THE EFFECT OF FOOD SAFETY INFORMATION ON CONSUMERS’ RESTAURANT PURCHASING BEHAVIOR A Thesis Presented to the Faculty of the Graduate School of Cornell University In Partial Fulfillment of the Requirements for the Degree of Master of Science in Hotel Administration by Wichayamas Paekul May 2025 © 2025 Wichayamas Paekul ABSTRACT Food safety can lead to fatal health risks; even so, the topic is rarely discussed with consumers unless an unfortunate incident of foodborne illnesses occurs. In the United States, foodservice establishments were responsible for a fourth of the foodborne illnesses cases reported between 2021 and 2023. The food-away-from-home consumption has also been on the rise during the same period. So, a food safety inspection program is one way to help consumers assess restaurant food safety risks. This study examined the effect of food safety information on consumers’ choices of fast-casual restaurants. Analyzing the choice experiment data from 500 participants, our findings suggested that consumers are willing to pay relatively more for a higher food safety letter grade. However, the effect of the type of violations produced a mixed result. The findings are important for policymakers, restauranteurs, and scholars, particularly offering insights into how consumers react to the assessment of food safety information. iii BIOGRAPHICAL SKETCH Wichayamas, or also known as Gift, was born and raised in Trang, Thailand. Growing up in a household where food is the primary love language and surrounded by relatives who work in the food industry, she developed her passion and love for the hospitality industry, specifically food and beverage, at a young age. Wichayamas graduated from Boston University with distinction (Magna Cum Laude) in Hospitality Administration. During her undergraduate years, she was fortunate to develop her interests in the foodservice industry further from leaders at the Aramark Collegiate Hospitality at Boston University and the Restaurant Associates at Harvard Business School. Following her undergraduate studies, she joined the Master of Science program at the Nolan School of Hotel Administration at Cornell University, focusing on food and beverage management. In the fall of 2025, Wichayamas will continue her graduate education at the University of Houston in the Hospitality Administration Ph.D. program. iv To the Paekul and Kunchong families. v ACKNOWLEDGMENTS This thesis will not be completed without the support of many parties. First, I would like to thank my advisor and the thesis committee chair, Professor Aaron Adalja, for his continuous support. This project would not have been able to run without his assistance and generous funding. I also would like to thank my thesis committee members, Professor Bradley Rickard and Dr. Jie Li, who helped shape the study's design and provided valuable inputs for manuscript editing. I want to take this opportunity to thank all the faculty at Cornell and the Nolan School, whom I would not have the tools to use in my thesis. To all my friends here at Cornell and my Thai Scholars community, thank you for always checking in while fighting our way through graduate school together. In addition to the support here in the United States, I want to give a big shout-out to my family back in Thailand, who continuously believe in me on the days I doubt myself. Thank you, Mom, for always being one phone call away despite the 11-hour time difference. Finally, I would like to thank SEVENTEEN and Tattoo Colour for the music that boosted me through the process of writing. vi TABLE OF CONTENTS Biographical Sketch………………………………………………………….... iii Dedication……………………………………………………………………… iv Acknowledgement……………………………………………..………………. v Table of Contents……………………………………………..……….……….. vi List of Figures……...………………………………..…………………….…… viii List of Tables………………………………………………….…………..…… ix 1 Introduction………………………………………………………………………. 1 2 Literature Review………………………………………………………………… 5 2.1 Literature Review…………………………………………………………. 5 2.1.1 Decision Making Process…………………………………………….. 5 2.1.2 Dining Experience Attributes………………………………………... 5 2.1.2.1 Food Quality……………………………………………….. 5 2.1.2.2 Price………………………………………………………... 6 2.1.2.3 Cleanliness and Hygiene…………………………………… 7 2.1.3 Food Safety in Restaurant……………………………………………. 8 2.1.4 Food Risk Communication…………………………………………... 8 2.1.5 Discrete Choice Experiment (DCE) and Willingness-To-Pay (WTP). 8 2.1.6 Willingness-To-Pay and Labelling…………………………………... 9 2.2 Hypotheses Development………………………………………………... 10 3 Data……………………………………………………………………….……… 12 3.1 Selection of choice attributes and levels…………………………….…… 13 3.1.1 Imputed Letter Grade and Type of Violations………………………. 13 3.1.2 Food item and Price data…………………………………………….. 15 3.2 Survey design……………………………………………………………. 15 3.3 Survey data collection…………………………………………………… 17 4 Empirical Methodology………………………………………………………….. 21 4.1 Conditional Logit Model…………………………………………..…..…. 21 4.2 Multinomial Logit Model...……………………………………………… 22 4.3 Mixed Multinomial Logit Model………………………………………… 23 vii 4.4 Willingness-To-Pay (WTP) Calculation……………………………….… 24 5 Results…………………………………………………………………………… 25 5.1 Distribution of levels in choice sets ……………………………….…….. 25 5.2 Importance of attributes…………………….……..……………………... 26 5.3 Logit Models………..……………………………………………………. 27 5.4 Willingness-To-Pay…………………………………………………….... 30 5.5 Analysis of Interaction Effects…………………………………………… 31 5.5.1 Food Safety Letter Grade B …………………………………………. 31 5.5.2 Food Safety Letter Grade C …………………………………………. 32 5.5.3 General Violation………………………………………….…………. 32 5.5.4 Critical Violation………..……………………………….…………... 33 5.5.5 Marginal WTP of Each Interaction Term…………………….……… 34 6 Conclusions…………………………………………………………………...….. 38 6.1 Discussion of the Results………………………………………………… 38 6.2 Theoretical Implication……………………………...…………………… 40 6.3 Policy and Managerial Implications………………………………………41 6.4 Limitations and suggestions for future research…………………………. 42 7 References……………………………..…………………………………………. 44 Appendix…………………………….…………………………………………...… 50 Appendix 1: Institutional review board for human participants: Notice of exemption…………………..……………………………………… 50 Appendix 2: Survey…………………………………………….………….… 51 Appendix 3: 2020 U.S. Census…………………………………...………….. 61 Appendix 4: Behavioral characteristics………...……………………………. 64 viii LIST OF FIGURES Figure 3.1: Information treatment…………………………………………………. 13 Figure 3.2: Explanation of food safety letter grade attributes and its levels………. 14 Figure 3.3: Explanation of food safety violation type attribute and its levels……... 14 Figure 3.4: Example choice task………………………………………..…………. 17 Figure 5.1: Importance of attributes………………………………………..……… 26 ix LIST OF TABLES Table 3.1: Attributes and levels used in the choice experiment…………………… 17 Table 3.2: Demographic characteristics…………………………...……………….. 19 Table 5.1: Distribution of levels seen and chosen…………………………………. 25 Table 5.2: Estimating demand parameters of Control group………………………. 29 Table 5.3: Estimating demand parameters for Treatment group ………………….. 29 Table 5.4: Consumer willingness-to-pay in USD………………………………….. 30 Table 5.5: Interaction effects with demographics and behaviors of Control group... 33 Table 5.6: Interaction effects with demographics and behaviors of Treatment group.…………………………………………………………………… 34 Table 5.7: Marginal WTP for interaction effects for Control group……..………… 36 Table 5.8: Marginal WTP for interaction effects for Treatment group……….…… 37 Table 6.1: Summary of hypothesis and results…………………………………….. 40 1 CHAPTER 1 INTRODUCTION People are making choices constantly in one day, and one of those decisions includes choosing which restaurant to dine at. One of the factors that drive consumers’ restaurant choices is food quality (Alonso et al., 2013; Namkung & Jang, 2007). Many food quality attributes, such as taste, are categorized as “experience attributes,” where consumers’ assessment comes only after they buy and consume the product (Caswell & Mojduszka, 1996). However, some food quality attributes like food safety and nutritional values fall into “credence attributes.” The reason is that food safety and nutritional values cannot be determined even after consumption; for example, linking foodborne pathogens or nutritional values cannot be done without scientific equipment (Caswell & Mojduszka, 1996). So, informational signage or labels are tools provided to make it more practical for consumers to assess these food qualities. A food safety inspection card is one of the signs that allows consumers to determine one of the food quality attributes, specifically food safety. Hence, the question is how the food safety inspection card impacts consumers’ purchasing behaviors. Food safety is an issue concerning the public and private sectors. Though food safety risks can happen across entities in the food supply chain, restaurants are usually the ones who take the most blame. The reason is that a restaurant is the last entity before the final processed dish reaches the consumer. Between 2021 and 2023, the number of foodborne illnesses in the United States totaled up to 154,383 cases, where the fourth of the cases (40,117) originated from consumption in foodservice establishments, including restaurants, dining services in schools, food trucks, hotels, and casino (CDC, 2 2025). During the same period, we also see a rise in food-away-from-home per capita spending in the U.S., making a new high of $4,485 in 2023 (Sinclair et al., 2024). A food safety inspection is a service adopted by many health departments of various jurisdictions across the U.S. to help ensure restauranteurs hold the highest food safety standards. The results of the inspection are public knowledge. Most jurisdictions publish them online, while some require restaurants to post the formatted printed results. The most well-recognized inspection card is the ABCEats (Food Establishment Inspections) by the New York City Department of Health and Mental Hygiene (NYC Health, 2025). The inspector would visit the restaurants unannounced once every six or twelve months for a food safety inspection to check for food safety compliances, which carries different penalty points. Those points later become a letter grade on a result card that consumers can see on every New York City foodservice establishment window. We have learned that food safety information affects consumer behavior in various ways. This study aims to add to the existing literature on the impact of letter grades and additional food safety information on purchasing decisions in terms of monetary value. One study used the information processing theory as a framework to understand the impact of posted food safety scores on consumer behavior (Choi et al., 2011). Through their pilot study, Choi et al. (2011) found that as the number of violations increased, the respondents’ perceived vulnerability and response efficacy score increased, as well as the fear of contracting foodborne illnesses. The purpose of this thesis is to find out how consumers respond to restaurants’ food safety information and whether it affects their purchasing decisions. A discrete