organic if they are 95% or more organic. Keywords and information usually appear on the front side of the product labels, while most of the nutritional information is displayed on the back. However, the protein, calorie, and fat content are sometimes highlighted on the front side. This front-side information tends to be what captures consumers’ attention most. In our analysis, we also focus on the front-side label information since it has the most impact on consumers at first glance. Table 3.1: Summary Statistics Variable Obs Mean Std. Dev. Min Max meat sub 1,692,925 0.730691 0.443609 0 1 plant based or veggie 1,692,925 0.5746291 0.4943993 0 1 protein 1,692,925 0.154086 0.3614013 0 1 protein per upc 1,692,925 52.82799 26.44629 10 288 fat per upc 1,692,925 23.05426 14.80269 0 200 calories per upc 1,692,909 575.9552 248.7366 160 4560 total sales($) 1,692,925 797.9848 1727.304 0.01 104992.3 price per ounce($) 1,692,925 0.4231709 0.1192561 0.00125 3.768462 total quantity(ounces) 1,692,925 2082.528 4971.306 0.08 324101.1 organic 1,692,925 0.0451798 0.2076983 0 1 less calorie claim 1,692,925 0.0001548 0.0124264 0 1 less fat claim 1,692,925 0.385186 0.4865301 0 1 chicken 1,692,925 0.0458372 0.2086038 0 1 Notes: This table provides summary statistics for the variables used in the analysis. The statistics include number of observations (Obs), mean, standard deviation (Std. Dev.), minimum (Min), and maximum (Max). Table 3.1 presents the descriptive statistics of the main variables. The average price 9 per ounce is $0.42, with a maximum value of $3.77. In contrast, the average price for dry-aged beef is about $45 per pound, which translates to approximately $2.80 per ounce. On the other hand, from U.S. Department of Agriculture (2023) the average cost for Beef Patties from the grocery store is approximately $6.84 per pound when on sale, or about $0.50 per ounce. Therefore, the price range of plant-based meat alternatives is generally consistent with that of traditional meat. In our analysis, the dependent variables of interest include total sales, which represents the total sales of a specific UPC (product) within a particular Retail Market Area (RMA) for a given week, the total volume of ounces sold (the quantity variable indicates the total ounces sold per week for each product) for an RMA in a specific week, and the price per ounce derived from the available data. The observation level of the dataset is at the UPC-RMA-week level. This means that each observation in the dataset corresponds to a unique combination of a specific product (UPC), a specific retail market area (RMA), and a specific week. This granularity allows us to analyze the sales performance and pricing dynamics of plant-based meat alternatives across different regions and time periods. 3.4 Market Size Accurately estimating market size is critical for our structural demand estimation. Ac- cording to Nevo (2000), it is important to define a market size that is sufficiently large to prevent negative outside market shares. They also suggested using a proportional factor related to the market size, and a constant factor that can be estimated (Berry, 1994). For our analysis, we use the population size as the proportional factor. We focus exclu- sively on plant-based foods as potential substitutes, assuming the potential market size for plant-based meat alternatives corresponds to the protein demand from animal-based sources in the U.S. market. Consequently, our outside shares reflect the portion of pro- tein demand from animal-based sources not satisfied by the top sixteen PBMA brands. Eshel et al. (2019) noted that plant-based substitutes need to supply the approximately 10 30 grams of protein per day that Americans currently get from beef, pork, and poultry (out of the total masses of 70, 30, and 74 grams of meat per day, respectively). In the scanner data, the amount of protein per serving is provided as a variable. Thus, we calculate the protein consumed from PBMA using the weekly transaction data. This amount was then divided by the market size, defined as 30 grams of protein per day times 7 days a week times the population size of that RMA. We use the calculated market shares for the structural demand estimation. 11 CHAPTER 4 METHODOLOGY 4.1 Hedonic Regression Models Hedonic regression is a statistical method used to analyze the relationship between the price of a product or service and its various characteristics. This method is based on the assumption that the price of a product can be decomposed into the sum of the values of its individual attributes or features. In this study, we use InfoScan retail scanner data from 2012 to 2021, aggregated to the weekly level for each RMA, to analysis the sales- weighted price per ounce (Pricejkrt), total quantity sold in ounces (Quantityjkrt), and total sales (Total salesjkrt). Here, j represents the product UPC, k is the manufacturer (since the product names and firm names for plant-based products are often highly consistent, we use firm ID here), r represents the Retail Market Area (RMA), and t represents a particular week. For simplicity, we use Y to stand for the three dependent variables we care about: the price per ounce, the quantity (of the total ounces sold), and the total sales. Yjkrt = β1 ·Meat substitutejkrt + β2 · Plant-based or Veggiejkrt + β3 · Proteinjkrt + β4 ·Organicjkrt + β5 · Low caloriejkrt + β6 · Low fatjkrt + β7 · Chickenjkrt + ϵjkrt (4.1) Log(Yjkrt) = β1 ·Meat substitutejkrt + β2 · Plant-based or Veggiejkrt + β3 · Proteinjkrt + β4 ·Organicjkrt + β5 · Low caloriejkrt + β6 · Low fatjkrt + β7 · Chickenjkrt + ϵjkrt (4.2) By utilizing specific product information of plant-based meat alternatives, we con- 12 struct multiple dummy variables based on different claims and key information of the products to estimate the average impact of labels on the prices of plant-based products. The independent variables