!     AN EXAMINATION OF US HOUSEHOLD EXPENDITURES ON HEALTHY FOOD  By  Andrea Leschewski                   A DISSERTATION   Submitted to Michigan State University in partial fulfillment of the requirements for the degree of  Agricultural, Food and Resource Economics ÑDoctor of Philosophy  2016          !ABSTRACT  AN EXAMINATION OF US HOUSEHOLD EXPENDITURES ON HEALTHY FOOD  By  Andrea Leschewski  This dissertation contains three chapters, each of which examine US household expenditures on healthy food. In the first chapter, determinants of householdsÕ expenditures on healthy food away from home (FAFH) are analyzed using the National Household Food Acquisition and Purchase Survey. For households purchasing FAFH, CraggÕs double-hurdle model is used to analyze their decision to purchase healthy FAFH and the share of total FAFH expenditures to allocate to healthy FAFH. Results indicate households receiving SNAP food assistance benefits are less likely to purchase healthy FAFH when dining away from home and allocate less of their total FAFH dollars to healthy items. In contrast, the healthy FAFH participation and expenditure shares of low-income households not receiving SNAP benefits do not significantly differ from those of high-income households. Other significant findings include that healthy FAFH participation and expenditure shares vary with the healthiness of householdsÕ food at home purchases, FAFH retailer type, nutritional information and time constraints, as well as other basic demographic factors.  The second chapter examines whether specific nutrients garner price premiums in fruit beverages sold in the US. Using the National Household Food Acquisition and Purchase Survey, hedonic price models for fruit juice and fruit drinks are estimated to determine whether specific nutrients, product characteristics, packaging type and acquisition characteristics are associated with price premiums. Based on the results from !the hedonic price models, three generalizations are made about the price premiums for nutrients and sugar in fruit beverages: (1) all nutrients garner premium prices in fruit juice, (2) sugar and select nutrients garner price premiums in non-diet fruit drinks and (3) all nutrients and sugar are associated with price discounts in diet fruit drinks. Findings further suggest that product attributes such as brand, flavor, organic labels, diet labels and package type, and acquisition characteristics such as store type, region, season and payment type are associated with price premiums in fruit beverages. The final chapter develops a group-based food diversity index, representative of diversity in household expenditures across food subgroups. This index is compared to the traditional product code-based food diversity index and applied to reassess expenditure and demographic determinants of food diversity demand. Results confirm that the group and product code indices capture different forms of food diversity. The indices are only moderately correlated and have varying means and skewness. Education, gender, age, household size, race, SNAP and food expenditures are found to significantly affect food diversity. However, the magnitude and direction of the effects vary between group and product code indices. Given these differences, it is essential that studies select a diversity index that corresponds to their objective. Results suggest that group-based indices are appropriate for informing food and nutrition policy, while product code-based indices are ideal for guiding food industry management decision-making.     !    Copyright by ANDREA LESCHEWSKI 2016                                   !"!                     I dedicate this dissertation to my father, Robert Leschewski, and my mother, Sandra Ryan.  Thank you for your constant love, support and encouragement, and for all of the sacrifices you made so that I could pursue my education.                       !"#!ACKNOWLEDGMENTS   First and foremost, thank you to my advisor Dr. Dave Weatherspoon for your mentorship over the past five years. I am grateful for the time and effort you spent helping shape my research interests, guiding me in my future career path and for always pushing me to overcome my weaknesses. Learning from and working with you has been the highlight of my graduate studies. I also wish to thank the members of my dissertation committee, Dr. Steven Miller, Dr. Brent Ross and Dr. Lorraine Weatherspoon, for their comments and suggestion for improving my dissertation. A special thanks to Dr. Miller for funding my graduate studies - thank you for always having your door open and your constant willingness to help address my questions. Also, a special thanks to Dr. Ross for the opportunity to TA your classes - your mentorship in using case studies in the classroom has been invaluable.  Thank you to the USDA-ERS, Michign AgBioResearch and NIFA Hatch Project #MICL02390 for funding my dissertation research. Also, a special thanks to Annemarie Kuhns of the USDA-ERS, for your guidance and help using the FoodAPS dataset. I would also like to thank the members of my graduate research group, Marie Steele, Gerard Taylor, Ben Sombatthira, Jongwoo Kim and Thasanee Satimanon, for their support and suggestions for improving my research. I am also indebted to the support of my graduate cohort, Serge Adjognon, Piyayut Chitchimnong, Jason Snyder, Joey Goeb, Mukesh Ray and Sarah Kopper. Finally thank you to my husband Brian, my parents, my extended family, Angela and Dale, who were my support system through this process and a needed reminder to enjoy life outside of my graduate studies.  !"##!TABLE OF CONTENTS   LIST OF TABLES ............................................................................................................. ix  LIST OF FIGURES .............................................................................................................x  KEY TO ABBREVIATIONS ............................................................................................ xi  CHAPTER 1 DETERMINANTS OF HOUSEHOLD EXPENDITURES ON           HEALTHY FOOD AWAY FROM HOME ............................................................1 1.1 Introduction .................................................................................................................1 1.2 Data .............................................................................................................................3 1.3 Theoretical Framework ...............................................................................................5 1.4 Methodology ...............................................................................................................6 1.4.1 The Zero Expenditure Problem ..........................................................................6 1.4.2 CraggÕs Double-Hurdle Model ..........................................................................8 1.4.3 Description of Explanatory Variables ................................................................9 1.5 Results .......................................................................................................................13 1.5.1 Descriptive Statistics ........................................................................................13 1.5.2 Double-Hurdle Model Estimates .....................................................................15  1.6 Discussion .................................................................................................................19  CHAPTER 2 A SEGMENTED HEDONIC ANALYSIS OF THE NUTRITIONAL COMPOSITION OF FRUIT BEVERAGES .........................................................23    2.1 Introduction ...............................................................................................................23    2.2 Background ...............................................................................................................25       2.2.1 US Fruit Beverage Industry ................................................................................25       2.2.2 Nutritional Composition of Fruit Beverages .......................................................25       2.2.3 Federal Programs, Policies and Guidelines Concerning Fruit Beverages ..........26       2.2.4 Changing Consumer Demand for Fruit Beverages .............................................28       2.2.5 Uniqueness of this Study ....................................................................................29    2.3 Hedonic Pricing Model .............................................................................................31       2.3.1 Hedonic Price Theory .........................................................................................31       2.3.2 Hedonic Price Model ..........................................................................................31       2.3.3 Application of Hedonic Price Model to Fruit Beverages ....................................33     2.4 Data ...........................................................................................................................37       2.4.1 Data Set ...............................................................................................................37       2.4.2 Descriptive Statistics ...........................................................................................38     2.5 Results .......................................................................................................................42       2.5.1 Nutrients ..............................................................................................................42       2.5.2 Product Attributes ...............................................................................................45       2.5.3 Packaging ............................................................................................................46       2.5.4 Acquisition Attributes .........................................................................................47    2.6 Conclusions ...............................................................................................................48 !"###!CHAPTER 3 RETHINKING HOUSEHOLD DEMAND FOR FOOD DIVERSITY ......50    3.1 Introduction ...............................................................................................................50    3.2 Literature Review ......................................................................................................52    3.3 Theoretical Background ............................................................................................53    3.4 An Alternative Group-Based Food Diversity Index .................................................54    3.5 Empirical Model .......................................................................................................60    3.6 Data ...........................................................................................................................62    3.7 Results .......................................................................................................................63       3.7.1 Descriptive Statistics ...........................................................................................63       3.7.2 Regression Results ..............................................................................................66       3.7.3 Expenditures .......................................................................................................67       3.7.4 Demographics .....................................................................................................68    3.8 Conclusions ...............................................................................................................71  APPENDICES ...................................................................................................................74    APPENDIX A Supplementary Material for Chapter 1 ...................................................75    APPENDIX B Supplementary Material for Chapter 2 ...................................................77    APPENDIX C Supplementary Material for Chapter 3 ...................................................78  REFERENCES ..................................................................................................................79                           !