choice experiment is employed to determine consumers’ preferences for restaurant food 3 item (udon), estimate price premiums, understand consumers’ perceptions toward relevant attributes, and explore consumers’ demographic information. The survey was launched through Prolific, a paid online survey distribution platform for data collection for U.S. consumers, to collect the data from 500 respondents in December 2024 (Prolific, 2025). We use three logit models (conditional, multinomial, and mixed) to analyze the conjoint data. The individual’s WTP per attribute is also assessed. The findings suggested that consumers’ valuation of a higher food safety letter grade is positive, such that letter grade A increases the likelihood of purchasing a dish from the restaurant over grades B and C. The analysis of the type of violations provides mixed results regarding additional food safety information. We learned that consumers are willing to pay a premium price for general violations over unknown violations and for unknown violations over critical violations. Following the introduction, the thesis is organized into five additional sections, including (ii) literature review, (iii) data description, (iv) empirical methodology, (v) results, and (vi) conclusions. Chapter 2 includes the study into previous literature about the decision-making process, dining experience attributes, food safety in restaurants, food risk communication, discrete choice experiments, and willingness-to-pay and labeling. Chapter 3 elaborates on the data used in this study. The next chapter, chapter 4, details the description of the data collection and analysis processes in chapter 4. Chapter 5 presents the results obtained from the analysis. Finally, chapter 6 concludes the findings and discusses the theoretical and practical implications. The last chapter ends with limitations and future research directions 4 CHAPTER 2 LITERATURE REVIEW 2.1 Literature Review 2.1.1 Decision Making Process Researchers have developed multiple versions of decision-making models in consumer decision-making processes. One is the stylized EKB model proposed by Pham and Higgins (2005). The stylized EKB model involves the following six steps: problem/need recognition, information search, consideration set formation, evaluation of alternatives, choice (or consumption), and post-choice processes. The step concerning the study is the "evaluation of alternatives", which is the subject of interest in consumer decision-making research, especially in the regulatory focus theory (Pham & Higgins, 2005). This step is crucial as it is adjacent (in a linear process world) to the result of what product, service, or restaurant is chosen. 2.1.2 Dining Experience Attributes 2.1.2.1 Food Quality Caswell and Mojduszka (1996) use the distinction concept developed by Nelson (1970, 1974) to separate food quality attributes into search, experience, and credence. Many food quality attributes, e.g., taste, are considered experience attributes, which consumers’ assessment comes in the post-consumption stage (Caswell & Mojduszka, 1996). However, this does not apply to food safety and nutritional values. These two attributes are considered to be “credence attributes” since product quality cannot be determined even after consumption. For example, linking foodborne pathogens or 5 nutritional values with food is hard to do without scientific equipment. So, informational labeling is important in “making it practicable for consumers to assess quality” (Caswell & Mojduszka, 1996). Beyond this aspect, food quality is the attribute that has been included in several studies concerning the effect of restaurant attributes on satisfaction and loyalty (Canny, 2014; Ha & Jang, 2010; Jeong & Jang, 2010; Mattila, 2001; Namkung & Jang, 2007; Sulek & Hensley, 2004). In the earliest study, Mattila indicated that food quality is among the top three reasons that motivate consumers to choose restaurants. Other researchers confirmed that food quality positively affects satisfaction and revisit intention (Namkung & Jang, 2007; Sulek & Hensley, 2004). 2.1.2.2 Price Price is an attribute that has been included in many studies related to consumers' dining choices. Zeithaml (1998) refers to price as a primary component in acquiring a product or service in an environment where monetary exchange is required. However, price cannot be evaluated in isolation from other attributes, namely quality, because of the complex nature of their relationship. Price acts as a cue for quality, and in turn, perceived quality can often moderate price (Andaleeb & Conway, 2006). In the restaurant context, price is defined as one of the "big four restaurant attributes" (Mathayomchan & Taecharungroj, 2020). Price often comes as a package with value, as consumers are concerned with the perceived value of the price paid for the meal (Kotler & Keller, 2012). 6 2.1.2.3 Cleanliness and Hygiene Cleanliness/ hygiene is one of the five components of the concept of meal experience (Campbell-Smith, 1970). The other four components include food and drink, level of service, value for money, and ambiance. One of the earliest findings found that cleanliness of restaurants, restrooms in particular, and food safety are important factors that influence consumers’ perception of the restaurant and affect their purchase and revisit intention (Barber & Scarcelli, 2009). Cleanliness/ hygiene is further confirmed in another study to be at the top of consumers’ minds. When customers are asked to rank the importance of factors when choosing restaurants to dine out, the mean score of “a clean production/ service environment” was ranked second behind “prior positive experience with a restaurant” (Duarte Alonso et al., 2013). 2.1.3 Food Safety in Restaurants Consumers think about food safety in general and in particular when eating at restaurants (Knight et al., 2007). Though consumers believe that restaurants can ensure food safety, they think other businesses in the food supply chain perform better in this aspect. Consumers’ perceptions of restaurants’ commitment to food safety also affect their dining frequency; consumers will be less likely to dine if the restaurants are perceived as “not at all” committed to food safety (Knight et al., 2007). Without any observable cues, food safety is hard to assess (Nelson, 1970). Many counties developed restaurant inspection programs to respond to rising concerns about foodborne illnesses and restaurant sanitation. The public posting of the hygiene grading program led to a decrease in the number of foodborne diseases in Los Angeles County 7 in the years following the implementation of the program (Jin & Leslie, 2003; Simon et al., 2005). The disclosure of a hygiene grading card, regardless of grades, positively affects the restaurants’ revenue and consumers’ choices of restaurants (Jin & Leslie, 2003). Another case of positive effect of the food safety inspection poster is in New York, NY. There was evidence of a decline in Salmonella Infections in NYC after the restaurants were required to post the inspection results, comparing the 2006-2010 and 2011-2015 (Firestone & Hedberg, 2018). 2.1.4 Food Risk Communication The key role of risk communication is to induce any concerns the public may have (Covello et al., 1987). Information search is part of consumer decision-making (Pham & Higgins, 2005). Information helps consumers weigh benefits and risks when making informed food consumption decisions (Walace & Oria, 2010). Food risks come with a degree of probability that would invoke fear among the public, which results in decreased consumption or loss of revenue for restaurants. Hence, transparency is one way to handle food risk communications, especially regarding food safety issues (Walace & Oria, 2010). 2.1.5 Discrete Choice Experiment (DCE) and Willingness-To-Pay (WTP) This study employs a discrete choice experiment (DCE) to measure Willingness- To-Pay (WTP). A discrete choice experiment (DCE) is a technique that is widely adopted in business (hospitality, tourism, and applied economics) literature in evaluating the monetary values behind the choice selection (Kemperman, 2021). The 8 discrete choice experiment is built upon the Random Utility Theory (Thurstone, 1927) and extended by McFadden (1974). Given the principle of utility-maximizing behavior, the random utility theory suggests that when an individual makes choices based on the characteristics shown, it comes with some degree of randomness (Willis, 2014). Assessing consumers' choices using DCE allows researchers to understand the likelihood of choosing a specific alternative over the availability of two or more alternatives (Friedel et al., 2022). The big advantage of DCE is the flexibility of combining any product's attributes regardless of its existence in the real market. Nevertheless, it still stimulates purchasing decisions that would happen in real life (Breidert et al., 2006). 2.1.6 Willingness-To-Pay and Labeling Willingness-to-Pay (WTP) is a method commonly used across disciplines, including the hospitality and retail sectors. In the retail industry, WTP assessment was on topics ranging from food technological advancement, food safety, and food waste in perishable retail products, such as “imported and hormone-treated beef,” “smart labeling and milk,” and “organic labels for fresh lettuce” (Alfnes, 2004; Endara et al., 2021; Wongprawmas & Canavari, 2017). In the hospitality industry, specifically, in the restaurant segment, WTP was used in evaluating the service attributes (e.g. food quality, ambiance), type of restaurants (e.g. full-service, quick service), performance level (e.g. high, low) and other attributes (e.g. locally sourced, menu, sustainability practices) (Binesh et al. 2023; Namkung & Jang, 2014; Parsa et al. 2017; Shin et al., 2018). There is currently only one study relating to 9 WTP and food safety by Alphonce et al. (2014) who studied the differences in WTP of when individuals are consumers vs. voting citizens. The voting citizen group exhibited a higher WTP when it comes to improving food safety standards in restaurants (Alphonce et al., 2014). 