include several product characteristics such as whether the product is a meat substitute, plant-based or veggie, high in protein, organic, low calorie, low fat, or chicken. The coefficients (β1, β2, . . . , β7) capture the impact of each of these characteristics on the dependent variables. To control for unobserved heterogeneity, we absorb fixed effects for year, RMA, and manufacturer. By including these fixed effects, we account for time-invariant factors specific to each year, regional characteristics specific to each RMA, and unchanging at- tributes specific to each manufacturer. This approach helps isolate the effect of the product characteristics on the dependent variables, providing more robust and reliable estimates. Considering the plant-based meat alternatives context, which includes the rise in market share of plant-based meat alternatives in recent years, differences in brand awareness and target audiences of various products, and varying dietary preferences across different regions, these fixed effects are particularly important. The fixed effects for the year capture the overall market trends and growth in the plant-based meat sector. The RMA fixed effects account for regional dietary preferences and other local factors, while the manufacturer fixed effects control for differences in brand strength and product po- sitioning among different companies. The error term ϵjkrt captures all other factors that affect the dependent variables but are not included in the model. 4.2 Structural Estimation 4.2.1 Logit Model Demand estimation typically involves a choice among a set of available alternatives (such as choosing which product to buy or selecting among competing products). In these scenarios, our dependent variable is usually discrete, making traditional linear regression 13 models inadequate. Consequently, discrete choice models (DCMs) have emerged as a solution. Luce (1959) derived the Logit model based on utility maximization theory, enabling rational consumers to choose the option that maximizes their utility from the available alternatives. We first estimate a logit model under the independence of irrelevant alternatives (IIA). Our indirect utility is given by Uijrt = αpjrt + xjrtβ ex + ξjrt + ϵijrt, (4.3) The utility model is represented as follows: Uijrt = αpjt + xjtβ ex + ξjt + ϵijrt, where Uijrt represents the utility that consumer i derives from product j at time t retailer marketing area r; α is the price coefficient, which measures the impact of price pjrt on utility; pjrt is the price of product j at time tretailer marketing area r; xjrt is a vector of other attributes of product j at time t retailer marketing area r; βex is the vector of coefficients for product attributes, which measures the impact of each product attribute xjrt on utility; ξjrt is the unobserved product-specific utility for product j at time t (also known as product fixed effects); ϵijrt is the error term, representing unobserved factors affecting the utility of consumer i for product j in retailer marketing area r at time t , and follows a Type I Extreme Value (Gumbel) distribution. The basic assumption of this model is that consumers choose the product that maximizes their utility. The coefficients α and βex are estimated to measure the impact of price and other product attributes on consumer choice. We normalize the mean utility of the outside good to zero so that Ui0tr = ϵi0tr, which generates the market share given by: sjrt = exp(αpjrt + xjrtβ ex + ξjrt) 1 + ∑ k exp(αpktr + xktrβex + ξktr) . (4.4) Taking logs, we get 14 log sjrt = αpjrt + xjrtβ ex + ξjrt − log ∑ k exp(αpktr + xktrβ ex + ξktr) (4.5) and log s0tr = − log ∑ k exp(αpktr + xktrβ ex + ξktr). (4.6) By differencing the above equations, we obtain a linear estimating equation: log sjrt − log s0tr = αpjrt + xjrtβ ex + ξjrt. (4.7) 4.2.2 Random Coefficients Logit Model The basic Logit model assumes that all consumers have identical preferences for the characteristics of alternative products. However, in the real world, consumers often choose products with asymmetric cross-price elasticities. Based on the research by Berry (1994) and Berry et al. (1995), the random coefficients Logit model was introduced for demand estimation to fully consider the different preferences of consumers for different products. Consumers with similar preferences have similar indirect utility functions, and the higher the similarity of alternative products, the greater their substitutability. If we still assume that ϵijrt follows a Type I Extreme Value distribution, then the market share of product j in market t is given by: sjt(δjt, θ) = ∫ exp(δjt + µijrt) 1 + ∑ k exp(δkt + µikt) f(µit|θ)dµit (4.8) Given an initial estimate of θ, we can solve the nonlinear equations system to obtain the vector δ, which aligns the observed and predicted market shares, such that sjt = sjt(δ, θ). Subsequently, we can execute a linear IV GMM regression represented as: 15 δjt(θ) = αpjt + xjtβ ex + ξjt. (4.9) To construct the moments, we interact the predicted residuals ξ̂jt(θ) with the instru- ments zjt, resulting in: ḡ(θ) = 1 N ∑ j,t zjtξ̂jt(θ). (4.10) In the random coefficients logit model, the correlation between choices can be cap- tured. Therefore, the correlation between choices is a function of product and consumer characteristics. Consumers with similar characteristics have similar product preference rankings, and thus similar substitution patterns, solving the problem of the independence of irrelevant alternatives (IIA) assumption. The price elasticity of market share can be expressed as: ηjk = ∂sjt ∂pkt · pkt sjt = αsjt(δjk − skt)dP ∗(D)dP v(ν)dP ϵ(ϵ) (4.11) In this equation, ηjk represents the elasticity of the market share of product j with respect to the price of product k. In other words, it reflects the sensitivity of the market share of product j to changes in the price of product k. The