#$!LIST OF TABLES    Table 1.1 Top 10 Healthy FAFH Items by Retailer Type (N=22,617) ...............................5   Table 1.2 Vuong Model Selection Test: Censored Tobit vs CraggÕs Double Hurdle .........7  Table 1.3 Variable Definitions ...........................................................................................10  Table 1.4 Descriptive Statistics for Participating and Non-Participating Households ......14  Table 1.5 Double-Hurdle Model Elasticities and Marginal Effects (N=3,894) .................17  Table 2.1 Top Fruit Beverage Front-of-Package Nutrition Labels ....................................29   Table 2.2 Variable Definitions ...........................................................................................34  Table 2.3 Descriptive Statistics By Fruit Beverage Type ..................................................39  Table 2.4 Fruit Beverage Hedonic Price Function Estimates ............................................43  Table 3.1 Share of Food Subgroups Comprised by the Product Code IndexÕs Basic and Convenience Food Categories ...........................................................................59  Table 3.2 Variable Definitions ...........................................................................................61  Table 3.3 Descriptive Statistics ..........................................................................................64  Table 3.4 OLS Regressions on Food Diversity Indices (N=4,341)  ..................................66  Table A.1 UK FSAÕs Nutrient Profiling Model A Points Allocation ................................75  Table A.2 UK FSAÕs Nutrient Profiling Model C Points Allocation ................................75  Table A.3 Double-Hurdle Model Estimates (N=3,894) ....................................................76  Table B.1 % Daily Value of Nutrients in 8oz of Assorted Fruit Juices Based on a 2,000 Calorie Diet .......................................................................................................77        !$!LIST OF FIGURES   Figure 1.1 FoodAPS Households Included in Study Sample by Participation Level ........13  Figure 2.1 Percent Daily Value of Key Nutrients in Fruit Beverages Based on a 2,000 Calorie Diet ......................................................................................................41    Figure 3.1 Group and Product Code-Based Food Categories ............................................57  Figure 3.2 Distribution of Group and Product Code Food Diversity Indices ....................65   Figure C.1 Correlation of Group and Product Code Food Diversity Indices ....................78                                 !$#!KEY TO ABBREVIATIONS   %DV  Percentage recommended daily value   ACQ  Acquisition characteristics  CES  Consumer Expenditure Survey   DEM  Demographics   DGA  Dietary Guidelines for Americans   ERS  Economic Research Service  EXP  Expenditures  FAFH  Food away from home  FAH  Food at home   FNS  Food and Nutrition Service   FOODAPS National Household Food Acquisition and Purchase Survey   FSA  Food Standards Agency  FSR  Full-service restaurant  HEI  Healthy Eating Index   HFD  Healthy Food Diversity   HPT  Household Production Theory   IMR  Inverse Mills ratio  IRI  Information Resource Inc.   NUTR  Nutrients  OLS  Ordinary Least Squares  PACK  Packaging  PROD  Product attributes !$##! QSR  Quick-service restaurant   RDV  Recommended daily values   SNAP  Supplemental Nutrition Assistance Program   UK  United Kingdom  US  United States   USDA  United States Department of Agriculture   WIC  Women, Infants and Children                                  !1 CHAPTER 1 DETERMINANTS OF HOUSEHOLD EXPENDITURES ON HEALTHY FOOD AWAY FROM HOME   1.1 Introduction Over the course of three generations, US dining habits have undergone a major transformation, with Americans cooking at home less and dining out more. From 1954 to 2013, the share of food dollars Americans allocated to food away from home (FAFH) increased from 25% to 50% (US 2013b). Defined as ready-to-eat meals, snacks and drinks purchased outside of the home, FAFH comprises approximately 35% of AmericansÕ caloric intake (Todd and Scharadin 2016; Lin and Guthrie 2012). Once synonymous with restaurants, alternative retailers such as supermarkets and convenience stores are expanding their FAFH offerings (Creel et al. 2008). In 2012, approximately 58%, 37%, 32% and 24% of households purchased FAFH from quick-service restaurants, grocery stores, convenience stores and sit-down restaurants, respectively (Data 2013). The increasing role of FAFH in AmericansÕ diets served as motivation for early studies on FAFH expenditures. Redman (1980), Kinsey (1983) and Yen (1993) focus on understanding the link between FAFH expenditures and demographic factors, placing particular emphasis on the effect of women entering the workforce. In the late 1980Õs and 1990Õs, the literature shifted to analyzing determinants of expenditures on different types of FAFH. McCracken and Brandt (1987) and Byrne et al. (1998) analyze FAFH expenditures by retailer type, distinguishing between quick-service and full-service restaurants, while Hiemstra and Kim (1995) and Jensen and Yen (1996) analyze FAFH expenditures by meal type, considering expenditures on breakfast, lunch, dinner and snacks. !2 Subsequent to these studies, FAFH has faced increasing scrutiny from the media, nutritionists and government. The nutrition literature find that FAFH consumption is associated with greater intake of calories, fat, cholesterol and sodium and lower intake of nutrients (Lin and Guthrie 2012; Jeffrey et al. 2006). Further, nutritionists find that FAFH consumption is linked to an increased risk for obesity (Kim et al. 2014). Mainstream media increased public awareness of FAFHÕs poor nutritional quality through books and documentaries such as Fast Food Nation: The Dark Side of the All-American Meal and Super Size Me (Bauer et al. 2012). Responding to this criticism, the 2010 Patient Protection and Affordable Care Act required all food vendors with more than 20 locations to display calorie counts on their menus (Bleich et al. 2015). With negative publicity adversely affecting sales, FAFH retailers have responded by improving the nutritional quality of their menu offerings (Jeffrey et al. 2006; Binkley 2006). Improvements include the addition of healthy menu lines, healthier childrenÕs meals, healthy indicator labels, fortified menu items and the discontinuation of Òsuper-sizedÓ portions (Bleich et al. 2015; Bauer et al. 2012; Harris et al. 2010). FAFH retailers have actively marketed these improvements, with nearly 50% of quick-service restaurantsÕ commercials focusing on healthy menu items (Kirkpatrick et al. 2014; Harris et al. 2010).  Criticism of the nutritional quality of FAFH has renewed interest in the determinants of FAFH expenditures in the literature. Stewart et al. (2005) explore the effect of preferences for a healthy diet and nutritional knowledge on FAFH demand, while Binkley (2006) considers how nutritional concerns and attitudes affect expenditures on FAFH. Liu et al. (2013a-b) analyze determinants of expenditures on FAFH by retailer, meal and household type in order to obtain a renewed understanding of the FAFH !3 market. Further, Richards and Mancino (2013) obtain FAFH price elasticities in order to inform FAFH tax policies aimed at curbing obesity.  While rising nutrition concerns have motivated recent studies to rexamine FAFH expenditures, no study has considered determinants of household expenditures on healthy FAFH. Thus, despite FAFH retailersÕ extensive efforts to improve the nutritional quality of their menu offerings, little is known about the factors affecting consumersÕ decision to purchase healthy items. Determining which demographic and economic factors affect expenditures on healthy FAFH can help inform future food and nutrition policy and help guide FAFH retailersÕ product development and marketing efforts. The objective of this study is to analyze householdsÕ expenditures in the growing healthy FAFH segment. Specifically, this study seeks to analyze determinants of: (1) householdsÕ decision to purchase healthy FAFH given that they are dining away from home and (2) the share of householdsÕ FAFH expenditures allocated to healthy FAFH.   1.2 Data This analysis of healthy FAFH expenditures is conducted using data from the National Household Food Acquisition and Purchase Survey (FoodAPS). Sponsored by the United States Department of AgricultureÕs Economic Research Service and Food and Nutrition Service, FoodAPS is a nationally representative data set of household food purchases and acquisitions. Administered by Mathematica Policy Research between April 2012 and January 2013, FoodAPS consists of one-week food acquisition diaries for 4,826 participating households, which document all food at home (FAH) and FAFH purchases  and acquisitions. Entry and exit surveys were conducted for each household and provide !4 demographic and socioeconomic data.  The FoodAPS data set is ideally suited for analyzing expenditures on healthy FAFH. Unlike the Consumer Expenditure Survey (CES) used in several past studies on FAFH, FoodAPS does not limit FAFH to items purchased from restaurants, work or school (Todd & Scharadin 2016; Liu et al. 2013a; Liu et al. 2013b; Jensen & Yen 1996). In addition to these retailers, FoodAPS includes FAFH purchases from supermarkets, grocery stores, convenience stores and gas stations. FoodAPS also improves upon the CES by oversampling low-income households and households receiving SNAP food assistance benefits, allowing for the analysis of food insecure household expenditures on healthy FAFH. Further, FoodAPS is ideal in that it provides detailed nutritional information for each FAFH purchase, including nutrient content and food group servings. This nutritional information is critical for the classification of healthy FAFH items in this study. Of the 4,826 households participating in FoodAPS, 165 do not report a food purchase or acquisition and are removed from the sample. Because we are interested in the share of FAFH expenditures households allocate to healthy FAFH, households who do not purchase FAFH are also excluded from the sample. Following Drewnowski and Fulgoni (2008), food items with limited nutritional content are removed, including condiments, sweeteners, creamers and alcohol. Further, free items not associated with coupon use are removed. This results in a sample of 3,894 households purchasing 57,779 FAFH items.  The healthfulness of each of the 57,779 FAFH items is evaluated using the UK Food Standards AgencyÕs Nutrient Profiling Model. Developed in 2005 to provide  !5  regulation on the advertising of unhealthy foods to children, the model identifies healthier food items using a point scoring system ranging from [-15, 40], with lower values associated with healthier items (Food 2011). Under the scoring system, ÔA PointsÕ are allocated based on the energy, saturated fat, total sugar and sodium content of each item. Similarly, ÔC PointsÕ are allocated based on each itemÕs fruit, vegetable, nut, fiber and protein content.1 An overall score is then calculated by subtracting the ÔC PointsÕ from the ÔA PointsÕ. Foods scoring less than 4 points and drinks scoring less than 1 point are classified as healthy options (Food 2011). Applying the Nutrient Profiling Model to the FoodAPS dataset results in the classification of 22,617 FAFH items as healthy. Examples of common healthy FAFH items by retailer type are presented in Table 1.1.  1.3 Theoretical Framework Following past studies on FAFH demand, Household Production Theory (HPT) is used to model FAFH expenditures (Liu et al. 2013a; Liu et al. 2013b; Jensen and Yen 1996; Hiemstra and Kim 1996). Developed by Becker (1965), HPT defines households as both !