2.2 Hypotheses Development As outlined in the abovementioned literature, they help us formulate hypotheses to test. Consumers generally think about food safety issues, and the belief of restaurants dealing with the issue is lower than other parties in the food supply chain (Knight et al., 2007). Food safety inspection results also provide consumers with an observable cue (Caswell & Mojduszka, 1996). The positive impacts were found that in the presence of food safety inspection cards, there was a decline in a specific foodborne disease in NYC and the increased revenue of restaurants in LA (Firestone & Hedberg, 2018; Jin & Leslie, 2003). Thus, we hypothesize the following: H1: Consumers are willing to pay more for higher food safety letter grade (vs. lower) Food safety communication comes with costs, so transparency is essential (Wallace & Oria, 2010). The core of the five dimensions of transparency communication is that relevant information should be shared and exist to improve consumers’ understanding (Rawlins, 2008). So, having additional information beyond a letter grade can help consumers evaluate the food safety inspection result beyond a letter grade. The availability heuristic also comes into play when assessing day-to-day 10 risks. Consumers rely on their personal experiences or what they have heard and seen in the mass media to make decisions (Schwarz et al., 1991). Therefore, we hypothesize that: H2a: Consumers are willing to pay more when additional information about food safety score is provided (general/critical vs. unknown) H2b: Consumers are willing to pay more when additional information about food safety score is perceived to be better (general vs. critical) H3: Consumers who are exposed to restaurant food safety risks information will exhibit a greater willingness to pay for restaurants with a highest letter grade 11 CHAPTER 3 DATA This chapter is divided into three sections: (i) selection of choice attributes and levels, (ii) survey design, and (iii) survey data collection. The data was collected in December 2024 through a self-administered incentivized survey on Prolific. A total of 500 participants were recruited. One participant failed the attention check question and another eight completed the experiment quicker than the estimated time of 3 minutes and 30 seconds. Hence, the data collected from 491 participants were used in this study. The study begins with a consent form, where only those who accept the terms can proceed to the first part of the survey. The survey consists of three parts: choice-based conjoint experiment, consumption and purchasing behavioral and demographic questions. Participants were randomly assigned to one of two information treatments about food safety and food- away-from-home spending in the United States in 2023 (Figure 3.1). Figure 3.4 shows an example of the choice set presented to participants, and Table 3.1 outlines the attributes and levels used in the choice experiment. 12 Figure 3.1: Information treatment 3.1 Selection of choice attributes and levels The primary data collection is based in Ithaca, NY. We pulled the restaurants’ food safety inspection results from the Food Program page on Tompkins County’s website in June 2024 (Tompkins County, 2024). Since the Tompkins County’s report only provides the violation description, to meet the objective of this study, we decided to create imputed grades by mapping the description with the guidelines provided by the NYC Health Department (NYC Health, 2016). 3.1.1 Imputed Letter Grade and Types of Violations The Self-Inspection Worksheet for Food Service Establishments separates type of violations into general and critical, where the critical violations carry a higher penalty points deduction. The reason for the point deduction severity is because the critical 13 violations are related to actions that carry substantial risks to public health hazards like foodborne illnesses (NYC Health, 2016). The information on the food safety letter grade scoring system and the meaning of each type of violation, as presented in the survey, are shown in Figures 3.2 and 3.3. Figure 3.2: Explanation of food safety letter grade attribute and its levels Figure 3.3: Explanation of food safety violation type attribute and its levels 14 3.1.2 Food item and Price data In this study, we chose udon, a Japanese noodle dish, as a representative food item. Participants had the authority over the protein choice (chicken, shrimp, or tofu) to eliminate any confounding effect that may occur from the selection of meat. We picked this dish as it is commonly sold at the three selected restaurants based on violation code mapping of restaurants in Ithaca, NY. These restaurants vary in food safety grades, number of violations, and type of violations. The price attribute had five levels: $10.99, $11.99, $12.99, $13.99, and $14.99. These prices were selected based on the pricing information of udon sold at the three selected restaurants in Ithaca, NY during our primary data collection, which varied from 11.40 to 11.95 to 13.99. In the end, we chose the 99-ending prices because of the left- digit bias tendency, where people perceived the one-penny difference as if it was more than a twenty-cent increase (Strulov-Shlain, 2023). 3.2 Survey design Upon entering the survey, participants were randomly assigned to one of two sets of information. The first part is the instruction, where they learn about the experiment's setting and are exposed to the photo of udon. Participants are asked to imagine that they are about to purchase udon with their choice of meat from a fast-casual Asian restaurant. They are asked to compare only the three restaurants within the choice set and select the one that they prefer the most. They are also told that characteristics beyond those listed on the page should be assumed to be identical across all three restaurants (e.g., 15 dishware, soup). The second part is the explanation of food safety scores, where letter grade cards and numeric score explanations are provided. Then, participants are equally and randomly put into either a treatment or control group. The information treatment was comprised of statistics about food-away-from- home spending in 2023, foodborne illness risks, and frequency of food safety risk occurrences by restaurant types. The information treatment was shown to participants in the treatment group only. Those in the control group saw information that are not related to food and food safety. A complete copy of the survey is provided in Appendix 2: Survey. The second part of the survey is DCE. The number of tasks and complexity were considered to minimize the risk of participants losing interest and cognitive burden (Mangham et al., 2009). The goal of this part is to gather the information to estimate consumers' willingness to pay based on eight choice tasks with four alternatives, including the no-buy option. The choices were generated through Qualtrics using the "randomized balance design" approach, which is similar to Sawtooth's Balanced Overlap Design (Qualtrics, 2025). The visual presentation of the letter-grade posters is included in the choice set to mimic the actual situation where people see the card at the restaurants and to account for the biased value estimates (Umberger & Mueller, 2010). The third part of the survey is to learn about the individuals' restaurant consumption and habits – dining frequency, average check, methods of purchasing, factors of restaurant selection, and prior food safety risk encounters. Otherwise specified, all questions were constructed on the 4-, 5- or 6-point Likert-type scales. An example of the Likert-type scale used is the answers ranging from 1 being "Not at all 16 important" to 5 being "Extremely important." The complete questionnaire for this section can be found in Appendix 2: Survey. In the fourth and final part of the survey, participants had to fill in the sociodemographic information. Figure 3.4: Example Choice task Table 3.1: Attributes and levels used in the choice experiment Attributes Levels Price $10.99 $11.99 $12.99 $13.99 $14.99 Food safety letter grade A B C Type of violations Unknown violation(s) General violation(s) Critical violation(s) 3.3 Survey data collection The survey was distributed on Prolific on December 11, 2024. Each participant received $2.00 for a completed survey response, as we expected the survey to be completed within 10 minutes, and the compensation rate was $12 per hour. After compiling the responses, we found that, on average, participants completed the survey 17 in 9 minutes. In order to participate in the study, participants must be at least 18 years old and currently reside in the United States. The total of 500 participants completed the survey. An attention check was included by asking the participants to select the country that was mentioned in the information treatment to ensure that they had read the passage. Eleven participants who failed the attention check or completed the survey under 3 minutes and 30 seconds were excluded from the analysis. The sample size was determined through the Sawtooth Software suggestions of minimum sample size, n (Orme & Chrzen, 2017) which the calculation is expressed as 𝑛 ≥ !"""# $% (3.1) where c is the maximum number of levels in any one attribute a is the number of choices in each task excluding no buy option q is the number of tasks Based on equation 3.1, we found that at least 209 respondents should be in each treatment condition. So, recruiting a total of 500 participants is sound for this study. Table 3.2 summarizes demographic characteristics for the overall, control, and treatment groups. The final sample size included 489 participants ( 𝑀%&' = 41.19, SD = 12.41; Female = 56%). 68% of the participants are 45 years old or younger. Half the participants are married (52%) and hold at least a bachelor’s degree or higher (49%). 18 Table 3.2: Demographic characteristics Variables Overall N = 4891 Control N = 2391 Treatment N = 2501 Age 25 or younger2 24 (4.9%) 11 (4.6%) 13 (5.2%) 26-35 158 (32%) 83 (35%) 75 (30%) 36-45 153 (31%) 73 (31%) 80 (32%) 46-55 90 (18%) 43 (18%) 47 (19%) 56-65 40 (8.2%) 17 (7.1%) 23 (9.2%) 65 or older 24 (4.9%) 12 (5.0%) 12 (4.8%) Gender Female2 275 (57%) 143 (60%) 132 (53%) Male 209 (43%) 94 (40%) 115 (47%) Education Less than high school degree 2 (0.4%) 2 (0.8%) 0 (0%) High school diploma/ GED 55 (11%) 30 (13%) 25 (10%) Some college but no degree 97 (20%) 53 (22%) 44 (18%) Associate degree in college (2-year)2 63 (13%) 28 (12%) 35 (14%) Bachelor's degree in college (4-year)2 184 (38%) 82 (34%) 102 (41%) Graduate Degree2 87 (18%) 43 (18%) 44 (18%) Marital status Divorced 37 (7.6%) 20 (8.4%) 17 (6.9%) Living with a partner 51 (11%) 23 (9.6%) 28 (11%) Married2 250 (52%) 117 (49%) 133 (54%) Never married 124 (26%) 66 (28%) 58 (24%) Separated 11 (2.3%) 6 (2.5%) 5 (2.0%) Widowed 12 (2.5%) 7 (2.9%) 5 (2.0%) Children (Yes) 214 (44%) 102 (43%) 112 (45%) Household income Less than $29,999 62 (13%) 29 (12%) 33 (13%) $30,000 to $59,999 147 (31%) 77 (33%) 70 (28%) $60,000 to $89,999 87 (18%) 40 (17%) 47 (19%) Over $90,0002 183 (38%) 87 (37%) 96 (39%) Ethnicity White2 339 (70%) 167 (70%) 172 (69%) Black or African American 94 (19%) 46 (19%) 48 (19%) Hispanic/Latino 15 (3.1%) 6 (2.5%) 9 (3.6%) Asian (including South Asian) 31 (6.4%) 14 (5.9%) 17 (6.8%) American Indian or Alaska 3 (0.6%) 1 (0.4%) 2 (0.8%) Other 4 (0.8%) 3 (1.3%) 1 (0.4%) Region Midwest 98 (20%) 49 (21%) 49 (20%) 19 The sample size in this study does not reflect the U.S. population, so generalizing the results to the broader context should be done carefully. There are notable differences comparing the sample size's demographic characteristics with the 2020 U.S. census data, which can be found in Appendix 3: 2020 U.S. Census data. The first difference is that most people in the sample have a bachelor's degree or higher (49%), whereas only 35% of the U.S. population falls in the same category. Ethnicity is the second difference. Though the sample size and the census data show that White is the predominant race (69% in sample size, 