first part of the equation, ∂sjt ∂pkt · pkt sjt , calculates the rate of change in market share sjt with respect to price pkt, multiplied by the ratio of price to market share. The parameter α in the equation measures overall price sensitivity. The indicator function δjk distinguishes between own-price elasticity (when j = k) and cross-price elasticity (when j ̸= k). By integrating over the probability distributions dP ∗(D), dP ν(ν), and dP ϵ(ϵ), we account for the distribution and heterogeneity of consumer preferences. In our model, we identify each market by the RMA and weekly transactions, specify- ing the market shares to be between 0 and 1 within a particular market, treating prices as endogenous. Using PyBLP (Conlon and Gortmaker, 2020), we generate demand instru- ments. We also assume the covariance matrix of the coefficients to be diagonal, meaning 16 that the random coefficients for different characteristics are independent. For instance, the randomness in price preferences is assumed to be unrelated to the randomness in brand attributes. By using a diagonal matrix, we simplify the model structure and re- duce the number of parameters that need to be estimated. For simplicity, we simulate 50 individuals from a random normal distribution for Monte Carlo draws. 17 CHAPTER 5 RESULTS 5.1 Hedonic Regression Results In Tables 5.1 and 5.2, Column 1, we present the impact of label claims on prices. The variables include plant-based or veggie, protein, and meat substitute, extracted from product descriptions, as well as health claim variables such as organic, low calorie, and low fat. All variables are binary. 5.1.1 Price Per Ounce The linear regression results in Table 5.1 show that products described as Meat substitute and Chicken have higher prices, with Meat substitute increasing the price per ounce by $0.04 and Chicken increasing it by $0.03, both highly statistically significant. This could be because consumers attach a premium to products that position themselves explicitly as a substitute traditional meat. Conversely, products described as Plant-based or Veggie have lower prices, reducing the price per ounce by $0.02. Products with a Protein claim reduce the price by $0.03 per ounce, Organic products by $0.01, Low calorie products by $0.10, and Low fat products by $0.01, all results being statistically significant at the 0.01 level. Although surprising, the Organic label’s negative effect on price might be explained by several factors. Firstly, consumers might perceive organic plant-based products differently than organic traditional meat products, possibly expecting lower prices due to the nature of plant-based ingredients. Secondly, the organic label could be more common among plant-based products, thus reducing its ability to command a premium price. In Table 5.2, the log-linear regression results show the impact of dummy variables changing from 0 to 1 on prices: products described as Meat substitute increase prices by 18 approximately 9% (e0.09 − 1), and Chicken increases prices by about 8% (e0.08 − 1), both highly significant. Products described as Plant-based or Veggie reduce prices by about 6% (e−0.06 − 1), products with a Protein claim reduce prices by about 7% (e−0.07 − 1), Organic products by about 1% (e−0.01−1), Low calorie products by about 2% (e−0.21−1), and Low fat products by about 2% (e−0.02 − 1), all results being statistically significant at the 0.01 level. The R-squared values for these two models are 0.51 and 0.47, respectively, indicating that the models have a high explanatory power for prices, with the linear regression model having slightly higher explanatory power than the log-linear regression model. Table 5.1: Linear Estimates of the Impact of Label Claims Price per ounce Total sales Quantity Meat substitute 0.04*** -94.44*** -411.58*** (0.00) (4.32) (12.92) Plant-based or Veggie -0.02*** 70.03*** 209.83*** 0(.00) (4.13) (12.32) Protein -0.03*** 134.05*** 428.91*** (0.00) (4.71) (14.07) Organic -0.01*** -178.96*** -544.20*** (0.00) (5.46) (16.30) Low calorie -0.10*** -356.99*** -529.17*** (0.00) (70.81) (211.46) Low fat -0.01*** 182.8*** 470.21*** (0.00) (2.83) (8.44) Chicken 0.03*** 9.00*** -183.49*** (0.00) (5.53) (16.53) Constant 0.41*** 1958.71*** -183.49*** (0.00) (2.95) (8.80) R-squared 0.51 0.53 0.54 Observations 1,521,303 1,521,303 1,521,303 Notes: All columns control for company and year fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. 5.1.2 Total Sales In Tables 5.1 and 5.2, Column 2, we present the impact of label claims on total sales. The linear regression results in Table 5.1 show that products described as Meat substitute and Chicken have lower total sales, with Meat substitute reducing total sales by $94.44 highly statistically significant. Conversely, products described as Plant-based or Veggie and those with a Protein claim and chicken increase total sales, with Plant-based or 19 Table 5.2: Log-Linear Estimates of the Impact of Label Claims Log(price per ounce) Log(total sales) Log(quantity) Meat substitute 0.09*** -0.09*** -0.18*** (0.00) (0.00) (0.00) Plant-based or Veggie -0.06*** 0.001*** 0.06*** (0.00) (0.00) (0.00) Protein -0.07*** 0.15*** 0.21*** (0.00) (0.00) (0.01) Organic -0.01*** -0.31*** -0.30*** (0.00) (0.00) (0.01) Low calorie -0.21*** -1.56*** -1.35*** (0.01) (0.75) (0.08) Low fat -0.02*** 0.18*** 0.19*** (0.00) (0.00) (0.00) Chicken 0.08*** 0.07*** -0.01*** (0.00) (0.01) (0.01) Constant -0.91*** 5.38*** 6.29*** (0.00) (0.00) (0.00) R-squared 0.47 0.55 0.55 Observations 1,521,303 1,521,303 1,521,303 Notes: All columns control for company and year fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Veggie increasing total sales by $70.03 and Protein by $134.05, both highly significant. Organic, Low calorie, and Low fat claims have a negative impact on total sales, with Organic reducing total sales by $178.96, Low calorie by $356.99, and Low fat by $182.80, all highly statistically significant. The log-linear regression results in Table 5.2 show similar patterns. Products de- scribed as Meat substitute have lower total sales, with Meat substitute reducing total sales by approximately 9% (e−0.09 − 1) and Chicken has higher total sales, increasing it by about 7% (e−0.07 − 1). Conversely, products described as Plant-based or Veggie and those with a Protein claim increase total sales, with Plant-based or Veggie increasing total sales by approximately 0.1% (e0.001 − 1) and Protein increasing it by about 15% (e0.15 − 1). Organic, Low calorie, and Low fat claims have a negative impact on total sales, with Organic reducing total sales by approximately 31% (e−0.31 − 1), Low calorie by about 156% (e−1.56 − 1), and Low fat increasing total sales by about 18% (e0.18 − 1). 