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!1 Tables A.1 and A.2 in Appendix A detail the allocation of A and C Points based on specific nutritional component levels. Table 1.1 Top 10 Healthy FAFH Items by Retailer Type (N=22,617)SupermarketsConvenienceQSRFSROthersBaked chipsBaked chipsBaked friesBaked friesBaked friesBaked chicken Baked friesBurritos/tacosLettuce/saladsMilk, lowfatOther vegetablesCoffeeBaked chicken Mixed poultry dishLettuce/saladsYeast breadsNuts/seedsPoultry sandwichMixed meat dishOther fruitVegetable dishBurritos/tacosBaked chipsDumplings/sushiMilk, lowfatMashed potatoesYeast breadsMixed poultry dishVegetable dishRolls/bunsMixed meat dishBananasCoffeeBeefMilk, reduced fatBananasFishOther sandwichRice/chow meinApplesPoultry dishCitrus juiceVegetable dishBaked chicken White potatoesLegumesMixed meat dishMashed potatoesMashed potatoesBaked chips!6 consumers and producers. In the context of FAFH, households have the option of producing their own home-cooked meals (i.e. FAH) or purchasing ready-to-eat meals (i.e. FAFH). Faced with a time constraint, a budget constraint and a production function, households maximize their utility by selecting optimal quantities of FAFH and time allocation. Following Binkley (2006), householdsÕ utility maximization yields: (1) !!!!!!!!!!!!!!!, where !! is a measure of household iÕs FAFH purchases, P  is vector of prices, I  is  household income, T is a householdÕs time cost, H is a vector of household health preferences and D is a vector of demographics and food acquisition characteristics. In this study, !! represents the share of householdsÕ FAFH expenditures allocated to healthy FAFH, which we refer to as healthy FAFH share in the remainder of the article. Given the cross-sectional nature of the FoodAPS data set, market prices are not available. Following Liu et al. (2013a) and Liu et al. (2013b) regional indicator variables are included to account for market price variation.     1.4 Methodology 1.4.1 The Zero Expenditure Problem A common issue when analyzing household expenditure data is the presence of a large number of zero expenditures. This problem is known as the censored dependent variable problem, or more commonly as the zero expenditure problem (Wooldridge 2010). There are three potential reasons householdsÕ healthy FAFH expenditure shares may equal zero: (1) corner solutions, (2) abstention and (3) infrequency of purchase (Pudney 1989). A corner solution implies that when eating out, households choose not to purchase healthy  !7  FAFH given the economic environment, while abstention implies that households purchasing FAFH choose not to purchase healthy items independent of economic factors.  Under infrequency of purchase, a householdsÕ purchase cycle for healthy FAFH is longer than the survey period. However, since we are analyzing the share of FAFH expenditures allocated to healthy FAFH, infrequency of purchase is not a logical explanation for zero expenditure shares in this study.  Two types of models are commonly used to correct for zero expenditure problems: (1) censored tobit models and (2) double-hurdle models. Censored tobit models imply that zero expenditure shares are the result of a corner solution (Wooldridge 2010). However, the censored tobit model is restrictive in that the same mechanism governs both the participation and expenditure decisions. The double-hurdle model overcomes this restriction by modeling a two-step decision process. In the first step, or the participation decision, a household dining away from home decides whether to purchase healthy FAFH. In the second step, or the expenditure decision, the household decides the share of its FAFH expenditures to allocate to healthy FAFH. This approach allows for zero expenditure shares to be the result of both economic factors and abstention.  In this study, a Vuong test for non-nested models is used to determine whether the censored tobit or CraggÕs double-hurdle model best fits the FoodAPS data. Shown in Table 1.2, results from the Vuong test strongly reject the null hypothesis that the two Table 1.2 Vuong Model Selection Test: Censored Tobit vs Cragg's Double HurdleVuong                Test StatisticTP>|t|Model                       Selection3,877.9618.480Cragg's Double Hurdle!8 models fit the data equally well in favor of the double-hurdle model. Thus, CraggÕs double-hurdle model is applied in this analysis.  1.4.2 CraggÕs Double-Hurdle Model In CraggÕs double-hurdle model, household participation and expenditure decisions are modeled as follows:  (2) !!!!!!!!"!!!!!!!"#!!!!!!!!!"!!"#$%!              where (3)!!!!!!!!!!!!!!!!!!!!!!!!!!! (4)!!!!!!!!!!!!!!!!!!!!!!!!!!!!!  where !!  is the observed healthy FAFH expenditure share for household i,  !!! is the latent healthy FAFH expenditure share, !! is a binary indicator of healthy FAFH participation among households purchasing FAFH, !! is a set of explanatory variables for the participation decision, !! is a set of explanatory variables for the expenditure decision and !!!and !! are error terms (Wooldridge 2010). While the set of explanatory variables included in the participation decision (!!) and expenditure decision (!!) are allowed to vary, it is common practice to include the same variables in both equations (Wooldridge 2010). Estimation of CraggÕs double-hurdle model is broken into two steps. In the first step, Equation 3 is estimated via a probit regression to model the participation decision. In the second step, Equation 4 is estimated using a truncated regression with a lower limit !9 of zero to model the expenditure decision. The first stageÕs inverse mills ratio is included in the second stage to correct for potential sample selection bias. Marginal effects and elasticities of the participation probability, conditional level and unconditional level are calculated for continuous and binary explanatory variables respectively. The participation probability !!!!!!!! represents the probability that households dining away from home purchase healthy FAFH, the unconditional level !!! is the expected value of healthy FAFH expenditure shares for all households purchasing FAFH and the conditional level !!!!!!! is the expected value of healthy FAFH expenditure shares for only households purchasing healthy FAFH.   1.4.3 Description of Explanatory Variables Table 1.3 details the explanatory variables comprising !! and !! in this study. Based on Equation 1 from Household Production Theory, the explanatory variables are grouped into five categories: (1) income, (2) acquisition characteristics, (3) demographics, (4) health preferences and (5) times costs.  The income category consists of three indicator variables that classify households based on income, Supplemental Nutrition Assistance Program (SNAP) participation and food access. Income is expected to positively affect healthy FAFH participation and expenditure shares, given that low-income households have lower availability of healthy FAFH and healthy eating index (HEI) scores (Hiza et al. 2013; Chau et al. 2013). A binary indicator variable is also included to indicate whether households receive SNAP benefits. SNAP benefits can be used to purchase select, non-heated FAFH items at  retailers other than restaurants, such as supermarkets and convenience stores. Unlike  !10  earned income, SNAP benefits are expected to inversely affect householdsÕ healthy  FAFH participation and expenditure shares. Nguyen et al. (2014) find that SNAP VariableDefinitionUnitBase  Healthy FAFH ShareShare of FAFH expenditures allocated to healthy FAFH%---   Low-Income/SNAPHousehold has income below the poverty threshold and receives SNAPDVaHigh   Income   Low-Income/Non-SNAPHousehold income below the poverty threshold, but does not receive SNAPDVHigh   Income   Low-AccessHousehold located in low-access tractDVHigh    Access   Healthy FAH ShareShare of FAH purchases allocated to healthy FAH%---   Supermarket ShareShare of FAFH bought at supermarkets%---   Convenience ShareShare of FAFH bought at conv. retailers%---   QSR ShareShare of FAFH bought at QSR%---   FSR ShareShare of FAFH bought at FSR%---   Other ShareShare of FAFH bought at other retailers%---   FAFH VisitsNumber of visits to FAFH retailers#---   AgeAge of primary respondentYears---   CollegePrimary respondent is college educatedDVNo College   African-AmericanPrimary respondent is African-AmericanDVWhite   AsianPrimary respondent is AsianDVWhite   HispanicPrimary respondent is HispanicDVWhite   Rural Household is located in a rural tractDVUrban   MidwestHousehold is located in the MidwestDVNortheast   SouthHousehold is located in the SouthDVNortheast   WestHousehold is located in the WestDVNortheast   Nutrition SearchPrimary respondent searched for nutrition information online in the last 2 DVNo Search   DietingA member of the household is dietingDVNo Diet   Helathy TimePrimary respondent is too busy to prepare healthy mealsDVHas TimeaDV: Discrete VariableTable 1.3 Variable DefinitionsDependent VariablesIncome Food Acquisition DemographicsHealth PreferencesTime Cost!11 households have lower HEIs than both low-income and high-income, non-SNAP households. Similarly, households with low-access to supermarkets are expected to have lower healthy FAFH expenditure shares and participation, with past research finding low-access is associated with lower HEI scores (Larson et al. 2008)2.  Acquisition characteristics are also expected to affect healthy FAFH participation and expenditure shares. The share of householdsÕ FAH expenditures allocated to healthy food is included to test the hypothesis that households purchasing a greater share of healthy FAH, will purchase a greater share of healthy FAFH. In order to account for differences in the availability, FAFH expenditure shares at supermarkets, convenience stores, sit-down restaurants and other retailers are included. The base variable is the share of FAFH expenditures at quick-service restaurants. Creel et al. (2005) and Binkley (2008) find that quick-service restaurants tend to have healthier FAFH options than supermarkets and convenience retailers, but less healthy options than sit-down restaurants. Thus, relative to quick-service restaurants, increasing FAFH expenditure shares at supermarkets and convenience retailers (sit-down restaurants and other retailers) is expected to decrease (increase) householdsÕ healthy FAFH participation and expenditure shares. Given Frank et al. 2009Õs finding that increased FAFH consumption lowers diet quality, FAFH visits are also included and expected to inversely affect healthy FAFH expenditure shares.  Following Binkley (2006), two variables representing householdsÕ health preferences are included: (1) nutrition search and (2) dieting. Nutrition search indicates that a household searched for nutrition information online prior to the survey period. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!2 A census tract is classified as low-access if at least 33% of the population is located more than 1 mile (urban) or 20 miles (rural) from the nearest supermarket (US 2016b). !12 While past studies find that nutrition information is inversely related to FAFH expenditures, we hypothesize it is positively related to healthy FAFH participation and expenditure shares (Stewart et al. 2005; Binkley 2006). Similarly, we expect that households indicating that one or more household member is dieting will be more likely to purchase and spend a greater share of their expenditures on healthy FAFH items. The literature further suggests that household characteristics impact healthy FAFH expenditure shares. Hiza et al. (2013) and Ervin (2011) find that dietary quality increases with age and education, suggesting that older households with college-educated heads will have greater healthy FAFH participation and expenditure shares. Findings by Hiza et al. (2013) and Ervin (2011) also suggest healthy FAFH participation and expenditures will vary by race, with Hispanics and Asians having higher HEI scores than Whites. In contrast, rural households are expected to have lower healthy FAFH participation and expenditure shares, with Sharkey et al. (2011) and Morton and Blanchard (2007) finding that rural households have lower HEI scores and lower access to healthy food in the US.  The final category of explanatory variables, time costs, accounts for householdsÕ opportunity cost of time. Serving as a proxy, healthy time is a dummy variable that indicates whether time constraints prevent households from preparing healthy meals at home. Binkley (2006) and McCracken and Brandt (1987) find that the importance households place on convenience is positively associated with FAFH expenditures. Similarly, householdsÕ time constraints in healthy meal preparation are expected to have greater healthy FAFH participation and expenditure shares.  !13  Figure 1.1 FoodAPS Households Included in Study Sample by Participation Level 1.5 Results 1.5.1 Descriptive Statistics Descriptive statistics are presented in Table 1.4 for all households in the sample. Mean  comparison tests are conducted to compare the acquisition characteristics, income, demographics, health preferences and time costs of households purchasing healthy FAFH  (participants) and households purchasing only unhealthy FAFH (non-participants). For clarification, Figure 1.1 provides a visual representation of the households in this studyÕs sample by participation level. Of the 3,894 households purchasing FAFH, 85% purchase healthy FAFH. On average, those purchasing healthy FAFH allocate 42% of their FAFH expenditures to healthy items. This finding mirrors FAH, where participants (non-participants) allocate an average of 45% (44%) of their FAH expenditures to healthy items.  Households  purchasing healthy FAFH made significantly more visits per week to FAFH retailers on  !14  average, 5.81, than non-participating households, 2.04. The distribution of FAFH expenditures across retailer types also differs significantly between participating and non-participating households. Participating households purchase a greater share of FAFH from full-service restaurants at 35% versus 20% for non-participating households. Table 1.4 Descriptive Statistics for Participating and Non-Participating HouseholdsVariableMean Participating          (N=3,322)Mean                             Non-Participating                (N=572)  Healthy FAFH Share0.42---   Low-Income/SNAP***0.270.42   Low-Income/Non-SNAP0.240.26   Low-Access0.310.3   Healthy FAH Share0.450.44   Supermarket Share***0.040.07   Convenience Share***0.060.13   QSR Share*0.430.46   FSR Share***0.350.20   Other Share**0.120.14   FAFH Visits***5.812.04   Age44.4845.49   College***0.230.17   African-American*0.150.17   Asian0.050.03   Hispanic*0.210.18   Rural **0.260.30   Midwest0.250.27   South0.370.39   West0.220.20   Nutrition Search***0.300.22   Dieting*0.320.28   Helathy Time***0.230.17Means differ at the * 0.10 level, ** 0.05 level and *** 0.01 levelDependent VariablesFood Acquisition DemographicsHealth PreferencesTime CostIncome !15 Further, participating households purchase a lower share of FAFH from supermarkets, convenience retailers, other retailers and quick-service restaurants at 4%, 6%, 12% and 43% versus 7%, 13%, 14% and 46% for non-participating households respectively.  Mean comparison tests also highlight the stark differences in the economic status of participating and non-participating households. Differing at the 1% level, 27% of participating households are low-income and receive SNAP benefits versus 42% of non-participating households. However, the share of low-income households, not receiving SNAP does not significantly differ by participation level. Unlike SNAP, participating and non-participating householdsÕ access to supermarkets does not significantly differ.  Demographics, health preferences and time costs further vary by participation level. Relative to non-participants, a greater share of participating households are college-educated, Hispanic, White and live in an urban census tract. A greater share of participating households also reported searching for nutrition information online and indicated that a household member was dieting. This finding suggests that health preferences may positively affect healthy FAFH participation. In addition to health preferences, times constraints also appear to positively affect healthy FAFH participation. The share of households reporting that time constraints prevent them from cooking healthy meals is greater among participants at 32%, than among non-participants at 28%.   1.5.2 Double-Hurdle Model Estimates CraggÕs double-hurdle model estimates are presented in Table A.3 in Appendix A. The inverse mills ratio is significant in the second stage, indicating that sample selection bias was present, but has been corrected. Further, variance inflation factors indicate no !16 presence of multicollinearity. The resulting estimates explain a significant amount of the variation in householdsÕ healthy FAFH expenditure shares, with a pseudo-R2 of 0.45. In order to facilitate interpretation of the double-hurdle model estimates, marginal effects and elasticities of the participation probability, unconditional expenditure level and conditional expenditure level are also calculated and presented in Table 1.5.  The double-hurdle model estimates indicate that SNAP participation affects householdsÕ healthy FAFH expenditure share. Marginal effects imply that, relative to high-income households, households receiving SNAP benefits are 3% less likely to purchase healthy FAFH and, given participation, allocate 5% less of their FAFH expenditures to healthy FAFH items. In contrast, healthy FAFH participation and expenditure shares do not differ between high-income households and low-income households that, despite eligibility, do not receive SNAP benefits. These findings coincide with the nutrition literature, in which Nguyen et al. (2014) find that SNAP households have lower HEIs than both low-income and high-income, non-SNAP  households. Results further indicate that food access does not affect healthy FAFH participation or expenditure shares.  Acquisition characteristics also significantly affect household healthy FAFH participation and expenditure shares. Conditional elasticities indicate that a 1% increase  in participating householdÕs healthy FAH share leads to an 8% increase in their healthy FAFH share. This finding supports the hypothesis that households purchasing a greater  share of healthy FAH will purchase a greater share of healthy FAFH. The frequency of  visits to FAFH retailers is also positively related to healthy FAFH participation. Elasticity  estimates indicate that a 1% increase in FAFH visits leads to a 17% increase in the  !17  probability of healthy FAFH participation. This finding suggests that frequent consumption of FAFH may result in variety-seeking behavior in FAFH purchases.  Relative to quick-service restaurants, the share of FAFH expenditures at supermarkets, full-service restaurants and other retailers (convenience retailers) positively (inversely) affects healthy FAFH participation. While the magnitude of the VariableParticipation ProbabilityabUnconditional        LevelConditional            Level   Low-Income/SNAP-0.03***-0.05***-0.05***   Low-Income/Non-SNAP-3.03E-03-0.01-0.01   Low-Access-3.08E-03-0.02-0.02   FAH Healthy Share-2.58E-030.07***0.08***   Share of FAFH Purchases at Supermarkets-1.64E-03**0.01***0.01***   Share of FAFH Purchases at Convenience-4.70E-03***-0.02***-0.01**   Share of FAFH Purchases at FSR0.01***0.16***0.15***   Share of FAFH Purchases at Other Retailers-4.36E-03***0.02***0.02***   FAFH Visits0.17***0.19***0.02Household Characteristics    Age0.010.20***0.19***   College-0.010.010.01   African-American1.63E-030.010.01   Asian0.020.12***0.12***   Hispanic0.02**0.010.01   Rural -0.01-0.04***-0.03***   Midwest-0.03***-0.05***-0.04***   South-0.02**-0.03**-0.02   West-0.02-0.02-0.01Health Preferences   Nutrition Search0.01*0.04***0.03***   Dieting0.010.03**0.02***Time Cost   Healthy Time3.20E-030.03***0.03**   Inverse Mills Ratio---------Constant---------aProbability/unconditional/conditional level elasticities calculated for continuous variables bProbability/unconditional/conditional level marginal effects calculated for discrete variablesSignificant at the * 0.10 level, ** 0.05 level and *** 0.01 levelTable 1.5 Double-Hurdle Model Elasticities and Marginal Effects  (N=3,894)IncomeFood Acquisition Characteristics!18 effect is small for supermarkets, convenience and other retailers, a 1% increase in the share of FAFH expenditures at full-service restaurants is associated with a 1% increase in the probability of purchasing healthy FAFH. Conditional elasticities further indicate that, relative to quick-service restaurants, 1% increases in the share of FAFH expenditures at supermarkets, full-service restaurants and other retailers (convenience retailers) lead to 1%, 15% and 2% (1%) increases (decreases) in the share of FAFH expenditures allocated to healthy items respectively. Paralleling the findings of Binkley (2008) and Creel et al. (2005), these findings suggest that supermarkets, full-service restaurants and other retailers may have greater availability of healthy FAFH items than quick-service restaurants and convenience retailers.  Largely mirroring the nutrition literature, several demographic factors affect householdsÕ healthy FAFH participation and expenditure shares. Age has a large, significant effect on healthy FAFH expenditure shares, with a 1% increase in age leading to a 19% increase in the allocation of FAFH expenditures to healthy FAFH. Healthy FAFH participation and expenditure shares are further found to vary among Hispanic, Asian and White households. Marginal effects indicate that Hispanic households are 2% more likely to purchase healthy FAFH, while Asian households allocate 12% more of their FAFH expenditures to healthy items. However, healthy FAFH participation and expenditure shares do not significantly differ between African-American and White households. Unlike age and race, results indicate that college education does not significantly affect healthy FAFH participation or expenditure shares.  Relative to urban households, participating households located in rural census tracts allocate 3% less of their FAFH expenditures to healthy FAFH items. Compared to !19 the Northeast region of the US, households located in the Midwest and Southern regions are 3% and 2% less likely to purchase healthy FAFH respectively. Further, participating Midwestern households allocated 4% less of their FAFH expenditures to healthy FAFH. These findings likely capture regional differences in FAFH prices, tastes and preferences.  Estimates for the final two categories of explanatory variables, health preferences and time costs, are as expected a priori. Households who search for nutrition information online are 1% more likely to purchase, and given participation, allocate 3% more of their expenditures to healthy FAFH. Similarly, participating households with a member dieting allocate 2% more of their FAFH expenditures to healthy FAFH than non-dieting households. As expected, these findings indicate that preferences for a healthy diet and nutrition information positively affect healthy FAFH participation and expenditures. Further, participating households that indicate time constraints prevent them from preparing healthy meals at home allocate 3% more of their expenditure to healthy FAFH.   1.6 Discussion Given FAFH retailersÕ recent efforts to improve the nutritional quality of their menu offerings, this study analyzes determinants of householdsÕ expenditures on healthy food away from home (FAFH) using the National Household Food Acquisition and Purchase Survey. For households purchasing FAFH, CraggÕs double-hurdle model is used to analyze determinants of their decision to purchase healthy FAFH and the share of total FAFH expenditures to allocate to healthy FAFH. Results indicate that healthy FAFH participation and expenditure shares vary with SNAP participation, the healthiness of householdsÕ food at home purchases, nutrition information, FAFH retailer type and time !20 constraints, as well as other basic demographic factors. Results from this study provide further insight into the food purchases of households receiving SNAP food assistance benefits. Relative to high-income households, SNAP households dining away from home are 3% less likely to purchase and allocate 5% less of their FAFH dollars to healthy FAFH. In contrast, the healthy FAFH participation and expenditure shares of low-income households not receiving SNAP do not significantly differ from those of high-income households. As with FAH, these results suggest that receiving SNAP benefits does not lead to improved dietary quality of FAFH purchases. Instead, it appears SNAPÕs income effect leads low-income households to purchase a greater share of unhealthy FAFH items. This suggests that SNAP householdsÕ decisions to purchase healthy FAFH are driven by factors other than income, such as tastes, preferences, cultural values and habits.  