75% in census data), the percentage is not comparable. The percentage of the Hispanic group in the sample size highlighted the difference since only 4.7% of the participants selected identified themselves as Hispanic/Latino whereas 19.5% of the U.S. population represent this group. The data on the sample's behavioral characteristics can be found in Appendix 4: Behavioral characteristics. The data came from the second part of the survey, where we Variables Overall N = 4891 Control N = 2391 Treatment N = 2501 Northeast 101 (21%) 44 (18%) 57 (23%) South 228 (47%) 113 (47%) 115 (47%) West 57 (12%) 32 (13%) 25 (10%) Employment status Working full-time2 271 (56%) 118 (49%) 153 (61%) Working part-time 93 (19%) 49 (21%) 44 (18%) A homemaker or stay-at-home parents 43 (8.8%) 31 (13%) 12 (4.8%) Retired 25 (5.1%) 12 (5.0%) 13 (5.2%) Unemployed and looking for work 27 (5.5%) 15 (6.3%) 12 (4.8%) Student 16 (3.3%) 7 (2.9%) 9 (3.6%) Other 13 (2.7%) 7 (2.9%) 6 (2.4%) Household size Mean (SD) 2.71 (1.35) 2.74 (1.34) 2.68 (1.37) 1n (%) 2reference level 20 asked consumers about their dining habits and behaviors. In terms of dining frequency, we found that there is no particular time that consumers choose to eat out or purchase a takeaway meal from a restaurant. When dining out, 61% of participants spend approximately $11-$20 per person when buying food at a restaurant. The data further shows that consumers in our sample rarely use third-party delivery service (73% use it less than once a week) compared to ordering online or in-store and dine-in or pick-up the food from a restaurant. Behaviors regarding ways to find restaurants in new areas revealed that Google Maps ranked first among the six options, followed by word of mouth or recommendations from friends or locals. Lastly, over half of participants rate the following dining attributes as important when choosing a restaurant: taste (74%), cleanliness of the store (54%), and health/ safety aspects of food items (56%). 21 CHAPTER 4 EMPIRICAL METHODOLOGY 4.1 Conditional Logit Model We use conditional logit model (CL) to fit the discrete choice data we obtained. The model is motivated by a random utility model (McFadden, 1986). It is valid under the assumption that the independence from irrelevant alternatives (IIA) holds, where we assume that overall choice preferences is not affected by the set of choices offered. Thus, terms that represent characteristics of individuals, which are the same for all choices, are not included. The result of the model follows the fact that the random utility model is based on comparisons of pairs of alternatives, not the alternatives themselves (Greene, 2012). The model can be specified as: 𝑈() = x()* 𝛽 + 𝜖()+ (4.1) where x() is the attribute of the choices 𝛽 represents preference for observed characteristics 𝜖()+ is independently and identically distributed extreme value Based on equation (4.1), the probability that consumer I faced with j = 1, 2, …, J alternatives Pr-𝑌( = 𝑗0𝑥(!, … , 𝑥(,4 = -./ (.!" # 2) ∑ -./ (.!" # 2)$ "%& (4.2) 22 4.2 Multinomial Logit Model We use the multinomial logit model (MNL) to analyze the choice experiment data. The MNL allows us to predict the probability of selecting one attribute over another and relaxes the IIA assumption held in the CL. It has been used in many food choice empirical studies such as Mwove et al. 2020; Recio-Roman et al. 2020; Talijiancic et al. 2021; Zanoli et al. 2015. The MNL is rooted in random utility theory, where the assumption is that consumers associate a certain utility that maximizes their utility (Train., 2009). Since we are only able to observe some attributes of the utility that consumer i obtains from alternative (restaurant) j in choice task t is 𝑈()+ = 𝑉()+ + 𝜀()+ ∀ 𝑗, 𝑡 where 𝑉()+ is the observable portion (representative utility) 𝜀()+ is the non-observable random component. The observable portion of the utility function can be expanded as: 𝑉()+ = ∑ 𝛽5(𝑋5()+6 57! + 𝛽"𝑝𝑟𝑖𝑐𝑒()+ (4.3) where 𝛽5( is consumer i’s attribute coefficient for attribute k 𝑋5)+ represents the kth non-price attribute of alternative (restaurant) j in task t 𝛽" is the price coefficient; 𝑝𝑟𝑖𝑐𝑒()+ is the price (continuous) variable generated by the price levels in our experimental design 𝑉(+) is the representative utility Based on equation (4.3), the probability that consumer i choosing an alternative (restaurant) j can be expressed as 𝑝() = 𝑝(𝑦( = 𝑗) = -./ (8!" # 9") ∑ -./ (8!" # 9')( '%& (4.4) 23 4.3 Mixed Logit Model The additional analysis allows us to assess the robustness of the empirical results. So, in addition to the conditional logit model, we use the mixed logit model (MIXL) to analyze the choice experiment data. The mixed logit model takes heterogeneity of population into account. It relaxes the IIA assumption, and the marginal utilities associated with choice characteristics vary between individuals. The choice utility can be expressed as: 𝑈()+ = ∑ 𝛽5(𝑋5()+6 57! + 𝛽"𝑝𝑟𝑖𝑐𝑒()+ + 𝜖()+ (4.5) where 𝑈()+ is consumer i utility from alternative (restaurant) j in task t 𝛽5( is consumer i’s attribute coefficient for attribute k 𝑋5)+ represents the kth non-price attribute of alternative (restaurant) j in task t 𝛽" is the price coefficient 𝑝𝑟𝑖𝑐𝑒()+ is the price (continuous) variable generated by the price levels in our experimental design 𝜖()+ is independently and identically distributed extreme value type 1 The parameters of the model are estimated by simulated maximum likelihood estimation techniques, using 500 Halton draws. The price coefficient is assumed to be fixed, while the random coefficients (food safety letter grades and type of violations) are assumed to follow a normal distribution. Based on equation (4.5), the probability that consumer i choosing an alternative (restaurant) j in choice task t, where the set of alternatives is 𝐶$ can be expressed as Pr (𝑈(+) ≥ 𝑈($+ , 𝑞 ∈ 𝐶$) = -./ (:!)") ∑ -./ (:!)")* ∈-* (4.6) 24 4.4 Willingness-to-pay (WTP) Calculation One application of choice models is to estimate how much consumers value the attributes of the choice (Greene, 2012). The result of WTP shows that while consumers keep their utility constant, how much consumers are willing to spend across different level within an attribute. WTP is computed by dividing the estimated coefficient of a utility model of the mixed logit model calculated in equation (4.5) by the price coefficient. 𝑊𝑇𝑃5 = −𝛽(5/𝛽" (4.7) where 𝛽" is the coefficient for price i denotes an individual consumer k denotes product attribute 25 CHAPTER 5 RESULTS In this chapter, the results are divided into x sections: (i) distribution of levels in the choice sets, (ii) importance of attributes, (iii) the logit models, (iv) willingness-to- pay, and (v) interaction effects. The model estimation was done using the ‘gmnl’ package in R (Sarrias & Daziano, 2017). 5.1 Distribution of levels in choice sets Table 5.1 summarizes the distributions of levels seen and chosen by all participants. Based on the distribution of levels, the experimental design was balanced, as levels within each attribute appear in similar frequencies. Looking at the chosen columns, we observe preferences’ heterogeneity in all variables. The food safety letter grade is the variable that shows the most significant heterogeneity, and the majority of participants who saw grade A ended up choosing it. Table 5.1: Distribution of levels seen and chosen Overall Control Treatment Seen (%) Chosen1 (%) Seen (%) Chosen1 (%) Seen (%) Chosen1 (%) Price $10.99 20.01 35.18 19.86 36.17 20.15 34.24 $11.99 20.02 30.40 19.93 30.97 20.10 29.85 $12.99 19.87 27.40 19.86 27.66 19.88 27.16 $13.99 20.12 25.58 20.31 25.49 19.93 25.67 $14.99 19.99 23.19 20.05 22.70 19.93 23.66 Letter grade A 33.33 56.01 33.32 55.31 33.35 56.67 B 33.37 23.60 33.35 24.99 33.38 22.27 26 C 33.30 5.42 33.33 5.44 33.27 5.41 Type of violations Unknown 33.19 31.94 33.14 32.93 33.23 30.99 General 33.47 37.81 33.47 37.81 33.47 37.80 Critical 33.34 15.28 33.39 14.99 33.30 15.57 1 Chosen shows the percentage from the total seen of each level. For example, we can interpret that of 20% who saw $10.99, 35.18% chosen that option. 5.2 Importance of attributes Figure 5.1 shows the importance of attributes from the choice sets. After completing eight choice tasks, we asked participants to indicate each attribute's importance on a Likert-type scale of 1-5, from Not at all important to Extremely important. Participants' opinions mainly were similar between the control and treatment groups except for the type of violations. In the treatment group, more participants rated the type of violation as an extremely important feature compared to those in the control group. Figure 5.1: Importance of attributes 27 5.3 Logit Models Tables 5.2 and 5.3 summarize the estimated demand parameters for the conditional (CL), multinomial (MNL), and mixed logit (MIXL) models separated by treatment information groups. As stated in chapter 3, we included both models to assess the robustness of the empirical results. The mixed logit model allows for heterogeneity, which is shown through the standard deviations. The price in MIXL is set to constant, whereas the other variables are set to be normally distributed random variables. The reference levels for this analysis were unknown violation(s) and grade A. The coefficients represent the utility of each level in the choice experiment. The coefficients in both models correspond to the utility for each attribute. In both groups, the three models show that all parameters, except the two-way interacting parameters, are highly significant, with the p-value < 0.05. As expected, the price variable has a negative coefficient, aligning with the downward-sloping demand curve representing the law of demand. This implies that if the price rises, their utility and probability of buying the dish will decrease. The sign of the estimated coefficients of letter grades is negative, which infers that the perceived utility decreases as a letter grade differs from and is lower than A. Furthermore, the difference in magnitude of coefficients when going from grades A to C is two times higher than when going from grades A to B, which supports H1. On the other hand, the signages for two levels of the type of violations are different. The sign of the estimated coefficients of general violation is positive, while it is negative for critical violation. These can be interpreted that, under the experiment 28 conditions, the likelihood of purchasing the dish is higher when the type of violation is general compared to when it is unknown. The estimated coefficient for the no-buy variable (choose not to buy any of the three options) is negative and significant. This implies that consumers are better off and willing to spend their money on udon rather than keeping the money unused. We can see that the information treatment negatively moderates the impact of the food safety letter grades because the estimated coefficients of grades B and C in the treatment groups are lower than the control group for both models. This implies that the information treatment increased the perceived preference for a higher letter grade. These results support H3. However, the information treatment has the opposite impact on the type of violations. The perceived preference for the presence of general violations is lower in the treated condition. It is also important to note that the difference in the estimated coefficients between general and critical became smaller in the treated condition. For example, in the conditional logit model the difference between general and critical is 1.9 (0.475- (-1.425)) in the control group, while it is 1.664 (0.297- (- 1.367)) in the other group. Finally, in the MIXL models, the standard deviations for grade B, grade C, general violations, and critical violations are highly statistically significant, with p < 0.001, which indicates that there is heterogeneity among consumers. The demographics and behavioral parameters may capture the heterogeneity. 