20 5.1.3 Quantity In Column 3 of Tables 5.1 and 5.2, the results shows the impact of various labels and nutritional content on the sales quantity (measured in total ounces). It is important to note that a single product package is typically comprised of several ounces of product, making the influence of these labels and nutritional information on the total sales ounces particularly significant. In table 5.1, Meat substitute (products identified as meat substitutes) significantly de- creased sales quantity (-411.58***). This indicates that being labeled as a meat substitute correlates with a significant reduction in sales quantity. Plant-based or Veggie (products identified as plant-based or veggie) significantly increased sales quantity (209.83***), sug- gesting that these labels can boost consumer purchase volume, thereby increasing the to- tal sales ounces. Protein (products identified as high protein) significantly increased sales quantity (428.91***), indicating that consumers are more inclined to purchase products labeled with high protein, increasing the total sales ounces. Organic (products identi- fied as organic) significantly decreased sales quantity (-544.20***), possibly due to the higher prices of organic products, suppressing purchase volume and reducing the total sales ounces. Low calorie (products identified as low calorie) significantly decreased sales quantity (-529.17***), possibly due to lower consumer acceptance of low-calorie prod- ucts, leading to a reduction in total sales ounces. Low fat (products identified as low fat) significantly increased sales quantity (470.21***), indicating high consumer demand for low-fat products, increasing the total sales ounces. Chicken (products identified as chicken) significantly decreased sales quantity (-183.49***), possibly because consumers prefer other substitutes, leading to a reduction in total sales ounces. In Table 5.2, we see similar results for the log-linear regressions. Meat substitute isassociated with an approximate 16.5% decrease in sales quantity (e−0.18 − 1). Plant- based or Veggie products are associated with an approximate 6.2% increase in sales quantity (e0.06 − 1). Being identified as high protein is associated with an approximate 21 23.4% increase in sales quantity (e0.21 − 1). Organic is associated with an approximate 25.9% decrease in sales quantity (e−0.30 − 1). Low calorie labels are associated with an approximate 74.1% decrease in sales quantity (e−1.35 − 1). Low fat labels are associated with an approximate 20.9% increase in sales quantity (e0.19 − 1). Chicken labels are associated with an approximate 1% decrease in sales quantity (e−0.01 − 1). 5.1.4 The Role of Nutrition Content The InfoScan scanner data also provides information on the nutritional content per serv- ing for the products. Since consumers typically do not calculate nutritional content per ounce, but rather consider the total content in a package or container, we focus on the total nutritional content of each specific product. Table 5.3 presents the combined effects of label information and nutritional content on the log of price per ounce, log of total sales, and log of quantity sold. In log-log regressions, coefficients represent elasticities, measuring the percentage change in the dependent vari- able for a 1% change in the independent variable. Through this analysis, we can observe the effects of different labels and nutrition content on price, total sales, and quantity sold. For the price per ounce regression results (Column 1), if a product is labeled as Meat or if fat per upc or sugar per upc increases by 1%, the price significantly increases by 8.4%, 13%, and 5.3%, respectively. Conversely, if a product is labeled as plant based or veggie products or chicken, or if protein per upc or calories per upc increases by 1%, the price significantly decreases by 3.3%, 8.8%, 4.7%, and 16.8%, respectively. Products labeled as meat substitutes have increased prices but significantly lower total sales and quantities sold. Plant-based or veggie products have lower prices, slightly lower total sales, but slightly higher quantities sold. Chicken-labeled products have significantly higher prices and total sales but lower quantities sold. Increases in protein content lead to higher prices and quantities sold, with a substantial impact on total sales. Increases in fat content significantly raise prices but decrease total sales and quantities sold. Increases 22 in calorie content reduce prices significantly but greatly increase total sales and quantities sold. We noticed a similar trend here as in our previous regressions. Table 5.3: Log-Log Estimates of the Impact of Label Claims and Nutrition Content (1) (2) (3) Log(price per ounce) Log(total sales) Log(quantity) meat sub 0.040∗∗∗ -0.024∗∗∗ -0.065∗∗∗ (0.001) (0.004) (0.004) plant based or veggie -0.027∗∗∗ -0.020∗∗∗ 0.007∗ (0.000) (0.003) (0.003) chicken 0.071∗∗∗ 0.096∗∗∗ -0.025∗∗∗ (0.001) (0.006) (0.006) log protein per upc -0.020∗∗∗ 0.218∗∗∗ 0.238∗∗∗ (0.001) (0.004) (0.004) log fat per upc 0.081∗∗∗ -0.275∗∗∗ -0.356∗∗∗ (0.000) (0.003) (0.003) log calories per upc -0.223∗∗∗ 0.774∗∗∗ 0.997∗∗∗ (0.001) (0.007) (0.007) Constant 0.327∗∗∗ 0.523∗∗∗ 0.197∗∗∗ (0.005) (0.031) (0.032) N 1.51e+06 1.51e+06 1.51e+06 R-squared 0.507 0.557 0.567 Notes: All regressions include firm, year, and RMA fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. 