This study also finds that the healthiness of participating householdsÕ food purchases is similar across FAFH and FAH, with 42% and 45% of FAFH and FAH shares allocated to healthy items respectively. A 1% increase in the share of FAH expenditures allocated to healthy items is also associated with an 8% increase in healthy FAFH expenditure shares. Given the conscious decision to purchase healthy FAFH, these results indicate that FAFH purchases are not inherently less healthy than FAH purchases. Thus, encouraging healthy decision-making may prove more effective in improving the nutritional quality of householdsÕ FAFH purchases than discouraging the consumption of FAFH altogether. Further, results also indicate that obtaining nutrition information alters both householdsÕ healthy FAFH participation and expenditure shares. Households who make !21 nutrition searches online are 1% more likely to purchase and allocate 3% more of their FAFH expenditures to healthy items. However, only 28% of households reported searching for nutritional information. This finding suggests that increasing householdsÕ use of nutrition information can help improve the nutritional quality of their FAFH purchases. Potential methods of increasing use of nutrition information include nutrition education programs with a focus on collecting and interpreting nutrition information, as well as increasing the transparency of nutritional information at FAFH retailers. In addition to implications for food and nutrition policy, findings from this study also have direct implications for FAFH retailers carrying healthy FAFH products. Estimates obtained from the double-hurdle model identify key consumer segments that allocate a greater share of their FAFH expenditures to healthy FAFH items. Results suggest that high-income, non-SNAP, urban, health-conscious and older householdsÕ are key consumer segments in the healthy FAFH industry. Identification of key consumer segments is essential for FAFH retailers to effectively market healthy FAFH products.  This study presents a first look at the determinants of householdsÕ expenditures on healthy FAFH. Based on the findings from this study, further research on healthy FAFH expenditures is needed to obtain a clearer picture of the role of healthy FAFH in householdsÕ diets. In future studies, analysis of healthy FAFH expenditures by household type would provide a better understanding of the effects of income and SNAP benefits on healthy FAFH expenditures. Analysis of healthy FAFH expenditures segmented by retailer type and meal type would also provide insight on the types of healthy FAFH households purchase and where householdsÕ choose to purchase healthy FAFH. Further, future research employing a systems demand approach would give valuable insight on !22 householdsÕ substitution among healthy FAFH, unhealthy FAFH, healthy FAH and unhealthy FAH.                               !23 CHAPTER 2 A SEGMENTED HEDONIC ANALYSIS OF THE NUTRITIONAL COMPOSITION OF FRUIT BEVERAGES   2.1 Introduction With the average American consuming nearly 40 liters of fruit beverages every year, the United States is one of the worldÕs largest fruit beverage consumers (Euromonitor 2015; Singh et al. 2015). Fruit beverages can be grouped into two categories: fruit juice and fruit drinks. Fruit juice is defined as pure, 100% juice with no added ingredients, while fruit drinks are fruit beverages containing ingredients other than fruit juice, such as sugar, and often have minimal nutritional value (Mintel Report 2015). On average, fruit drinks contain only 10% fruit juice (Harris et al. 2011).  Currently, a significant shift in fruit beverage consumption is occurring in the US, due largely to concerns over its sugar content. Wang et al. (2008) explain that fruit beveragesÕ sugar content is similar to that of soft drinks and other sugar sweetened beverages. Studies have further found evidence that fruit beverage consumption is associated with an increased risk for obesity, heart disease and diabetes (Dennison et al. 1997; Wojcicki and Heyman 2012; Imamura et al. 2015; Eshak et al. 2013). Reflecting these concerns, the Dietary Guidelines for Americans have been revised and now recommend abstaining from fruit drink consumption and limiting fruit juice consumption (US 2015). As a result, the United States Department of Agriculture has cut back on fruit beverage provisions in food assistance programs such as Women, Infants, and Children (WIC), and has begun regulating fruit beverage sales in schools. ConsumersÕ reactions to these concerns and changing federal guidelines/programs are reflected in the sales declines for fruit juice and non-diet fruit drinks from 2010 to 2015 (Mintel Report 2015; !24 Okrent and MacEwan 2014). Despite its high sugar content, many nutritionists view fruit beverages as an important source of vitamins and minerals and as a cost effective way for consumers to meet their daily fruit intake recommendations (OÕNeil et al. 2012; Clemens et al. 2015). Among its consumers, fruit beverages are increasingly purchased for their functional attributes i.e. the nutrients they contain (Mintel 2014). According to Mintel, over 40% of Americans depend on fruit juice as a source of added nutrients in their diets (2015). Manufacturers have responded to the demand for functional fruit beverages by emphasizing the naturally occurring nutrients in their products and introducing fruit beverages fortified with vitamins and minerals (Siro et al 2008; Bishai and Nalubola 2002). Given the increased importance consumers and manufacturers have placed on the functional nutrients found in fruit beverages, as well as the changing federal guidelines on fruit beverage consumption, this study seeks to determine whether specific nutrients garner price premiums in fruit beverages sold in the US. Specifically, this study seeks to answer the following questions: (1) which nutrients found in fruit beverages garner price premiums, (2) do the specific nutrients that garner price premiums vary by fruit beverage type and (3) what other attributes of fruit beverages garner a price premium. This study adds to the literature in that it is the first to consider whether specific nutrients garner price premiums in fruit beverages sold in the US. Further, this study is the first to estimate separate hedonic models for fruit juice and fruit drinks, thus allowing price premiums for nutrients to differ between the two fruit beverage types.  !25 2.2 Background 2.2.1 US Fruit Beverage Industry  The US is one of the largest consumers of fruit beverages, with 8.4 and 4.2 billion liters of fruit juice and fruit drinks purchased in 2014 (Euromonitor 2015; Singh et al. 2015). According to Mintel, approximately 75% (49%) of US consumers reported drinking fruit juice (drinks) in 2015. Orange (mixed fruit) is the most popular flavor of fruit juice (fruit drink), with a 60% (29%) market share (Euromonitor 2015). Other leading fruit juice (drink) varieties in order of market share include apple, mixed fruit, tomato, grape, cranberry, grapefruit, prune and lemon (citrus, berry, lemonade, grape and apple) (Euromonitor 2015). The fruit beverage industry in the US is relatively concentrated, with ten major companies accounting for 70% of fruit beverage sales. These companies and their respective market shares are as follows: Coca-Cola Co. (18.2%), PepsiCo Inc (13.4%), Campbell Soup (7.7%), Kraft Foods Group Inc. (6.6%) Ocean Spray Cranberries Inc. (6.5%), Dr. Pepper Snapple Group (5.3%), National Grape Cooperative Association Inc. (3.7%), Citrus World (3.3%), Beverage Holdings (2.4%) and Nestle (2.2%) (Euromonitor 2015).   2.2.2 Nutritional Composition of Fruit Beverages  In the US, sugar-sweetened beverages are the single greatest source of added sugars in the American diet, with fruit drinks alone accounting for 10% of the added sugar consumed every year (Krebs-Smith 2001; US 2015). On average, an eight ounce fruit drink serving contains thirty-two grams of sugar or approximately 100% of oneÕs !26 recommended daily sugar intake (Harris et al. 2011). A 2014 report by YaleÕs Rudd Center for Food Policy and Obesity further explains that the average fruit drink sold in the US contains only 10% fruit juice, with the remaining 90% of the drink comprised of water and sugar (Harris et al. 2011). Correspondingly, fruit drinks are described as providing empty calories, in that they are high in energy from added sugars, but low in nutrients such as vitamins, minerals and fiber (Reedy & Krebs-Smith 2010).  Unlike fruit drinks, fruit juice has historically been viewed as an important source of nutrients in the American diet. A detailed summary of the nutritional composition of seven common varieties of fruit juice is provided in Table B.1 in Appendix B. In general, fruit juice is a significant source of Vitamin C, Potassium, Magnesium, Iron and Phosphorus (OÕNeil and Nicklas 2008). However, despite being a natural source of vitamins and minerals, all fruit juice varieties have high sugar contents, ranging from 49% of the recommended daily sugar intake for an 8oz serving of grapefruit juice to 119% for grape juice (OÕNeil and Nicklas 2008; US 2013c).   2.2.3 Federal Programs, Policies and Guidelines Concerning Fruit Beverages Over the past decade, federal programs, policies and guidelines have been altered or enacted in response to concerns over the high sugar content of fruit beverages in the US. Issued every five years, the Dietary Guidelines for Americans (DGA) provide consumers with guidance on maintaining a healthy diet and serve to inform food, health and nutrition policy (US 2015). The DGA recommendations on fruit beverage consumption have evolved considerably over the past decade. In 2005, the DGA recommended choosing fruit beverages with little added sugar (US 2005). By 2010, the DGA !27 specifically stated to abstain from consuming fruit drinks and suggested limiting childrenÕs intake of fruit juice, especially if children are overweight or obese (US 2010). In the 2015 DGA, specific limits were placed on added sugar consumption, with no more than 10% of oneÕs calories to be derived from added sugar (US 2015). In 2007, the USDAÕs nutrition program for Women, Infants and Children (WIC) was revised in response to the 2005 DGAÕs recommendation to choose beverages with little added sugar (Cole et al. 2011). Established in 1966, the goal of WIC is to provide supplemental foods containing nutrients known to be lacking in the diets of at-risk women and children (Oliveira et al. 2002). Since its inception, fruit juice has been among the items provided by WIC due to its vitamin content. To be deemed WIC eligible, a product must contain only 100% unsweetened, pasteurized juice and contain a minimum of 20 mg of Vitamin C per 100ml of juice (US 2016c). In compliance with the 2005 DGA, revisions made to WIC in 2007 include the removal of fruit juice from all infant packages and a nearly 50% reduction in the maximum fruit juice prescription for women and children (Cole et al. 2011). The USDA has also taken steps to regulate beverages sold in US schools. In July of 2014, the USDA implemented the Smart Snacks in School Standards which defined nutritional standards that all foods and beverage items sold in schools must satisfy (US 2013a). The standards effectively banned the sales of SSBs in schools, including fruit drinks. Among fruit beverages, only 100% fruit juice or 100% fruit juice diluted with water and with no added sugar can be sold in schools. The standards also limit the portion size of fruit juice that can be sold to 8oz and 12oz in elementary and middle/high schools respectively (US 2013a).  !28 2.2.4 Changing Consumer Demand for Fruit Beverages Consumers have reacted to the concerns over the sugar content in fruit beverages, as well as the changing federal guidelines and programs, by altering their fruit beverage consumption (Okrent and MacEwan 2014). Fruit juice expenditures in the US declined by 5% from 2010 to 2015, with approximately 34% of consumers who stopped drinking fruit juice doing so because of its high sugar content (Mintel 2015; Mintel 2014). During the same time period, fruit drink expenditures increased by 6%, driven primarily by the development of products containing fewer calories and less sugar (Mintel 2015; Taylor 2014; Okrent and MacEwan 2014). Among consumers, fruit beverages are increasingly viewed as functional foods (Mintel 2014). The Functional Food Center defines functional foods as Ònatural or processed foods that contain known or unknown biologically-active compounds; which, in defined, effective non-toxic amounts, provide a clinically proven and documented health benefit for the prevention, management, or treatment of chronic diseaseÓ (Martirosyan and Singh 2015). According to Mintel, 40% (24%) of US consumers who purchase fruit juice (fruit drinks) look for vitamin or mineral enhanced formulas (Mintel 2014). Leading functional ingredients consumers seek in fruit beverages include Vitamin C, Vitamin D and Calcium (Euromonitor 2016). In addition to added nutrients, approximately 43% of fruit juice and fruit drink consumers are interested in no sugar added or low sugar varieties. (Mintel 2014). In response to consumer demand for functional fruit beverages, manufacturers are emphasizing the naturally occurring nutrients in its products and introducing new fruit beverages fortified with vitamins and minerals (Siro et al 2008; Bishai and Nalubola  !29  2002). Key nutrients manufacturers are fortifying their fruit juice (drink) products with include Calcium, Vitamin D and Vitamin C (Vitamin C and Vitamin E) (Euromonitor 2016). In addition to functional attributes, a main area of focus for fruit beverage manufacturers is sugar reduction in its products (Mintel 2015). Manufacturers are conveying the nutritional benefits of their fruit beverages to consumers through the use of front-of-package labels. Detailed in Table 2.1, common front-of-package nutrition labels on fruit beverages include: good source of vitamins/antioxidants, % daily value of vitamins/minerals, natural source of antioxidants and no added/reduced/less sugar.   2.2.5 Uniqueness of this Study  Given the increased importance consumers and manufacturers place on the functional Table 2.1 Top Fruit Beverage Front-of-Package Nutrition Labels100% Fruit JuiceFruit DrinksVitamin C! % Daily Value Vitamin C ! An Excellent Source of Vitamin C ! With Vitamin C ! With Vitamin C ! % Vitamin C Per Serving ! Excellent/Good Source of Vitamin C Vitamin D! An Excellent Source of Vitamin D  ! Plus Calcium & Vitamin D! Plus Vitamin DVitamin E! % Daily Value of Vitamin E! Great Source of Vitamin EAntioxidants! Antioxidant Advantage ! Packed with Antioxidants A & C ! Essential Antioxidants ! Natural Source of Antioxidants ! Antioxidants Vitamin C & E ! 100% Daily Value of the Antioxidant Vitamin CMultiple Vitamins!With Vitamins A,B,C,D,E ! Packed with Vitamins ! Excellent Source of Vitamins ! Good Source of Vitamins A, C, E Calcium! % Daily Value of Calcium ! An Excellent/Good Source of Calcium ! Plus Calcium & Vitamin D! NoneSugar! 1/2 the Sugar ! No Sugar Added ! Less Sugar ! No High Fructose Corn Syrup! % less sugar  ! reduced sugar ! No High Fructose Corn SyrupFiber!High Fiber ! Good Source of Fiber ! With Fiber! None!30 nutrients found in fruit beverages, as well as changing federal guidelines on fruit beverage consumption, this study seeks to determine whether key nutrients garner price premiums in fruit beverages. Several past studies have considered price premiums for nutrients in foods other than fruit beverages. Looking at breakfast cereal, Morgan et al. (1979) and Stanley et al. (1991) collectively find that protein, minerals, vitamins and sugar garner a premium price, while fiber and calories are associated with a price discount. Similarly, Angulo et al. (2006) and Harris (1997) conclude that meat with greater fat, protein and fiber content commands a premium price. Gulseven and Wohlgenant (2014) further find a price premium for lactose and cholesterol free milk.  Two past studies have analyzed whether nutrients garner price premiums in fruit beverages. Weemaes and Riethmuller (2001) considered the price premium associated with quality attributes, including nutrients, in Australian fruit beverages. Findings include that sugar is associated with a negative price premium and fruit beverages labeled with the Australian Heart Foundation seal garner a price premium. In 2014, Szathvary and Trestini analyzed the effects of nutrition and health claims on the prices of fruit beverages in Northeast Italy. Results suggest that fruit beverages containing a nutrition and/or health claim are associated with a price premium.  This study adds to the literature in that it is the first to consider whether specific nutrients garner price premiums in fruit beverages sold in the US. Building off of Weemaes and Riethmuller (2001) and Szathvary and TrestiniÕs (2014) analysis of select nutrition claims, this analysis seeks to determine the price premiums associated with all key nutrients found in fruit beverages, including Vitamin C, Vitamin D, Antioxidants, Calcium, and sugar. This study is also the first to perform a segmented hedonic analysis !31 of fruit beverages, with separate models estimated for fruit juice and fruit drinks.  2.3 Hedonic Pricing Model 2.3.1 Hedonic Price Theory  In their formative works, Lancaster (1966) and Rosen (1974) questioned the traditional notion that consumers obtain utility from goods themselves. Instead, they explain that goods are made up of a set of attributes and it is these attributes that provide utility to the consumer. This concept serves as the basis for hedonic price theory. Under this theory, the observed prices and quantity of attributes for a specific good define a set of hedonic prices (Rosen 1974). There are three key assumptions made by hedonic theory: (1) consumers are aware of all available versions of a product, (2) there is significant variation within a product segment and (3) it is costless to switch between products (Costanigro et al. 2011).   2.3.2 Hedonic Price Model Following Rosen (1974), the hedonic price function for a good is defined as follows: (1)  !!!!!!!!!!!!  !!!!!!!!!!!!  where z is the product and !! is a row vector of the attributes for the ith product. Given this price function, consumers choose a bundle of attributes to maximize the following utility function (3) subject to their budget constraint (4): (3)  !!!!!!!!!!!! (4)  !!!!!!!!  !32 where y is income and x represents all other goods and has a unit price. Maximization of the utility function subject to the budget constraint results in the following first order condition: !!!!!!!!!!!.  This first order condition yields the implicit price for a specific attribute, !!!, and implies that consumers are indifferent between paying the implicit price for an additional unit of an attribute and using the money to purchase all other goods x (Costanigro et al. 2011). Analogously, producers choose a bundle of attributes and the number of goods to produce containing a particular attribute, !!!!, to maximize the following profit function: (6)  !!!"!!!!!!!!!!, where !!!!!!!! is the producerÕs cost function and ! is a parameter representing the producerÕs factor prices and production technologies. Maximization of this profit function results in the following first order condition: !!!!!!!!!!.  This first order condition implies that the marginal cost of adding an additional unit of an attribute to a product equals the implicit price of the attribute (Costanigro et al. 2011). Thus, at equilibrium, the market clearing implicit price for a particular attribute represents both producersÕ costs of providing the attribute and consumersÕ willingness-to-pay for the attribute.  There are several common issues associated with hedonic models, the most important of which of which is model misspecification. Economic theory provides no guidance on choosing the appropriate functional form for the hedonic price function !33 (Chau and Chin 2003; Halvorsen and Pollakowski 1981). Following Yim et al. (2014) and Teuber and Hermann (2012), the Box-Cox Test was used to determine the appropriate functional form for the hedonic price functions in this study (Box and Cox 1964). Three functional forms were considered: linear, log-linear and inverse square root. Results from the Box-Cox Test suggest that the log-linear functional form outperforms the other specifications and was thus used in this study. Other common issues present in hedonic analyses include heteroscedasticity and multicollinearity (Constanigro et al. 2011). In this analysis, the Breusch-Pagan-Godfrey test and variance inflation factors are used to detect the presence of heteroscedasticity and multicollinearity respectively.   2.3.3 Application of Hedonic Price Model to Fruit Beverages In this analysis, we estimate hedonic models for 100% fruit juice and fruit drinks. The following hedonic price functions are estimated: !!!"!"#$%&'#$%!!!!!!!"#$!!!!!!!!"#$!"!!!!!!!"#$!!!!!!!!"#!"!!!!!! !!!"!"#$%&"#'(!!!!!!!"#$!!!!!!!!"#$!"!!!!!!!"#$!!!!!!!!"#!"!!!!!! where JuicePrice and DrinkPrice are the price per ounce for fruit juice and fruit drink purchases respectively. Attributes of fruit beverages included in the hedonic price function are classified into four categories: (1) nutrients (nutr), (2) product attributes (prod), (3) packaging (pack) and (4) acquisition attributes (acq). The variables included  in these categories are detailed in Table 2.2.  The first category of attributes, nutrients, is comprised of the key nutrients found  !34  Table 2.2 Variable DefinitionsVariableDefinitionUnitBase VariableDependent Variables   Per Unit PriceFruit juice/drink price per oz$/oz---Nutrients   AntioxidantsAntioxidant content mg/100g---   Calcium and Vitamin DCalcium and Vitamin D contentmg/100g---   Vitamin CVitamin C Contentmg/100g---   Total SugarSugar Contentg/100g---Product Attributes   Brand10 top brand name dummiesDVOther Brands   Private LabelPrivate label product DVNon-Private Label   DietDiet/low-calorie productDVNon-Diet   FlavorFlavor of fruit beverage:other citrus, berry, lemonade, apple, mixed fruit, vegetable, grape, other flavorsDVsOrange   OrganicOrganic productDVNon-OrganicPackaging   Package SizeSet of 3 package size dummies: oversized  (! 89oz), standard (59-64oz) and single serve (" 24 oz)DVsOther SizesAcquisition Attributes   Low-Access TractAcquisition in low-access census tract at 1/10 mi urban/ruralDVNon-Low-Access   Low-Income TractAcquisition in low-income census tractDVNon-Low-Income   RegionItem purchsed in the West, South or MidwestDVsNortheast   SeasonItem purchased in fall, winter or springDVsSummer   Store TypeSet of 4 dummies for store type: convenience, club store, discount store and supermarketDVsGrocery Store   WICWIC payment used DVNon-WIC   Coupon UsedAmount of coupon(s) applied $---   Store SavingsAmount of store savings applied$---!35 in fruit beverages that are sought by consumers and advertised by manufacturers. These  include: antioxidants, Vitamin C, Vitamin D, Calcium and sugar3. Calcium and Vitamin D are combined into a single variable as fruit beverage manufacturers tend to fortify fruit  beverage products with Vitamin D in conjunction with Calcium (Biancuzzo et al. 2010; De Lourdes et al. 2012; Table 2.1). With the exception of sugar, a price premium is expected for each of these nutrients due to the health benefits they provide consumers, as well as the added costs manufacturers incur when fortifying fruit beverages. Conversely, we hypothesize that sugar will garner a negative price premium as consumers and manufacturers seek to limit its content in fruit beverages. Interaction terms between the nutrients and a diet (zero or low-calorie) fruit drink dummy variable, are also included in the nutrients category. These interaction terms are included to distinguish between the price premium for nutrients in diet and non-diet fruit drinks. The second category, product attributes, consists of five variables representative of the productsÕ characteristics: flavor, brand name, private label, diet and organic. In their studies on fruit beverages, Szathvary and Trestini (2014) and Weemaes and Riethmuller (2001) found that price premiums for fruit beverages varied by flavor. In this study we include the following top-selling fruit beverage flavors: orange, other citrus, berry, apple, lemonade4, mixed fruit, vegetable5, grape and other flavors (Euromonitor 2015); orange is the reference flavor. In addition to flavor, dummy variables for brands with a market share greater than 5% in the fruit juice and drink markets are included in the model. Depending on the brandsÕs reputation, prior hedonic analyses have found that !