29 Table 5.2: Estimating demand parameters for Control group Variables Control CL1 MNL1 MIXL1 Price -0.233 *** (0.025) -0.269*** (0.026) -0.362*** (0.034) Grade B -1.356*** (0.120) -1.556*** (0.134) -2.015*** (0.190) Grade C -3.148*** (0.186) -3.527*** (0.199) -5.734*** (0.453) General 0.495*** (0.122) 0.475*** (0.134) 0.745*** (0.193) Critical -1.258*** (0.120) -1.425*** (0.133) -1.874*** (0.210) No buy -5.214*** (0.338) -5.790*** (0.372) -7.668*** (0.491) Grade B * General -0.406 (0.174) -0.229 (0.187) -0.390 (0.237) Grade B * Critical 0.025 (0.195) 0.134 (0.208) -0.093 (0.268) Grade C * General -0.320 (0.256) -0.273 (0.267) -0.405 (0.360) Grade C * Critical 0.368 (0.321) 0.568 (0.329) 0.144 (0.439) Standard deviation estimates Grade B 1.450*** (0.159) Grade C 2.749*** (0.327) General Violation(s) 1.349*** (0.162) Critical Violation(s) 1.871*** (0.196) Log-Likelihood 1,860 -1,720.5 -1,579.6 Number of observations 1,912 1,912 1,912 1 Mean (Std. Err) Notes: ‘.’ p<0.1, ‘*’ p<0.05, ‘**’ p<0.01, ‘***’ p<0.001 Table 5.3: Estimating demand parameters for Treatment group Variables Treatment CL1 MNL1 MIXL1 Price -0.173 *** (0.024) -0.211 *** (0.026) -0.292*** (0.032) Grade B -1.614*** (0.122) -1.941*** (0.137) -2.409*** (0.183) Grade C -3.488*** (0.190) -3.814*** (0.202) -5.954*** (0.463) General 0.269* (0.118) 0.297* (0.131) 0.536** (0.187) Critical -1.275*** (0.117) -1.367*** (0.127) -1.871*** (0.216) No buy -4.474*** (0.330) -5.081*** (0.361) -6.740*** (0.467) Grade B * General -0.038 (0.172) 0.103 (0.187) -0.055 (0.234) Grade B * Critical 0.142 (0.196) 0.256 (0.207) -0.201 (0.273) Grade C * General 0.333 (0.251) 0.343 (0.260) 0.192 (0.343) Grade C * Critical 0.515 (0.334) 0.497 (0.340) 0.054 (0.457) Standard deviation estimates Grade B 1.185*** (0.154) Grade C 2.749*** (0.391) General Violation(s) 1.352*** (0.157) Critical Violation(s) 2.032*** (0.201) Log-Likelihood 1,902 -1,802.1 -1,654.3 Number of observations 2,000 2,000 2,000 1 Mean (Std. Err) Notes: ‘.’ p<0.1, ‘*’ p<0.05, ‘**’ p<0.01, ‘***’ p<0.001 30 5.4 Willingness-To-Pay (WTP) In the wtp.gmnl function, each attribute's marginal WTP (mWTP) is directly estimated rather than the marginal utility (Sarrias & Daziano, 2017). The developer of this package noted that wtp.gmnl does not include the negative sign in the calculation process ('gmnl' version 1.1.3.2 (2020-05-27)). So, we flipped the WTP signs in Table 5.3 from the original results shown in R. Table 5.3 summarizes consumer willingness-to-pay separated by information treatment groups. We found that the willingness-to-pay for each level is very high and does not fit the real-world context. Thus, the interpretation of the WTP will not be a direct dollar value but a relative value compared to the reference level. In general, both control and treatment groups have similarities in the signs and statistically significant levels of each variable. We found that consumers are willing to pay more for a higher food letter grade. As such, the WTP for grade A is relatively 5 times higher than grade B and relatively 15 times higher than grade C. The difference in WTP between grades B and C is 3 times. The results further reject H2a but support H2b. Having the type of violations indicated on the food safety inspection poster does not positively impact consumers’ choices. Consumers are willing to pay a premium when general is indicated, while expected otherwise when violations are critical. Table 5.4: Consumer willingness-to-pay in USD Variables Control Treatment Grade B -5.559*** (0.688) -8.243*** (1.029) Grade C -15.819*** (1.788) -20.371*** (2.541) General Violation(s) 2.055*** (0.557) 1.834** (0.664) Critical Violation(s) -5.171*** (0.723) -6.401*** (0.9565) No buy -21.155*** (0.805) -23.062*** (1.133) 31 Grade B * General -1.077 (0.659) -0.187 (0.801) Grade B * Critical -0.256 (0.740) -0.687 (0.937) Grade C * General -1.117 (1.000) 0.657 (1.173) Grade C * Critical 0.398 (1.212) 0.185 (1.564) 1 Mean (Std. Err) Notes: ‘.’ p<0.1, ‘*’ p<0.05, ‘**’ p<0.01, ‘***’ p<0.001 Delta method was used for the estimation of the standard errors 5.5 Analysis of Interaction effects We developed MIXL models with interaction effects between each attribute and demographic and behavioral data to further investigate the heterogeneity among consumers when choosing a restaurant to purchase udon (Table 5.4). Through the analysis, we found that some demographic and behavioral variables, such as income, household size, and education, do not show statistical significance in any of the attributes’ levels. Hence, those variables are not reported in Tables 5.4 to 5.7. 5.5.1 Food Safety Letter Grade B In a control condition, the interaction effects with a negative and statistically significant coefficient for grade B over grade A are ages 36-45, 65 or older, those who have seen or heard about food safety inspection posters, and those who have experienced pain or sickness after the meal. The results make sense because these groups are more vulnerable and susceptive to food safety risks in restaurants. On the other hand, the interaction effects with a positive and statistically significant coefficient for grade B are average checks of $31-$40. In a treatment condition, an $11-$20 average check is the only term that shows a positive and statistically significant coefficient with grade B. 32 5.5.2 Food Safety Letter Grade C In the control condition, the interaction effects with a negative and statistically significant coefficient for grade C are all age groups except 65 or older, homemakers, those who have seen or heard about the food safety inspection poster, and those who have experienced pain or sickness after the meal. Meanwhile, the interaction effects with a positive and statistically significant coefficient for grade C are Black and an average check of $31-$40. In the treatment condition, the statistically significant coefficients are all age groups except 65 or older and Black; they also have the same sign as their counterparts in the control condition. 5.5.3 General Violation All statistically significant interaction effects in the control condition show a negative sign. Those variables are age 65 or older, those who identified as Black, and the other racial groups. Age 65 or older is also a negative and statistically significant in the treatment condition. The sign of the estimated coefficients of general violation in the MIXL model (Tables 5.2 and 5.3) is positive. Thus, the direct interpretation is that the perceived utility and probability of purchasing decrease when consumers in these groups see the general violation over the unknown violation. 33 5.5.4 Critical Violation In the control condition, the interaction effects with a negative and statistically significant coefficient for critical violation over unknown violation are age 56-65, retired, and those that experienced pain or sickness after the meal. In the treatment condition, the interaction effects with statistically significant coefficients are retired (negative) and an average check of $50 or more (positive). The positive sign is interesting because it can be inferred that as consumers pay more for a meal, they expect to see information transparency on the food safety inspection poster (critical violation over unknown violation). Table 5.5: Interaction effects with demographics and behaviors of Control group Control Variable Interaction Grade B Grade C General Critical Age 26 - 35 (-)* 36 - 45 (-)* (-)* 46 - 55 (-)** 56 - 65 (-)** (-)* 65 or older (-)** (-)* Ethnicity Black (+)* (-)* Asian Hispanic Native Other (-)* Employment Status Part-time Homemaker (-)* Retired (-)* Unemployed Student Other Average check $11 - $20 $21 - $30 $31 - $40 (+)** (+)* $41 - $50 $50 or more Have seen or heard about food safety inspection poster Yes (-)* (-)* Experiencing pain or sickness after the meal Yes (-)* (-)* (-)* 34 Notes: *p<0.05, **p<0.01, ***p<0.001, estimated (std. err) (-) and (+) indicate the sign of the estimated coefficients Reference level: 25 or younger, White, Working full-time, $10 or less Table 5.6: Interaction effects with demographics and behaviors of Treatment group Treatment Variable Interaction Grade B Grade C General Critical Age 26 - 35 (-)* 36 - 45 (-)* 46 - 55 (-)** 56 - 65 (-)** 65 or older (-)* Ethnicity Black (+)** Asian Hispanic Native Other Employment Status Part-time Homemaker Retired (-)* Unemployed Student Other Average check $11 - $20 (+)* $21 - $30 $31 - $40 $41 - $50 $50 or more (+)* Have seen or heard about food safety inspection poster Yes Experiencing pain or sickness after the meal Yes Notes: .p<0.1, *p<0.05, **p<0.01, ***p<0.001, estimated (std. err) (-) and (+) indicate the sign of the estimated coefficients Reference level: 25 or younger, White, Working full-time, $10 or less 5.5.5 Marginal WTP of each interaction term In addition to estimating the demand parameters, we also compute the marginal WTP for the interaction effects between each attribute and demographic and behavioral data (Tables 5.6 and 5.7). The following variables: gender, income, household size, and college degree are not statistically significant at any level with all attributes. The 35 interaction effects are explained compared to the marginal WTP in the main MIXL model. In the control condition, for age, consumers expect discounted WTP when grade B, grade C, or critical violation is present on the food safety inspection poster. The discounted rates varied across the age groups with no clear pattern, with those aged 65 or older having the highest discounted WTP for grades B and C over A. For ethnicity, people identifying as Black and Other expected discounted WTP when a general violation is present. The results could imply that they do not value the indication of general violation over unknown violation. However, interestingly, the Black variable shows a premium WTP for grade C. For employment status, a homemaker exhibits a lower WTP for grade C. Similarly, retirees show a negative WTP for critical violation. For average checks, consumers are willing to pay a premium in the presence of grades B and C. The marginal WTP for people who have seen or heard about food safety inspection posters and experienced pain or sickness is lower in grade B, grade C, and critical interaction terms. Similar to the