23 5.2 Logit Results Since the market shares in the left-hand side of Equation 4.7 consists of observed data, this model can be estimated using linear IV GMM. Then we use PyBLP (Conlon and Gortmaker, 2020), a Python package for structural demand estimation, which yields a result for the parameter on prices: α = −6.5 (Table 5.4). The coefficient α = −6.5 indicates that for every unit increase in price, the log-odds of a product being chosen decreases by 6.5 units. When the price changes by ∆p, the change in log-odds is given by ∆(log-odds) = α · ∆p = −6.5 · ∆p. For small price changes, we can approximate the coefficient as the percentage change in the probability of selection: %∆P ≈ (α ·∆p) ·100%. For example, when the price per ounce increases by $0.01 (increasing price 0.16 dollars per pound): %∆P ≈ (−6.5 · 0.01) · 100% = −6.5%. This indicates that a small increase in price could lead to a significant decrease in the probability of selection. We can observe that consumers are highly sensitive to price changes for plant-based products, where even a small price increase could lead to a substantial drop in demand. Conversely, a price reduction strategy could significantly increase market share. Table 5.4: Logit Results Summary Problem Results Summary GMM Step 2 Objective Value +29000 Clipped Shares 0 Weighting Matrix Condition Number +660000 Beta Estimates (Robust SEs in Parentheses) Prices -6.5*** (0.20) 24 5.3 Random Coefficients Logit Results 5.3.1 Market Structure Assumptions In the context of the plant-based product market, our analysis is based on a comprehensive dataset covering nearly all relevant plant-based brands nationwide. We rely on a Bertrand competition model, assuming that firms compete through prices rather than quantities. Each firm sets prices to maximize its profits while considering the pricing strategies of other firms. We assume a static Nash equilibrium, meaning that the market is in a state of equilibrium where each firm’s pricing strategy is the optimal response, and no firm has an incentive to unilaterally change its pricing strategy. We also assume perfect information, where firms are aware of the market demand structure and the cost structures of their competitors. Profit maximization is our core assumption, with each firm’s objective being to max- imize its profits. We assume pricing at the brand level rather than at the individual product level. In the short term, we assume that marginal costs remain constant. Con- sidering the existence of multi-product firms, our model reflects that firms consider the overall profitability of their product portfolio when making pricing decisions. Regarding the demand structure, we use the estimated random coefficient logit model to calculate market shares and price elasticities. This model reveals significant heterogeneity in con- sumer preferences, particularly concerning price, meat substitutability, and low-calorie claims. Although our data covers the entire US, we still assume that geographic mar- kets are somewhat independent and that firms make pricing decisions separately for each market. Given the comprehensiveness of our data, we adopt the assumption of no supply constraints, meaning firms can meet any level of demand. Our model ignores dynamic considerations, such as the long-term effects of product innovation, which could be a limitation. Finally, we assume that product characteristics are exogenously given and do not change in the short term. 25 5.3.2 Demand Parameters Table 5.5: Beta Estimates (Robust SEs in Parentheses) prices -8.5*** (0.82) The price coefficient α in the random coefficients logit model is -8.5 (SE: 0.82), which is consistent with the Logit results and indicates a significant negative impact of price on product choice (Table 5.5). In other words, consumers are highly sensitive to price changes. The random coefficients matrix (Table 5.6) reveals heterogeneity in consumer preferences. The constant term (-4.3 (SE: 0.49)) represents baseline preference hetero- geneity, with the negative value indicating that some consumers are less likely than others to choose any product. This suggests a significant segmentation in the market, where some groups are more interested in the entire product category while others are not. Meat substitutes (coefficient 2.9(SE: 0.13)) and low-calorie claims (coefficient 5.5(SE: 0.54)) show the largest and most significant preference variations, indicating that these features may be key to market segmentation. Protein content (coefficient -1.6(SE: 0.14)) also shows importance, but consumer reactions vary. In contrast, the direct “plant based” label, organic, and low-fat claims are less significant. 5.3.3 Marginal Costs We can deduce the marginal costs from the firms’ first-order conditions. We assume that prices are determined at the firm level, which coincides with the brand level, where each firm sets prices to maximize its total profits. The first-order conditions (FOC) in this context form a vector ∂πf ∂pf , with the element corresponding to product j in the set Ff of products sold by firm f given by (omitting the t subscript, assuming prices are set at the individual market level): 26 T ab le 5. 6: N on li n ea r C o effi ci en t E st im at es (R ob u st S E s in P ar en th es es ) S ig m a: 1 m ea t su b p la n t b as ed p ro te in or ga n ic le ss ca lo ri e cl ai m le ss fa t cl ai m 1 -4 .3 ** * (0 .4 9) m ea t su b 0 + 2. 9* ** (+ 0. 13 ) p la n t b as ed 0 0 -0 .2 0 (+ 0. 62 ) p ro te in 0 0 0 -1 .6 ** * (+ 0. 14 ) or ga n ic 0 0 0 0 + 0. 20 (+ 2. 