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!3 For brevity, sugar is included in the nutrients variable category despite it classification as a carbohydrate. 4A lemonade dummy variable is not included in the fruit juice price functions as no lemonade is 100% juice. 5 A vegetable dummy variable is included only in the juice price functions as all vegetable beverages are 100% juice !36 brand names garner both positive and negative price premiums (Morgan et al. 1979; Szathvary and Trestini 2014). A dummy variable is also included for private label products, with the expectation that these products are associated with negative price premiums (Sethuraman & Cole 1999). Two additional product attributes are included in the analysis: organic and diet6. Past studies have found that organic beverages garner significant price premiums (Szathvary and Trestini 2014; Gulseven and Wohlgenant 2014). Diet fruit beverages are also expected to garner a price premium given their value-added attribute of having fewer calories. In the third category of variables, packaging, three variables are included to characterize each fruit beverageÕs package size: standard, single serve and oversized. In their analysis of soda prices, Fox and Melser (2014) found that the relationship between package size and price is non-linear. In general, single-serving containers of soft drinks cost more per ounce. than standard sized containers (two liters). The authors further find that oversized packages (24 packs) cost less per ounce than standard sized packages. Analogous to the findings of Fox and Mesler (2014), we expect that single-serving fruit beverages will garner a positive price premium and that standard and oversized fruit beverage containers will garner a negative price premium relative to other sizes.  The final category of variables describes the attributes of the acquisition, including where, when and how the fruit beverages were purchased. Store type, region and census tract characteristics are included to characterize where the fruit beverages were purchased. Szathvary and Trestini (2014) found that fruit beverages sold at supermarkets garner a price premium over other retailer types. Past studies have also !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!6 Note that only fruit drinks can be classified as ÒdietÓ !37 found significant heterogeneity in the regional consumption of food products (Morgan et al. 1979; Drescher et al. 2008; Singh et al. 2015). Dummy variables for acquisitions made in low-income and low-access census tracts are also included in the model. Due to a lack of competition from other retailers, food price tend to be higher in low-access census tracts (Ver Ploeg 2010). Low-income census tracts are also expected to charge higher prices in that they have fewer chain retailers and supermarkets (Ver Ploeg 2007, Powell et al. 2007). Seasonal dummy variables are included in the price functions to account for price variation due to the seasonality of fruit production and demand. We also account for whether WIC was used as payment for fruit juice. Because the size, flavor and brand that WIC participants can purchase is predetermined, these consumers likely do no not consider price when purchasing fruit juice. Finally, the dollar amount of coupons and store savings applied to fruit beverages are included, with the intuitive hypothesis that coupon usage and store savings are associated with lower prices.   2.4 Data 2.4.1 Data Set  The National Household Food Acquisition and Purchase Survey (FoodAPS) data set was used for the analysis in this study (2016b). Funded by the United States Department of Agriculture (USDA) Economic Research Service (ERS) and the Food and Nutrition Service (FNS), FoodAPS is a national survey of 4,826 households. Collected between April 2012 and January 2013, the FoodAPS dataset contains a record of each householdÕs food at home (FAH) and food away from home (FAFH) acquisitions over a one-week period. Entry and exit surveys were administered to households in order to collect !38 demographic and socioeconomic data. The FoodAPS dataset also contains supplemental data on the nutritional composition of all food items purchased, food acquisition characteristics, payment methods and product attributes.  During the one-week acquisition period, 1,852 households in the FoodAPS dataset purchased fruit beverages for at home consumption. These households made a total of 4,166 fruit beverage purchases, of which 42% were fruit juice and 58% were fruit drink purchases. Fruit beverage items that had a price of zero and were not associated with coupons or store discounts were removed from the dataset. Each fruit beverage item purchased was then classified as either 100% fruit juice or as a fruit drink based on the percentage of juice it contained and its sugar content. This resulted in a final sample size of 1,362 fruit juice and 1,832 fruit drink purchases.   2.4.2 Descriptive Statistics Descriptive statistics for the fruit beverage prices and attributes are presented in Table 2.3. Comparing fruit juice to fruit drinks, we find that fruit juice is slightly more expensive, with an average price of $0.07 per ounce versus $0.06 per ounce for fruit drinks. Of particular interest to this study are the differences in nutritional composition of fruit juice and fruit drinks. The descriptive statistics reveal that fruit juice has significantly higher levels of all key nutrients in comparison to fruit drinks. In particular, fruit juice contains approximately 500% more antioxidants, 400% more Calcium and Vitamin D, and 225% more Vitamin C than fruit drinks. Despite having different vitamin  and mineral contents, fruit juice and fruit drinks contain similar amounts of sugar. On average, fruit juice and fruit drinks contain 10.24 and 11.63 grams of sugar per 100g  !39  Table 2.3 Descriptive Statistics By Fruit Beverage TypeVariableMean (Juice)Mean (Drinks)Dependent Variables   Per Unit Price**0.070.06Nutrients   Antioxidants***0.830.16   Calcium and Vitamin D***31.287.4   Vitamin C***30.0813.08   Total Sugar***10.2411.63Product Attributes   Brand***0.05-0.080.05-0.10   Private Label***0.210.05   Diet---0.05   Flavor***0.06, 0.05, ---, 0.17, 0.12, 0.07, 0.06, 0.070.02, 0.10, 0.13, 0.02, 0.44, ---, 0.04, 0.14   Organic0.050.06Packaging   Package Size*, ***, ***0.11, 0.13, 0.450.13, 0.22, 0.33Acquisition Attributes   Low-Access Tract***0.350.29   Low-Income Tract***0.480.59   Region***0.26, 0.31, 0.250.26, 0.38, 0.19   Season***0.37, 0.04, 0.130.33, 0.06, 0.16   Store Type***0.02, 0.04, 0.02, 0.860.03, 0.02, 0.05, 0.84   WIC0.08---   Coupon Used0.020.01   Store Savings**0.170.14*Means for fruit juice and drinks differ at the 0.10 level, ** 0.05 level and *** 0.01 level!40 serving respectively.  Putting these numbers into perspective, Figure 2.1 presents the percentage recommended daily value (%DV) of key nutrients provided by the fruit beverages in the data set (US 2013c). On average, an 8oz serving of fruit juice provides 115%, 7% and 6% of the %DV of Vitamin C, Calcium and Vitamin D, and antioxidants, while fruit drinks provide 50%, 2% and 1% of the %DV respectively. Comparing sugar content, an 8oz serving of fruit juice contains 73% of the %DV, compared to 83% for fruit drinks. In addition to nutrients, the descriptive statistics reveal key differences in the product attributes of fruit juice and drinks. The distribution of flavors varies significantly between fruit juice and fruit drink purchases. For fruit juice, orange is the top-selling flavor, followed by apple, mixed fruit, vegetable/other flavors, grape/other citrus and berry. Mixed fruit is the top selling fruit drink flavor, followed by lemonade/other flavors, orange, berry, grape and apple/other citrus. These distributions are similar to those reported by Euromonitor (2015), suggesting that the fruit beverage purchases in the FoodAPS dataset are representative of all US fruit beverage acquisitions. We also find that while the market share of the top five fruit juice and drink brands are similar, private label products comprise 21% of fruit juice purchases, but only 5% of fruit drink purchases. According to Abate and Peterson (2005), the narrow price difference between private label and branded juice drinks is a possible explanation for private label productsÕ low market share in the fruit drink segment. Considering packaging, a greater share of fruit drinks are purchased in single serve and oversized  packages, 22% and 13%, versus 13% and 11% for fruit juice. Conversely, a greater share of fruit juice purchases are in standard size packages, 45%, versus 33% for fruit drinks. !41  Figure 2.1. Percent Daily Value of Key Nutrients in Fruit Beverages Based on a 2,000 Calorie Diet The characteristics of fruit beverage acquisitions also differ significantly between fruit juice and fruit drinks, with both regional and seasonal heterogeneity. Fruit drinks purchases are more prevalent in the Southern portion of the United States, while fruit juice purchases are more prominent in the Midwest. Where acquisitions are made also varies significantly by fruit beverage type. While the shares of fruit drinks and juice purchased at supermarkets are similar, a greater share of fruit drink purchases are made at convenience retailers and discount stores, while a greater share of fruit juice purchases are made at club stores. We also find that fruit drink (fruit juice) purchases are more common in low-income (low-access) census tracts. Looking at payment type, 8% of fruit juice purchases were made using WIC benefits. While savings from coupons are comparable, store savings are, on average, 20% greater for fruit juice than fruit drinks.  0%10%20%30%40%50%60%70%80%90%100%110%120%Vitamin C!AntioxidantsCalcium and Vitamin DSugarFruit JuiceFruit Drinks!42 2.5 Results Estimates of the log-linear fruit juice and fruit drink hedonic price equations are obtained using ordinary least squares regression techniques and are presented in 2.3. The estimated models explain a significant portion of the variation in fruit juice and fruit drink prices, with r-squared values of 0.64 and 0.62 respectively. Breush-Pagan test results suggest the presence of heteroskedasticity, thus we calculate White-Huber standard errors.   2.5.1 Nutrients  Of particular interest to this study, are the price premiums associated with nutrients and sugar in fruit beverages. The hedonic price estimates in Table 2.4 show that price premiums for nutrients and sugar vary between fruit juice, non-diet fruit drinks and diet fruit drinks.  