control group, in the treatment condition, for the age and retired variables, consumers expect a discounted rate when grade C, general, and critical are present. On the other hand, Black and average check variables have a positive WTP. 36 Table 5.7: Marginal WTP for interaction effects for Control group Control Variable Interaction Grade B Grade C General Critical Age 26 - 35 -10.791*** (2.964) 36 - 45 -10.152** (2.944) 46 - 55 -9.931* (3.175) 56 - 65 -9.280* (3.666) -5.313* (2.612) 65 or older -6.731** (2.372) -12.341** (3.871) Ethnicity Black 3.684* (1.684) -2.121* (0.871) Asian Hispanic Native Other -6.193* (2.834) Employment Status Part-time Homemaker -7.033* (3.096) Retired -4.481* (2.278) Unemployed Student Other Average check $11 - $20 $21 - $30 $31 - $40 5.555** (1.856) 8.080* (3.725) $41 - $50 $50 or more Have seen or heard about food safety inspection poster Yes -2.184* (0.919) Experiencing pain or sickness after the meal Yes -1.493* (0.736) -3.452* (1.540) -2.285* (0.910) Notes: .p<0.1, *p<0.05, **p<0.01, ***p<0.001, estimated (std. err) Reference level: 25 or younger, White, Working full-time 37 Table 5.8: Marginal WTP for interaction effects for Treatment group Treatment Variable Interaction Grade B Grade C General Critical Age 26 - 35 -6.494* (3.253) 36 - 45 -6.655* (3.216) 46 - 55 -10.294** (3.725) 56 - 65 -13.314** (4.955) 65 or older -5.649* (2.625) Ethnicity Black 5.843** (2.014) Asian Hispanic Native Other Employment Status Part-time Homemaker Retired -5.601* (2.760) Unemployed Student Other Average check $11 - $20 $21 - $30 $31 - $40 $41 - $50 $50 or more Have seen or heard about food safety inspection poster Yes Experiencing pain or sickness after the meal Yes Notes: .p<0.1, *p<0.05, **p<0.01, ***p<0.001, estimated (std. err) Reference level: 25 or younger, White, Working full-time 38 CHAPTER 6 CONCLUSIONS 6.1 Discussion of the Results The present study investigated the effect of food safety information on consumers’ restaurant purchasing decisions. The main objective was to explore the marginal value of additional information on the food safety inspection card beyond a letter grade. The first hypothesis (H1) was to test whether consumers are willing to pay more for higher food safety grades. The results from the empirical model supported this hypothesis as consumers showed a positive preference toward a higher food safety letter grade (A vs. B and A vs. C). The WTP for grade A over grade B is $5.56, and grade A over grade C is $15.82. The direct dollar value may not appeal to real-world applications. The results could also be interpreted as a relative value. These results align with the findings in the existing literature regarding the effect of the number of food safety violations on consumers’ behavioral intentions (Choi et al., 2011). The authors found that the response efficacy increased as the number of violations increased. One measurement item under this category is “If I choose a restaurant with a higher inspection score, it will greatly reduce my chances of contracting foodborne illness” (Choi et al., 2011). Our study also demonstrates that a higher grade is considered a safer choice. Hence, consumers will pay more for restaurants with higher food safety letter grades. The second hypothesis (H2a and H2b) was that consumers are willing to pay more for additional information beyond the food safety letter grade. We found that 39 consumers are willing to pay more for general violations over unknown violations, but this is not true in the case of critical violations. According to Schwarz et al. (1991), consumers rely on the information existing in mass media to make decisions. One possible explanation could be that consumers perceived the food safety risks in the following order: general, unknown, and critical (from most to least positive). We found that providing information regarding restaurant food safety risks and food-away-from-home spending effectively increases consumers’ relative WTP and differentiates the effect among the two groups. Under the treatment condition, the impact of lower grades and specified violations is more severe. For example, when grade C is present, consumers expect the discounted WTP of $15.82 and $20.37 in the control and treatment conditions, respectively. Hence, the last hypothesis, H3, that consumers in the treatment condition are willing to pay more was supported. In conclusion, consumers value the higher food safety letter grade and type of violation that is comparably perceived as safer. We failed to reject H1 and H2b but rejected H2a and H3. However, the comparison between unknown vs. general and critical violations is ambiguous and needs further exploration. The reasons behind the ambiguous results can be how the type of violation was described, or there could be a moderator that was not covered in the scope of this study. The treatment information was effective in this study. This could signal that the parties involved in food safety issues in restaurants use similar educational information to effectively educate consumers on the seriousness of food safety risks in restaurants. The findings of the hypotheses test are summarized in Table 6.1. 40 Table 6.1: Summary of hypothesis and results Hypothesis Empirical results Outcome H1 Consumers are willing to pay more for higher food safety grade (vs. lower) Consumers are willing to pay relatively $5.56 more for grade A over B, and $15.82 relatively more for grade A over C Supported H2a Consumers are willing to pay more when additional information about food safety score is provided (general or critical vs unknown) Consumers are willing to pay relatively $2.06 more for general violation over unknown violation Consumers are willing to pay relatively $5.17 less for critical violation over unknown violation Rejected H2b Consumers are willing to pay more when additional information about food safety score is perceived to be better (general vs. critical) Consumers are willing to pay relatively $7.23 more for the general over critical violations Supported H3 Consumers who are exposed restaurant food safety risks information will exhibit a greater willingness to pay for restaurants with highest letter grade Consumers are willing to pay relatively more for grade A over grades B and C in the treatment condition Supported 6.2 Theoretical Implications This thesis analyzes consumers' assessments of the elements of a food safety inspection poster and how each attribute impacts their choice of restaurants. Thus, this research contributes to the existing hygiene/ food safety literature. This study investigates consumers' restaurant choices based on food safety information in a monetary value. Several literature look at the impact of food safety/ 41 hygiene scorecards through the lenses of psychological value; for example, Choi et al. (2011) explored how the number of violations affects consumers' intention to change behaviors through the Protective Motivation framework. One study similar to this thesis is the study by Henson et al. (2006), which uses different types of food safety-related characteristics to test how consumers assess food safety in food establishments. However, they only touch on the visible indicators related to food safety in restaurants. Though WTP is widely used in hospitality and applied economic papers, this study is the first to apply the concept to the food safety elements on a food safety inspection poster. One literature that applies a similar empirical methodology to compare WTP between consumers and citizens when it comes to food safety risks in restaurants found that both groups exhibit higher WTP for reduced food safety risks (Alphonce et al., 2014). The difference between that study and this one is the specification. The previous study concerns the differences between individual roles associated with food safety risk reduction, while this one focuses on how consumers value each specific attribute. 6.3 Policy and Managerial Implications This study has implications for policymakers and fast-casual restaurant managers and owners. Overall, the research findings suggest that consumers value high food safety letter grades, but having the type of violations indicated provides mixed results. Many jurisdictions that have already conducted food safety inspections can adopt the use and display of the food safety inspection card. The local government can explore additional indicators, in addition to the letter grade on the card, as a penalty for restaurants that do not adhere to food safety practices. However, these adoptions should 42 be exercised with caution, and communication of the information listed on the card should be thoroughly explained. Finally, since consumers value the additional information when perceived to be more positive (higher relative willingness to pay for general violation type than critical), self-explanatory information behind letter grade could be added to explain the sanitary evaluation. 6.4 Limitations and suggestions for future research This thesis has several limitations, which provide opportunities for future studies to obtain more robust and generalizable findings. First is the setting of the experiment. The Asian fast-casual restaurant was selected for the experiment, so the result may not be generalizable to other types of restaurants, such as fine dining and other ethnic quick-service restaurants. One suggestion is to have the type of restaurants as a between-subject variation to see the differences in responses. This study was conducted using an incentivized online experiment. Adding the in-person incentivized experiment would change the analysis of the results from hypothetical to applicable results in the real-world setting. Relating to the type of restaurant, the study chose udon as a representative dish; the results may not be generalizable to other dishes. Udon is a unique dish. Even though a photo and description were shown to participants, there might be some unfamiliarity. We could add a pre-screening question or survey to ask about familiarity with the dish before allowing participants to proceed to the main survey. Another pivot on dishes selected to consider is the ingredients and cooking method. In future research, it would 43 be interesting to test if consumers would respond differently toward uncooked dishes (e.g., salad, sashimi, or raw fish) or partially uncooked (e.g., hamburger, poke bowl). The second limitation is the characteristics chosen in the choice set. This study focused on two elements of food safety inspection results: food safety letter grade and type of violations. The letter grading system is a straightforward measurement that is commonly used and familiar to people. It would be interesting to test the threshold effect between two different grading formats: numeric vs. letter grades. The wording of the type of violation could also be changed. In this study, the unknown violation was chosen to describe the unavailability of violation type on the resulting poster, which could be confusing for participants to understand. Another problem is the possibility of combination, which was not well-explained in the study. For example, it may not be clear to participants that a critical violation does not eliminate the possibility of a restaurant having a grade A. These problems could be handled through an in-person experiment, where participants can ask clarifying questions before beginning the experiment. The last limitation is in the dining habits and behaviors section. We did not ask in-depth questions about participants' prior food safety knowledge and how the information treatment was perceived. In the future, it would be better to get more information on the topic prior to the beginning of the survey about how much consumers know about food safety inspection or results. 