9) le ss ca lo ri e cl ai m 0 0 0 0 0 + 5. 5* ** (+ 0. 54 ) le ss fa t cl ai m 0 0 0 0 0 0 -0 .4 4 (+ 1. 0) 27 0 = ∂πf ∂pj = ∂ ∂pj ∑ n∈Ff sn(pn −mcn) = sj + ∑ n∈Ff ∂sn ∂pj (pn −mcn) (5.1) which can be rewritten as: 0 = s+∆(p−mc) (5.2) where ∆ is a J ×J matrix with ∆n,j equal to ( ∂sn ∂pj ) if both n and j are owned by the same firm, and zero otherwise. Thus, the vector of marginal costs for all products is: mc = ∆−1s+ p (5.3) From Figure 5.1, the distribution of marginal costs exhibits a clear right skew, pri- marily concentrated within the 0 to 1 range, with a peak close to 0. This distribution shape reveals that the marginal costs of most plant-based products in the market are rel- atively low, aligning with theoretical expectations of economies of scale and competitive markets. The right skew of the distribution indicates that the majority of plant-based meat substitutes have relatively low marginal costs, likely due to the use of common plant protein ingredients (such as soy and pea protein) and the advantages of large-scale production. However, the long-tail characteristic of the distribution suggests that some products have higher marginal costs, which may indicate a focus on premium products for certain items. 5.3.4 Price Elasticities For a specific market t, the price elasticities of product j concerning a price change in product k within the same market are expressed by 28 Figure 5.1: Marginal Cost (dollars) ηkj := ∂sj/sj ∂pk/pk = ( pk sj ) ∂ ∂pk (∫ ∫ exp(δj + σBviBj + σIIipj) 1 + ∑ m exp(δm + σBviBm + σIIipm) dF v(v)dF I(I) ) (5.4) ≈ pk sj ns∑ i=1 (α + σIIi)(−sijsik + 1{k=j}sik) (5.5) From Figure 5.2, we find that the negative own-price elasticities indicate that the law of demand holds, meaning that as prices increase, the quantity demanded decreases. Most PBMAs have own-price elasticities clustered between -4 and -3. This suggests that a 1% increase in price would result in a decrease in demand by approximately 3% to 4%, indicating high price sensitivity for these products. The high own-price elasticities could be attributed to the high substitutability of plant-based meat alternatives. If the prices of these products increase, consumers are likely to switch to other meat substitutes. The blue vertical line in the figure, representing the aggregate elasticity, peaks close to zero, indicating low overall market elasticity. This implies that while individual plant- 29 based meat alternatives exhibit high price sensitivity, the overall market is less responsive to price changes. This could be because, in the face of rising prices for plant-based meat alternatives, consumers switch between different substitutes, thereby dampening the overall market elasticity. Figure 5.2: Mean Own Elasticities and Aggregate Elasticities 5.3.5 Markups The markup ratio is typically calculated using the Lerner index: P−MC P , where MC is the marginal cost. From Figure 5.3, we find that the markup distribution in the plant-based meat alternatives market exhibits an extreme left-skewed pattern, highly concentrated near zero, reflecting a unique market structure and competitive landscape. Such low markup rates suggest intense market competition, with firms generally facing significant challenges in maintaining profitability. This phenomenon likely stems from multiple in- teracting factors. Firstly, high production costs exert substantial pressure on companies, limiting their ability to raise prices. Secondly, the market may have entered a relatively 30 mature stage where price competition has become the dominant strategy. Some firms might leverage economies of scale to maintain low markup rates, while high consumer price sensitivity forces companies to keep prices low to secure market share. Product homogenization may further intensify price competition, making it difficult for firms to increase profits through differentiation. This distribution pattern may also indicate that the industry is in a rapid growth or market penetration phase, where companies prioritize expanding market share over short-term high profits. Figure 5.3: Markups, % 5.3.6 Market Concentration The market for PBMAs is complex. Although the Diversion Ratio distribution graph in Figure 5.4 indicates low diversion ratios between most products, suggesting a highly differentiated and fragmented market, we also observe high own-price elasticities for indi- vidual products (-4 to -3), indicating that consumers are sensitive to price changes. This 31 seemingly contradictory phenomenon reflects the multi-layered nature of the market. Al- though consumers may remain relatively loyal to specific brands or product types, they still consider other options when there are significant price changes. The overall market elasticity is low (with the peak of the blue vertical line in Figure 5.2 close to zero), further indicating that although consumers are sensitive to the price of individual products, they tend to switch within the plant-based category rather than completely abandoning it. This explains why price changes for individual products may cause consumers to turn to other plant-based options while keeping the overall market stable. Products may be similar in basic functionality (leading to some degree of homogeneity), but differences in brand, taste, texture, and other details result in a highly fragmented and interrelated market structure. Figure 5.4: Diversion Ratio 32 CHAPTER 6 CONCLUSIONS This study analyzes the impact of label claims on the prices, total sales, and quantities of plant-based meat alternatives using InfoScan retail scanner data from 2012 to 2021. The regression analyses show that products labeled as “Meat substitute” and “Chicken” command higher prices, while “Plant-based or Veggie” and “Protein” claims are associ- ated with lower prices. The logit and random coefficients logit models indicate that price has a significant negative impact on product choice, with an