44 CHAPTER 7 REFERENCES Alfanes, F. (2004). Stated preferences for imported and hormone-treated beef: application of a mixed logit model. European Review of Agricultural Economics, 31(1), 19-37. https://doi.org/10.1093/erae/31.1.19 Alphonce, R., Alfnes, F., & Sharma, A. (2014). Consumer vs. Citizen willingness to pay for restaurant food safety. Food Policy, 49, 160–166. https://doi.org/10.1016/j.foodpol.2014.06.009 Andaleeb, S. S., & Conway, C. (2006). Customer satisfaction in the restaurant industry: An examination of the transaction-specific model. Journal of Services Marketing, 20(1), 3-11. https://doi.org/10.1108/08876040610646536 Barber, N. & Scarcelli, J. (2009). Clean restrooms: how important are they to restaurant consumer? Journal of Foodservice, 20(6), 309-320. https://doi.org/10.1111/j.1748- 0159.2009.00155.x Binesh, F., Belarmino, A., & Bai, Y. (Mabel). (2023). Exploring which factors impact restaurant willingness-to-pay by menu course. Journal of Foodservice Business Research, 1–24. https://doi.org/10.1080/15378020.2023.2281196 Breidert, C., Hahsler, M. & Reutterer, T. (2006). A Review of Methods for Measuring Willingness-To-Pay. Innovative Marketing, 2(4), 8-31. Campbell-Smite, G. (1970). Marketing the Meal Experience. Cornell Hospitality Quarterly, 11(1), 73-102. https://doi.org/10.1177/001088047001100116 Canny, I.U. (2014). Measuring the Mediating Role of Dining Experience Attributes on Customer Satisfaction and Its Impact on Behavioral Intentions of Casual Dining Restaurant in Jakarta. Internation Journal of Innovation, Management and Technology, 5(1), 25-29. https://doi.org/10.7763/IJIMT.2014.V5.480 Caswell, J. A. (1998). How Labeling of Safety and Process Attributes Affects Markets for Food. Agricultural and Resource Economics Review, 27(2), 151–158. https://doi.org/10.1017/S106828050000647X https://doi.org/10.1093/erae/31.1.19 https://doi.org/10.1016/j.foodpol.2014.06.009 https://doi.org/10.1108/08876040610646536 https://doi.org/10.1111/j.1748-0159.2009.00155.x https://doi.org/10.1111/j.1748-0159.2009.00155.x https://doi.org/10.1080/15378020.2023.2281196 https://doi.org/10.1177/001088047001100116 https://doi.org/10.7763/IJIMT.2014.V5.480 https://doi.org/10.1017/S106828050000647X 45 Choi, J., Nelson, D. C., & Almanza, B. (2011). The impact of inspection reports on consumer behavior: A pilot study. Food Control, 22(6), 862–868. https://doi.org/10.1016/j.foodcont.2010.11.007 Covello, V. T., McCallum, D. B., & Pavlova, M. T. (1987). Effective risk communication: The role and responsibility of government and nongovernment organizations. New York and London: Plenum Press. Duarte Alonso, A., O’neil, M., Liu, Y. & O’shea, M. (2013). Factors Driving Consumer Restaurant Choice: An Exploratory Study From the Southeastern United States. Journal of Hospitality Marketing & Management, 22(5), 547-567. https://doi.org/10.1080/19368623.2012.671562 Endara, P., Wiedmann, M., & Adalja, A. (2023). Consumer willingness to pay for shelf life of high-temperature, short-time-pasteurized fluid milk: Implications for smart labeling and food waste reduction. Journal of Dairy Science, 106(9), 5940–5957. https://doi.org/10.3168/jds.2022-22968 Firestone, M. J., & Hedberg, C. W. (2018). Restaurant Inspection Letter Grades and Salmonella Infections, New York, New York, USA. Emerging infectious diseases, 24(12), 2164–2168. https://doi.org/10.3201/eid2412.180544 Friedel, J., Foreman, A. & Wirth, O. (2022). An introduction to “discrete choice experiments” for behavior analysts. Behavioral Processes, 198, https://doi.org/10.1016/j.beproc.2022.104628 Ha, J., & Jang, S. S. (2013). Attributes, consequences, and consumer values. International Journal of Contemporary Hospitality Management. Ha, J. & Jang, S.S. (2013). Attributes, consequences, and consumer values. International Journal of Contemporary Hospitality Management, 25(3), 383-409, https://doi.org/10.1108/09596111311311035 Hanck, C., Arnold,M., Gerber, A., & Schmelzer M. (2024). Introduction to Econometrics with R. https://www.econometrics-with-r.org/ITER.pdf Greene, W. (2012) Econometric Analysis. 7th Edition, Prentice Hall, Upper Saddle River. Jeong, E. & Jang, S.S. (2011). Restaurant experiences triggering positive electronic word- of-mount (eWOM) motivations. International Journal of Hospitality Management, 30(2), 356-366. https://doi.org/10.1016/j.ijhm.2010.08.005 https://doi.org/10.1016/j.foodcont.2010.11.007 https://doi.org/10.1080/19368623.2012.671562 https://doi.org/10.3168/jds.2022-22968 https://doi.org/10.3201/eid2412.180544 https://doi.org/10.1016/j.beproc.2022.104628 https://doi.org/10.1108/09596111311311035 https://www.econometrics-with-r.org/ITER.pdf https://doi.org/10.1016/j.ijhm.2010.08.005 46 Jin, G., & Leslie, P. (2003). The Effect of Information on Product Quality: Evidence from Restaurant Hygiene Grade Cards. The Quarterly Journal of Economics, 118(2), 409- 451, https://doi.org/10.1162/003355303321675428 Orme B. & Chrzen, K. (2017). Becoming an Expert in Conjoint Analysis: Choice Modeling for Pros. Sawtooth Software, inc. Kemperman, A. (2021). A review of research into discrete choice experiments in tourism: Launching the Annals of Tourism Research Curated Collection on Discrete Choice Experiments in Tourism. Annals of Toursim Research, 87, https://doi.org/10.1016/j.annals.2020.103137 Knight, A. J., Worosz, M. R., & Todd, E. C. D. (2007). Serving food safety: Consumer perceptions of food safety at restaurants. International Journal of Contemporary Hospitality Management, 19(6), 476–484. https://doi.org/10.1108/09596110710775138 Kotler, P. and Keller, K.L. (2012) Marketing Management. 14th Edition, Pearson Education. McFadden, D. (1974) Conditional Logit Analysis of Qualitative Choice Behavior. Frontiers in Econometrics, 105-142. Mangham, L., Hanson, K. & McPake, B. (2009). How to do (or not to do) ... Designing a discrete choice experiment for application in a low-income country. Health Policy Plan, 24(2), 151-158. https://doi.org/10.1093/heapol/czn047 Mathayomcham, B. & Taecharungroj, V. (2020). “How was your meal?” Examining customer experiemce using Google maps reviews. Internation Journal of Hospitality Management, 90. https://doi.org/10.1016/j.ijhm.2020.102641 Mattila, A. (2001). Emotional bonding and restaurant loyalty. The Cornell Hotel and Restaurant Administration Quarterly, 42(6), 73–79. https://doi.org/10.1016/S0010- 8804(01)81012-0 Mwove, J., Imathiu, S, Orina, I. & Karanja, P. (2020). Food safety knowledge and practices of street food vendors in selected locations within Kiambu County, Kenya. African Journal of Food Science, 14(6), 174-185. https://doi.org/10.5897/AJFS2020.1929 https://doi.org/10.1162/003355303321675428 https://doi.org/10.1016/j.annals.2020.103137 https://doi.org/10.1108/09596110710775138 https://doi.org/10.1093/heapol/czn047 https://doi.org/10.1016/j.ijhm.2020.102641 https://doi.org/10.1016/S0010-8804(01)81012-0 https://doi.org/10.1016/S0010-8804(01)81012-0 https://doi.org/10.5897/AJFS2020.1929 47 Namkung, Y., & Jang, S. (2007). Does Food Quality Really Matter in Restaurants? Its Impact On Customer Satisfaction and Behavioral Intentions. Journal of Hospitality & Tourism Research, 31(3), 387–409. https://doi.org/10.1177/1096348007299924 Namkung, Y., & Jang, S. (Shawn). (2014). Are Consumers Willing to Pay more for Green Practices at Restaurants? Journal of Hospitality & Tourism Research, 41(3), 329-356. https://doi.org/10.1177/1096348014525632 National Research Council (US) Committee on the Review of Food and Drug Administration's Role in Ensuring Safe Food, Wallace, R. B., & Oria, M. (Eds.). (2010). Enhancing Food Safety: The Role of the Food and Drug Administration. National Academies Press (US). Nayga, R. (1999). Toward an understanding of consumers’ perceptions of food labels. The International Food and Agribusiness Management Review, 2(1), 29–45. https://doi.org/10.1016/S1096-7508(99)00011-7 Nelson, P. (1970) Information and Consumer Behavior. Journal of Political Economy, 78, 311-329. http://dx.doi.org/10.1086/259630 Nelson, P. (1974) The Economic Value of Advertising. In: Brozen, Y., Ed., Advertising and Society, New York University Press, New York, 43-66. NYC Health. (2016). What to Expect When You’re Inspected: A Guide for Food Service Operators. NYC Health. (2016). Self-Inspection Worksheet for Food Service Establishments. New York City Department of Health and Mental Hygiene Parsa, H.G., Dutta, K. & Njite, D. (2017). Consumer Behavior in Restaurants: Assessing the Importance of Restaurant Attributes in Consumer Patronage and Willingess to Pay. Hospitality Marketing and Consumer Behavior (pp.213-236). Apple Academic Press. Pham, M.T. & Higgins, E. (2004). Promotion and Prevention in Consumer Decision Making: State of the Art and Theoretical Propositions. Inside Consumption: Consumer Motives, Goals, and Desires. https://doi.org/10.4324/9780203481295 Qualtrics. (2025). Conjoint Analysis Security Statement. https://www.qualtrics.com/support/conjoint-project/getting-started-conjoints/getting- started-choice-based/conjoint-analysis-white-paper/#ExperimentalDesign https://doi.org/10.1177/1096348007299924 https://doi.org/10.1177/1096348014525632 https://doi.org/10.1016/S1096-7508(99)00011-7 http://dx.doi.org/10.1086/259630 https://doi.org/10.4324/9780203481295 https://www.qualtrics.com/support/conjoint-project/getting-started-conjoints/getting-started-choice-based/conjoint-analysis-white-paper/#ExperimentalDesign https://www.qualtrics.com/support/conjoint-project/getting-started-conjoints/getting-started-choice-based/conjoint-analysis-white-paper/#ExperimentalDesign 48 Rawlins, B. (2008). Measuring the relationship between organizational transparency and employee trust. Public Relations Journal, 2(2), 1-21. Recio-Román, A., Recio-Menéndez, M., & Román-González, M. V. (2020). Food Reward and Food Choice. An Inquiry Through The Liking and Wanting Model. Nutrients, 12(3), 639. https://doi.org/10.3390/nu12030639 Sarrias, MA and RA Daziano. 