estimated price coefficient of -8.5. Consumers highly value low-calorie claims and meat substitute attributes, while the “plant-based” attribute shows a less pronounced impact. In other words, label claims and nutritional content significantly influence the prices, total sales, and quantities of plant-based meat alternatives. The distribution of marginal costs suggests that most plant-based products have rel- atively low marginal costs due to economies of scale and competitive markets. However, some products incur higher marginal costs, indicating a focus on premium items. The price elasticity analysis reveals high own-price elasticity for plant-based meat alternatives, suggesting that these products are highly price-sensitive. In contrast, the overall market elasticity is low, implying that while individual products are sensitive to price changes, the market as a whole is less responsive due to consumer substitution among different products. This study has several limitations worth noting. First, we currently face endogeneity issues with the model—there may be endogeneity between prices and certain product characteristics, potentially leading to estimation bias. The dynamic nature of consumer behavior, such as habit formation or brand loyalty, may have been overlooked as well. The impact of market trends and seasonal variations, especially the changes in market structure pre- and post-COVID, were not considered in our study. Additionally, we may have underestimated the competitive relationship between plant-based products and 33 traditional meat products. Lastly, we did not fully account for the impact of consumer environmental awareness and health considerations on their choices, particularly in terms of labeling information. These findings have significant implications for the marketing strategies of plant- based products. It is essential to consider developing and promoting products targeted at different market segments, particularly focusing on meat substitutes and low-calorie options. Emphasizing the protein content of products is important, but it is also crucial to recognize the variability in consumer demand. Pricing strategies should be carefully considered to address high price sensitivity. The marketing emphasis on organic and low- fat claims may need to be reevaluated. Additionally, our results provides directions for further research, such as exploring the relationship between consumer characteristics and preferences, investigating the reasons behind the lack of significant preference differences for certain attributes, and studying the combined effects of different attribute combina- tions on consumer choices. From a state legislation perspective, prohibiting the use of the term ”meat” could harm consumer welfare by potentially misleading consumers and leading to higher spending on products that may be less environmentally friendly. Future work includes considering the incorporation of product ratings or review data (if available) as proxies for quality, analyzing the substitution relationship between plant- based products and traditional meat products, and examining the impact of different retail channels (e.g., supermarkets, specialty stores) on the demand for plant-based prod- ucts. Additionally, integrating store aisle product location information into the analysis to examine the impact of in-store product placement should be considered if data allows. Future research should also incorporate more detailed demographic data to further elu- cidate consumer behavior and preferences. Furthermore, future research should analyze counterfactual policy scenarios, such as restrictions on the use of terms like “meat” and “burger” or store aisle restrictions. 34 APPENDIX A APPENDIX Table A.1: The Impact of Label Claims and Nutrition Content (1) (2) (3) Price per ounce Total sales Quantity meat sub 0.013∗∗∗ 409.755∗∗∗ 551.881∗∗∗ (0.001) (10.932) (40.319) plant based or veggie -0.002∗∗∗ -86.412∗∗∗ -89.746∗∗∗ (0.000) (5.862) (21.621) chicken -0.002∗ 345.368∗∗∗ 780.497∗∗∗ (0.001) (11.371) (41.937) protein per upc 0.000∗∗∗ -8.860∗∗∗ -12.692∗∗∗ (0.000) (0.675) (2.283) fat per upc 0.003∗∗∗ -22.852∗∗∗ -86.170∗∗∗ (0.000) (0.534) (1.971) sugar per upc 0.001∗∗∗ -12.009∗∗∗ -61.004∗∗∗ (0.000) (0.675) (2.979) calories per upc -0.000∗∗∗ -11.843∗∗∗ 5.003∗∗∗ (0.000) (0.159) (0.643) N 2.77e+05 2.77e+05 2.77e+05 R-squared 0.441 0.626 0.631 Notes: All regressions include firm and year fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. 35 Table A.2: Linear Estimates of the Impact of Nutrition Content (1) (2) (3) Price per ounce Total sales Quantity protein per upc 0.000∗∗∗ -5.868∗∗∗ -9.913∗∗∗ (0.000) (0.158) (0.583) fat per upc 0.003∗∗∗ -20.252∗∗∗ -83.082∗∗∗ (0.000) (0.531) (1.953) sugar per upc 0.001∗∗∗ 1.014 -42.083∗∗∗ (0.000) (0.730) (2.687) calories per upc -0.000∗∗∗ 1.195∗∗∗ 4.301∗∗∗ (0.000) (0.039) (0.143) N 2.77e+05 2.77e+05 2.77e+05 R-squared 0.440 0.624 0.630 Notes: All regressions include firm, year and RMA fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10 36 Table A.3: Log-Linear Estimates of the Impact of Nutrition Content (1) (2) (3) Log(price per ounce) Log(total sales) Log(quantity) log protein per upc 0.022∗∗∗ -0.315∗∗∗ -0.338∗∗∗ (0.002) (0.010) (0.010) log fat per upc 0.118∗∗∗ -0.491∗∗∗ -0.609∗∗∗ (0.002) (0.012) (0.012) log sugar per upc 0.045∗∗∗ 0.071∗∗∗ 0.026∗∗∗ (0.001) (0.007) (0.007) log calories per upc -0.259∗∗∗ 1.163∗∗∗ 1.422∗∗∗ (0.004) (0.024) (0.025) Constant 0.220∗∗∗ 0.658∗∗∗ 0.438∗∗∗ (0.017) (0.097) (0.099) N 2.24e+05 2.24e+05 2.24e+05 R-squared 0.426 0.621 0.628 Notes: All regressions include firm, year and RMA fixed effects. Standard errors are in parentheses. *, **, and *** indicate significance at the 10 37 BIBLIOGRAPHY Adalja, A. A. (2022). Voluntary quality disclosure in credence good markets. Berry, S., Levinsohn, J., and Pakes, A. (1995). Automobile prices in market equilibrium. Econometrica, 63(4):841–890. Berry, S. T. (1994). Estimating discrete-choice models of product differentiation. The RAND Journal of Economics, pages 242–262. Campos, S., Doxey, J., and Hammond, D. (2011). Nutrition labels on pre-packaged foods: a systematic review. Public health nutrition, 14(8):1496–1506. Conlon, C. and Gortmaker, J. (2020). Best practices for differentiated products demand estimation with PyBLP. The RAND Journal of Economics, 51(4):1108–1161. DeMuth, B., Malone, T., McFadden, B. R., and Wolf, C. A. (2023). Choice effects asso- ciated with banning the word “meat” on alternative protein labels. Applied Economic Perspectives and Policy, 45(1):128–144. Eden, S. (2011). Food labels as boundary objects: How consumers make sense of organic and functional foods. Public understanding of science, 20(2):179–194. Ericsson, K. A. and Kintsch, W. (1995). Long-term working memory. Psychological review, 102(2):211. Eshel, G., Stainier, P., Shepon, A., and Swaminathan, A. (2019). Environmentally opti- mal, nutritionally sound, protein and energy conserving plant based alternatives to us meat. Scientific reports, 9(1):10345. FMCG Gurus (2020). Global consumer trends in the protein market. Accessed: 2024- 07-18. Fund, A. L. D. (2023). Miyoko’s victorious in first amendment lawsuit against california department of food and agriculture. Accessed: 2024-07-15. Gleckel, J. A. (2020). Are consumers really confused by plant-based food labels? an empirical study. J. Animal & Env’t L., 12:1. He, J., Evans, N. M., Liu, H., and Shao, S. (2020). A review of research on plant-based meat alternatives: Driving forces, history, manufacturing, and consumer attitudes. Comprehensive reviews in food science and food safety, 19(5):2639–2656. 38 Hu, F. B., Otis, B. O., and McCarthy, G. (2019). Can plant-based meat alternatives be part of a healthy and sustainable diet? Jama, 322(16):1547–1548. Inguglia, E. S., Song, Z., Kerry, J. P., O’Sullivan, M. G., and Hamill, R. M. (2023). Addressing clean label trends in commercial meat processing: strategies, challenges and insights from consumer perspectives. Foods, 12(10):2062. Katare, B., Yim, H., Byrne, A., Wang, H. H., and Wetzstein, M. (2023). Consumer will- ingness to pay for environmentally sustainable meat and a plant-based meat substitute. Applied Economic Perspectives and Policy, 45(1):145–163. Keoleian, G. A. and Heller, M. C. (2018). Beyond meat’s beyond burger life cycle as- sessment: a detailed comparison between a plant-based and an animal-based protein source. Technical report. Kyriakopoulou, K., Dekkers, B., and van der Goot, A. J. (2019). Plant-based meat analogues. In Sustainable meat production and processing, pages 103–126. Elsevier. Luce, R. D. (1959). Individual choice behavior, volume 4. Wiley New York. Nelson, P. (1970). Information and consumer behavior. Journal of political economy, 78(2):311–329. Neuhofer, Z. T. and Lusk, J. L. (2022). Demand for plant-based meat alternatives and the role of habit formation and variety seeking. Nevo, A. (2000). A practitioner’s guide to estimation of random-coefficients logit models of demand. Journal of economics & management strategy, 9(4):513–548. Ortega, D. L., Sun, J., and Lin, W. (2022). Identity labels as an instrument to reduce meat demand and encourage consumption of plant based and cultured meat alternatives in china. Food Policy, 111:102307. Piqueras-Fiszman, B. and Spence, C. (2015). Sensory expectations based on product- extrinsic food cues: An interdisciplinary review of the empirical evidence and theoret- ical accounts. Food Quality and Preference, 40:165–179. Ruby, M. B. (2012). Vegetarianism. a blossoming field of study. Appetite, 58(1):141–150. Segovia, M. S., Yu, N.-Y., and Van Loo, E. J. (2023). The effect of information nudges 39 on online purchases of meat alternatives. Applied Economic Perspectives and Policy, 45(1):106–127. Spronk, I., Kullen, C., Burdon, C., and O’Connor, H. (2014). Relationship between nutrition knowledge and dietary intake. British journal of nutrition, 111(10):1713– 1726. Taylor, S. (2020). Meat wars: The unsettled intersection of federal and state food labeling regulations for plant-based meat alternatives. U. Mass. L. Rev., 15:269. Turtle Island Foods, SPC, e. a. (2024). Turtle island foods, spc, et al., plaintiffs, v. locke thompson, et al., defendants. Case No. XYZ, Court Name. Accessed: 2024-07-15. U.S. Cattlemen’s Association (2018). Petition 18-01. Accessed: 2024-07-15. U.S. Department of Agriculture (2023). National retail report - beef advertised prices for beef at major retail supermarket outlets ending during the period of 07/12 thru 07/18. Accessed: 2024-07-15. Van Loo, E. J., Caputo, V., and Lusk, J. L. (2020). Consumer preferences for farm-raised meat, lab-grown meat, and plant-based meat alternatives: Does information or brand matter? Food Policy, 95:101931. Zhao, S., Wang, L., Hu, W., and Zheng, Y. (2023). Meet the meatless: Demand for new generation plant-based meat alternatives. Applied Economic Perspectives and Policy, 45(1):4–21. Zimmerman, F. J. and Shimoga, S. V. (2014). The effects of food advertising and cognitive load on food choices. BMC public health, 14:1–10. 40 Biographical Sketch Dedication Acknowledgements Table of Contents List of Tables List of Figures Introduction Literature Review Information Interventions and Consumer Behavior Market Demand and Consumption Behavior Labelling Effect Data Data Source Selection of Plant Based Meat Categories Key Variables and Summary Tables Market Size Methodology Hedonic Regression Models Structural Estimation Logit Model Random Coefficients Logit Model Results Hedonic Regression Results Price Per Ounce Total Sales Quantity The Role of Nutrition Content Logit Results Random Coefficients Logit Results Market Structure Assumptions Demand Parameters Marginal Costs Price Elasticities Markups Market Concentration Conclusions Appendix