2017. “Multinomial Logit Models with Continuous and Discrete Individual Heterogeneity in R: The gmnl Package.” Journal of Statistical Software, 79(2), 1-46. http://dx.doi.org/10.2139/ssrn.3065365 Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simmons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61, 195–202. Shin, Y. H., Im, J., Jung, S. E., & Severt, K. (2017). Locally Sourced Restaurant: Consumers Willingness to Pay. Journal of Foodservice Business Research, 21(1), 68– 82. https://doi.org/10.1080/15378020.2016.1276319 Simon, P. A., Leslie, P., Run, G., Jin, G. Z., Reporter, R., Aguirre, A., & Fielding, J. E. (2005). Impact of restaurant hygiene grade cards on foodborne-disease hospitalizations in Los Angeles County. Journal of environmental health, 67(7), 32– 60. Sinclair, W., Rivera-Cintron, D. & Zeballos, E. (2024 ). U.S. Consumers Increased Spending on Food Away From Home in 2023, Driving Overall Food Spending Growth. U.S. Department of Agriculture, Economic Research Service. https://www.ers.usda.gov/amber-waves/2024/october/u-s-consumers-increased- spending-on-food-away-from-home-in-2023-driving-overall-food-spending-growth Strulov-Shlain, A. (2023). More Than a Penny’s Worth: Left-Digit Bias and Firm Pricing. The Review of Economic Studies, 90(5), 2612-2645. https://doi.org/10.1093/restud/rdac082 Sulek, J.M. & Hensley, R.L. (2004). The Relative Importance of Food, Atmosphere, and Fairness of Wait: The Case of a Full-service Restaurant. Cornell Hospitality Quarterly, 45(3), 235-247. https://doi.org/10.1177/0010880404265345 https://doi.org/10.3390/nu12030639 http://dx.doi.org/10.2139/ssrn.3065365 https://doi.org/10.1080/15378020.2016.1276319 https://www.ers.usda.gov/amber-waves/2024/october/u-s-consumers-increased-spending-on-food-away-from-home-in-2023-driving-overall-food-spending-growth https://www.ers.usda.gov/amber-waves/2024/october/u-s-consumers-increased-spending-on-food-away-from-home-in-2023-driving-overall-food-spending-growth https://doi.org/10.1093/restud/rdac082 https://doi.org/10.1177/0010880404265345 49 Thurstone, L. L. (1927). A law of comparative judgment. Psychological Review, 34(4), 273–286. https://doi.org/10.1037/h0070288 Tompkins County (2024). Food Program: Food Service Inspection Results. Tompkins County. https://www.tompkinscountyny.gov/All-Departments/Whole- Health/Environmental-Health-Division/Food-Program Train, K. (2009) Discrete Choice Methods with Simulation. Cambridge University Press, New York. http://dx.doi.org/10.1017/CBO9780511805271 Umberger, W. & Mueller, S. (2010). Is Presentation Everything? Using Visual Presentation of Attributes in Discrete Choice Experiments to Measure the Relative Importance of Intrinsic and Extrinsic Beef Attributes. 2010 Annual Meeting, July 25-27, 2010, Denver, Colorado 61856, Agricultural and Applied Economics Association. Willis, K. (2014). The Use of Stated Preference Methods to Value Cultural Heritage. Handbook of the Economics of Art and Culture, 2, 145-181. https://doi.org/10.1016/B978-0-444-53776-8.00007-6 Wongprawmas, R., & Canavari, M. (2017). Consumers’ willingness-to-pay for food safety labels in an emerging market: The case of fresh produce in Thailand. Food Policy, 69, 25–34. https://doi.org/10.1016/j.foodpol.2017.03.004 Zeithaml, V.A. (1988). Consumer Perceptions of Price, Quality, and Value: A Means-End Model and Synthesis of Evidence. Journal of Marketing, 52(3), 2-22. https://doi.org/10.2307/1251446 https://doi.org/10.1037/h0070288 https://www.tompkinscountyny.gov/All-Departments/Whole-Health/Environmental-Health-Division/Food-Program https://www.tompkinscountyny.gov/All-Departments/Whole-Health/Environmental-Health-Division/Food-Program http://dx.doi.org/10.1017/CBO9780511805271 https://doi.org/10.1016/B978-0-444-53776-8.00007-6 https://doi.org/10.1016/j.foodpol.2017.03.004 https://doi.org/10.2307/1251446 50 APPENDIX Appendix 1: Institutional review board for human participants: Notice of exemption 51 Appendix 2: Survey 52 53 54 55 56 57 58 59 60 61 Appendix 3: 2020 U.S. Census Data Fact Fact Note United States Population estimates, July 1, 2024, (V2024) 340,110,988 Population estimates, July 1, 2023, (V2023) 334,914,895 Population estimates base, April 1, 2020, (V2024) 331,515,736 Population estimates base, April 1, 2020, (V2023) 331,464,948 Population, percent change - April 1, 2020 (estimates base) to July 1, 2024, (V2024) 2.6% Population, percent change - April 1, 2020 (estimates base) to July 1, 2023, (V2023) 1.0% Population, Census, April 1, 2020 331,449,281 Population, Census, April 1, 2010 308,745,538 Persons under 5 years, percent 5.5% Persons under 18 years, percent 21.7% Persons 65 years and over, percent 17.7% Female persons, percent 50.5% White alone, percent 75.3% Black alone, percent (a) (a) 13.7% American Indian and Alaska Native alone, percent (a) (a) 1.3% Asian alone, percent (a) (a) 6.4% Native Hawaiian and Other Pacific Islander alone, percent (a) (a) 0.3% Two or More Races, percent 3.1% Hispanic or Latino, percent (b) (b) 19.5% White alone, not Hispanic or Latino, percent 58.4% Households, 2019-2023 127,482,865 Persons per household, 2019-2023 2.54 High school graduate or higher, percent of persons age 25 years+, 2019-2023 89.4% Bachelor's degree or higher, percent of persons age 25 years+, 2019- 2023 35.0% Median households income (in 2023 dollars), 2019-2023 $78,538.00 Per capita income in past 12 months (in 2023 dollars), 2019-2023 $43,289.00 Persons in poverty, percent 11.1% FIPS Code "1" 62 NOTE: FIPS Code values are enclosed in quotes to ensure leading zeros remain intact. Value Notes None Fact Notes (a) Includes persons reporting only one race (b) Hispanics may be of any race, so also are included in applicable race categories N Data for this geographic area cannot be displayed because the number of sample cases is too small. 63 64 Appendix 4: Behavioral characteristics Characteristic Overall N = 4891 Control N = 2391 Treatment N = 2501 Dine Out Frequency Lunch Weekday Hardly ever 163 (33%) 75 (31%) 88 (35%) Once a month 83 (17%) 49 (21%) 34 (14%) 2-3 times a month 78 (16%) 38 (16%) 40 (16%) Once a week 71 (15%) 35 (15%) 36 (14%) A few times a week 70 (14%) 30 (13%) 40 (16%) Everyday 23 (4.7%) 11 (4.6%) 12 (4.8%) Dinner Weekday Hardly ever 89 (18%) 37 (15%) 52 (21%) Once a month 85 (17%) 50 (21%) 35 (14%) 2-3 times a month 120 (25%) 61 (26%) 59 (24%) Once a week 112 (23%) 57 (24%) 55 (22%) A few times a week 64 (13%) 23 (9.6%) 41 (16%) Everyday 18 (3.7%) 10 (4.2%) 8 (3.2%) Lunch Weekend Hardly ever 134 (27%) 56 (23%) 78 (31%) Once a month 104 (21%) 58 (24%) 46 (18%) 2-3 times a month 112 (23%) 55 (23%) 57 (23%) Once a week 81 (17%) 40 (17%) 41 (16%) A few times a week 38 (7.8%) 17 (7.1%) 21 (8.4%) Everyday 19 (3.9%) 12 (5.0%) 7 (2.8%) Dinner Weekend Hardly ever 79 (16%) 33 (14%) 46 (18%) Once a month 106 (22%) 55 (23%) 51 (20%) 2-3 times a month 128 (26%) 67 (28%) 61 (24%) Once a week 110 (22%) 54 (23%) 56 (22%) A few times a week 46 (9.4%) 18 (7.5%) 28 (11%) Everyday 19 (3.9%) 11 (4.6%) 8 (3.2%) Average Check $10 or less 41 (8.4%) 21 (8.8%) 20 (8.0%) $11 - $20 300 (61%) 151 (63%) 149 (60%) $21 - $30 102 (21%) 45 (19%) 57 (23%) $31 - $40 27 (5.5%) 11 (4.6%) 16 (6.4%) $41 - $50 11 (2.2%) 6 (2.5%) 5 (2.0%) $50 or more 6 (1.2%) 4 (1.7%) 2 (0.8%) 65 Characteristic Overall N = 4891 Control N = 2391 Treatment N = 2501 Ordering medium frequency Third party delivering service Less than once a week 356 (73%) 171 (72%) 185 (74%) Once a week 58 (12%) 30 (13%) 28 (11%) 2-3 times a week 50 (10%) 25 (10%) 25 (10%) 4-5 times a week 12 (2.5%) 7 (2.9%) 5 (2.0%) Everyday 8 (1.6%) 5 (2.1%) 3 (1.2%) Order online for a pick up order Less than once a week 273 (56%) 129 (54%) 144 (58%) Once a week 136 (28%) 68 (28%) 68 (27%) 2-3 times a week 47 (9.6%) 25 (10%) 22 (8.8%) 4-5 times a week 21 (4.3%) 11 (4.6%) 10 (4.0%) Everyday 8 (1.6%) 5 (2.1%) 3 (1.2%) Order in-store and dine-in Less than once a week 245 (50%) 128 (54%) 117 (47%) Once a week 150 (31%) 68 (28%) 82 (33%) 2-3 times a week 64 (13%) 29 (12%) 35 (14%) 4-5 times a week 14 (2.9%) 8 (3.3%) 6 (2.4%) Everyday 12 (2.5%) 5 (2.1%) 7 (2.8%) Order in-store and takeaway Less than once a week 226 (46%) 113 (47%) 113 (45%) Once a week 158 (32%) 84 (35%) 74 (30%) 2-3 times a week 67 (14%) 30 (13%) 37 (15%) 4-5 times a week 27 (5.5%) 9 (3.8%) 18 (7.2%) Everyday 6 (1.2%) 2 (0.8%) 4 (1.6%) Method used to find new restaurants (ranking 6 items) Google Maps 1 227 (46%) 113 (47%) 114 (46%) 2 91 (19%) 44 (18%) 47 (19%) 3 64 (13%) 34 (14%) 30 (12%) 4 53 (11%) 20 (8.4%) 33 (13%) 5 40 (8.2%) 21 (8.8%) 19 (7.6%) 6 14 (2.9%) 7 (2.9%) 7 (2.8%) Yelp/ TripAdvisor or similar platform 1 62 (13%) 30 (13%) 32 (13%) 2 99 (20%) 51 (21%) 48 (19%) 3 86 (18%) 37 (15%) 49 (20%) 4 99 (20%) 50 (21%) 49 (20%) 66 Characteristic Overall N = 4891 Control N = 2391 Treatment N = 2501 5 91 (19%) 43 (18%) 48 (19%) 6 52 (11%) 28 (12%) 24 (9.6%) Word of Mouth/ Recommendations from friends or locals 1 115 (24%) 45 (19%) 70 (28%) 2 131 (27%) 70 (29%) 61 (24%) 3 126 (26%) 64 (27%) 62 (25%) 4 65 (13%) 33 (14%) 32 (13%) 5 38 (7.8%) 19 (7.9%) 19 (7.6%) 6 14 (2.9%) 8 (3.3%) 6 (2.4%) Delivery app suggestion 1 23 (4.7%) 15 (6.3%) 8 (3.2%) 2 46 (9.4%) 20 (8.4%) 26 (10%) 3 49 (10%) 24 (10%) 25 (10%) 4 81 (17%) 43 (18%) 38 (15%) 5 125 (26%) 63 (26%) 62 (25%) 6 165 (34%) 74 (31%) 91 (36%) Wandering around 1 27 (5.5%) 18 (7.5%) 9 (3.6%) 2 66 (13%) 26 (11%) 40 (16%) 3 97 (20%) 49 (21%) 48 (19%) 4 103 (21%) 52 (22%) 51 (20%) 5 89 (18%) 42 (18%) 47 (19%) 6 107 (22%) 52 (22%) 55 (22%) Social media – TikTok/ Instagram/ Facebook/ X 1 35 (7.2%) 18 (7.5%) 17 (6.8%) 2 56 (11%) 28 (12%) 28 (11%) 3 67 (14%) 31 (13%) 36 (14%) 4 88 (18%) 41 (17%) 47 (19%) 5 106 (22%) 51 (21%) 55 (22%) 6 137 (28%) 70 (29%) 67 (27%) Importance of dining attributes Taste Extremely important 360 (74%) 175 (73%) 185 (74%) Very important 106 (22%) 49 (21%) 57 (23%) Moderately important 18 (3.7%) 11 (4.6%) 7 (2.8%) Slightly important 3 (0.6%) 3 (1.3%) 0 (0%) Price Extremely important 182 (37%) 89 (37%) 93 (37%) 67 Characteristic Overall N = 4891 Control N = 2391 Treatment N = 2501 Very important 182 (37%) 90 (38%) 92 (37%) Moderately important 100 (20%) 48 (20%) 52 (21%) Slightly important 20 (4.1%) 10 (4.2%) 10 (4.0%) Not at all important 3 (0.6%) 1 (0.4%) 2 (0.8%) Quality of service Extremely important 160 (33%) 78 (33%) 82 (33%) Very important 189 (39%) 82 (34%) 107 (43%) Moderately important 116 (24%) 66 (28%) 50 (20%) Slightly important 18 (3.7%) 9 (3.8%) 9 (3.6%) Not at all important 1 (0.2%) 1 (0.4%) 0 (0%) Convenience (time and location) Extremely important 146 (30%) 63 (26%) 83 (33%) Very important 206 (42%) 100 (42%) 106 (42%) Moderately important 113 (23%) 59 (25%) 54 (22%) Slightly important 19 (3.9%) 13 (5.4%) 6 (2.4%) Not at all important 3 (0.6%) 2 (0.8%) 1 (0.4%) Availability of food items Extremely important 152 (31%) 68 (28%) 84 (34%) Very important 213 (44%) 113 (47%) 100 (40%) Moderately important 93 (19%) 44 (18%) 49 (20%) Slightly important 24 (4.9%) 9 (3.8%) 15 (6.0%) Not at all important 4 (0.8%) 3 (1.3%) 1 (0.4%) Cleanliness of the store Extremely important 262 (54%) 125 (52%) 137 (55%) Very important 151 (31%) 68 (28%) 83 (33%) Moderately important 61 (12%) 36 (15%) 25 (10%) Slightly important 6 (1.2%) 4 (1.7%) 2 (0.8%) Not at all important 1 (0.2%) 0 (0%) 1 (0.4%) Health/ safety aspects of food items Extremely important 275 (56%) 130 (54%) 145 (58%) Very important 128 (26%) 61 (26%) 67 (27%) Moderately important 56 (11%) 31 (13%) 25 (10%) Slightly important 16 (3.3%) 9 (3.8%) 7 (2.8%) Not at all important 3 (0.6%) 2 (0.8%) 1 (0.4%) 1n (%)