Looking first at fruit juice, we find that all nutrients garner a price premium. Adding an additional mg of antioxidants (Vitamin C) to fruit juice leads to a 5% (0.01%) increase in the price per ounce. For a standard 60oz container, this corresponds to a $0.21 and $0.01 premium for an additional mg of antioxidants and Vitamin C respectively. While Calcium/Vitamin D also garners a price premium, the premium itself is extremely small. Adding an additional mg of Calcium/Vitamin D increases the per ounce price of fruit juice by just 0.0005%, or a $0.002 premium for the standard 60oz container. These price premiums for nutrients in fruit juice likely reflect both manufacturersÕ costs and consumersÕ willingness-to-pay. For manufacturers, fortifying fruit juice with vitamins  and minerals leads to increased production costs. On the demand side, consumers may  pay a premium for fruit juice containing more nutrients given their positive health  !43  Table 2.4 Fruit Beverage Hedonic Price Function Estimates VariableCoeff.SECoeff.SENutrients   Antioxidants0.05*0.030.010.04   Diet*Antioxidants-------0.36**0.17   Vitamin C2.20E-03***6.33E-030.01***1.32E-03   Diet*Vitamin C-------0.02**0.01   Calcium/Vitamin D4.95E-04**2.20E-040.02***0.01   Diet*Calcium/Vitamin D-------0.06***0.03   Total Sugar2.75E-046.33E-040.01***3.06E-03   Diet*Total Sugar-------0.06**0.03Product Attributes   Brand 10.070.06-0.31***0.04   Brand 20.25***0.03-0.08*0.05   Brand 30.15***0.030.39***0.05   Brand 40.050.04-0.010.04   Brand 50.29***0.040.050.05   Private Label-0.25***0.03-0.22***0.04   Other Citrus0.21***0.050.100.07   Berry0.18***0.050.23***0.05   Apple-0.020.040.33***0.13   Lemonade------0.24***0.06   Mixed Fruit0.040.040.20***0.04   Vegetable-0.42*0.24 ------   Grape0.16***0.060.050.09   Other Flavors0.19***0.070.010.05   Diet------1.43***0.44   Organic0.30***0.070.86***0.06Packaging   Oversized Package-0.38***0.04-0.53***0.04   Single Serve Package0.67***0.060.66***0.06   Standard Size Package-0.30***0.04-0.06*0.04Acquisition Attributes   Low-Access Tract9.03E-030.020.020.03   Low-Income Tract-0.08***0.02-0.12***0.02   Fall0.04*0.02-0.020.03   Spring0.12***0.03-0.06*0.04   Winter0.010.04-0.12**0.06   Midwest-0.08**0.03-0.030.04   South-0.06*0.04-0.06*0.04   West3.43E-030.040.020.04   WIC0.12***0.03------   Convenience 0.26***0.070.180.12   Club Store-0.140.100.020.08   Discount Store-0.42***0.120.030.07   Supermarket-4.20E-030.06-0.010.05   Coupon Used-0.52*0.32-0.100.09   Store Savings-0.18***0.02-0.15***0.04Constant-2.87***0.07-3.82***0.07R-Squared*, ** and *** denote significance at the 10%, 5% and 1% levelFruit Juice        (N=1,362)Fruit Drinks     (N=1,832)0.640.62!44 benefits. Unlike nutrients, sugar is not associated with a price premium in fruit juice. This  finding is likely the result of the fact that manufacturers do not incur the cost of adding sugar to fruit juice, as juice naturally contains large amounts of sugar (OÕNeil and Nicklas 2008). Consumers may also pay a premium price for fruit juice containing more sugar given the growing public concern over the adverse health effects of sugar  consumption. Unlike fruit juice, an additional gram of sugar is associated with a 1% price  premium for non-diet fruit drinks. For the standard 60oz container, this corresponds to $0.04 for each additional gram of sugar. This premium is partly attributable to the added costs manufacturers face when adding sugar to non-diet fruit drinks. On the demand side, consumers that prefer the taste of sweeter drinks may also be pay a premium for non-diet fruit drinks containing additional sugar.  Also differing from fruit juice, only select nutrients garner price premiums in non-diet fruit drinks; Vitamin C and Calcium/Vitamin D are associated with a price premium, while antioxidants are not. Adding an additional mg of Vitamin C (Calcium/Vitamin D) to a non-diet fruit drink leads to a 1% (2%) increase in the price per ounce. For the standard 60oz container, this corresponds to a $0.04 and $0.07 premium for Vitamin C and Calcium/Vitamin D respectively. As with fruit juice, these premiums likely reflect the costs incurred by manufacturers to fortify the fruit drinks with nutrients, as well as consumersÕ willingness to pay for nutrients given their positive health benefits. However, the price premiums for Vitamin C and Calcium/Vitamin D in non-diet fruit drinks are larger than those for fruit juice. One plausible explanation for this difference is the fact that fruit drinks contain less naturally occurring nutrients than fruit juice. (Harris et al. !45 2011; Empty 2015). Thus, to achieve the same level of nutrients, non-diet fruit drink manufacturers must incur higher fortification costs than fruit juice manufacturers.  Differing from both non-diet fruit drinks and fruit juice, nutrients and sugar in diet fruit drinks are associated with negative price premiums. An additional mg of antioxidants, Vitamin C and Calcium/Vitamin D leads to a 35%, 1% and 4% decrease in the price per ounce respectively. For the standard 60oz container, this corresponds to a $1.26, $0.04 and $0.14 discount for an additional mg of antioxidants, Vitamin C and Calcium/Vitamin D. Similarly, an additional gram of sugar leads to a 5% decrease in the price per ounce for diet fruit drinks, or a $0.18 discount for the standard 60oz container. Given that manufacturers still incur additional costs when adding nutrients and sugar to diet fruit drinks, these negative price premiums suggest that diet fruit drink consumers pay a premium to reduce nutrients and sugar. Given the nature of diet fruit drinks, consumers intuitively seek to reduce to nutrients and sugar in diet fruit drinks in order to reduce the fruit beveragesÕ caloric content; by reducing the nutrient and sugar content of a diet fruit drink, one also decreases the calories in the drink.  Based on these results, three main generalizations are made about the price premiums for nutrients and sugar in fruit beverages: 1. All nutrients garner premium prices in fruit juice 2. Sugar and select nutrients garner premium prices in non-diet fruit drinks 3. All nutrients and sugar are associated with price discounts in diet fruit drinks  2.5.2 Product Attributes In addition to nutrients, several product attributes also garner price premiums in fruit !46 beverages. As found by Szathvary and Trestini (2014), nearly all of the top fruit juice and fruit drink brands garner a price premium, ranging from 15% to 39%. However, fruit drink Brands 1 and 2 have negative coefficients, suggesting that these are discount or value brands. Unlike branded products, private label fruit beverages are associated with price discounts. Relative to branded products, private label fruit juice and fruit drink products cost 25% and 22% less per ounce respectively. Results also highlight flavorsÕ effect on fruit beverage prices. Relative to orange juice, berry, grape, other citrus and other flavors garner price premiums ranging from 16% to 21%. Conversely, vegetable flavored juice is shown to cost 42% less per ounce than orange juice. Considering fruit drinks, nearly all flavors are associated with higher prices than orange flavored drinks, with price premiums ranging from 20% for mixed fruit to 33% for apple flavored drinks. The estimation results also indicate that organic and diet fruit beverages are associated with significant price premiums. Compared to non-organic fruit beverages, organic fruit juice and fruit drinks price is 30% and 86% higher per ounce respectively. This finding is comparable to that of Szathvary and Trestini (2014), who found a 48% price premium for organic fruit beverages sold in Australia. Diet fruit drinks also garner a substantial price premium, with prices 143% higher than those of non-diet fruit drinks.  2.5.3 Packaging Similar to Fox and MelserÕs (2014) findings for soft drinks, fruit beverages sold in single serve packages are associated with higher prices, relative to other package types. Single serve packages garner 67% and 60% price premiums for fruit juice and fruit drinks. Also mirroring Fox and Melser (2014), fruit beverages sold in standard sized and oversized !47 packages are associated with lower per ounce prices than other package types. This negative price premium is greater for oversized packages than for standard sized packages, with oversized packages priced 38% (53%) less per ounce for fruit juice (drinks) and standard sized packages priced 30% (6%) less per ounce respectively.   2.5.4 Acquisition Attributes Several attributes of the acquisition event also affect the price of fruit beverages. The estimation results indicate there is both seasonal and regional variation in fruit juice and fruit drink prices. Further, while fruit drinks prices appear to be similar across retailer types, the type of store fruit juice is purchased at has a significant impact on its price. Relative to grocery stores, fruit juice prices are 26% higher at convenience retailers and 42% less at discount retailers.  Further, the estimation results confirm that store savings and coupon usage are associated with lower prices for fruit beverages. When store savings are applied to an item, prices decrease by 18% and 15% for fruit juice and fruit drinks. Similarly, fruit juice prices decrease by 52% when a coupon is used; for fruit drinks, the coefficient for coupon usage is negative but not significant. The hedonic price equation estimates further indicate a 12% price premium for fruit juice purchased using WIC benefits. This finding supports this studies hypothesis that because the size, flavor and brand that WIC participants can purchase is predetermined, WIC consumers likely do not consider price when purchasing fruit juice. Dummy variables for acquisitions made in low-income census tracts also significantly affect fruit beverage prices. Low-income census tracts are associated with !48 fruit juice and fruit drink prices 8% and 12% less than those in non-low-income census tracts. This is likely attributable to retailers charging lower prices in low-income areas where households have less disposable income.   2.6 Conclusions Given the increased importance consumers and manufacturers have placed on the functional nutrients found in fruit beverages, as well as the changing federal guidelines on fruit beverage consumption, this study sought to determine whether specific nutrients garner price premiums in fruit beverages sold in the US. Using the National Household Food Acquisition and Purchase Survey, hedonic price models for fruit juice and fruit drinks are estimated to determine whether specific nutrients, product characteristics, packaging type and acquisition characteristics are associated with price premiums. Based on the results from the hedonic price models, three generalizations are made about the price premiums for nutrients and sugar in fruit beverages: (1) all nutrients garner premium prices in fruit juice, (2) sugar and select nutrients garner price premiums in non-diet fruit drinks and (3) all nutrients and sugar are associated with negative price premiums in diet fruit drinks. Findings further suggest that product attributes such as brand, flavor, organic labels, diet labels and package type, and acquisition characteristics such as store type, region, season and payment type are associated with price premiums in fruit beverages. This studyÕs price premium estimates for nutrients can provide valuable insight to fruit beverage manufacturers, particularly in their design of future marketing initiatives and new product development. Given the growing concern over the healthfulness of fruit !49 beverages in recent years, manufacturers are employing marketing tools such as front-of-package labels and advertisements to emphasize the nutrients found in fruit beverages. Estimates of price premiums for these nutrients can help fruit beverage manufacturers determine which specific nutrients to emphasize in these marketing initiatives. Assuming that the marginal costs of different nutrients are similar, fruit beverage manufacturers should emphasize the nutrients that garner the largest price premium. Results from this study suggest that fruit juice marketing initiatives should focus on antioxidants, while non-diet drink marketing efforts should emphasize Vitamin C and Calcium/Vitamin D. For diet fruit drinks, all nutrients are associated with a negative price premium, suggesting that marketing efforts should focus on calorie content instead of nutrient content.  Estimates of nutrient price premiums can also guide fruit beverage manufacturers in product development. In developing a new fruit beverage product, manufacturers must determine whether or not to fortify the product with nutrients and, in the case of fortification, which specific nutrients should be used. Price discounts suggest that fortification will not lead to increased returns for diet fruit drink manufacturers. However, positive price premium estimates from this study suggest that fruit beverage manufacturers should consider fortifying fruit juice and non-diet fruit drinks with certain nutrients. Fruit juice and non-diet fruit drink manufacturers can compare the price premiums for specific nutrients estimated in this study to their marginal costs of fortification to determine which specific nutrients to use in fortifying their product.    !50 CHAPTER 3 RETHINKING HOUSEHOLD DEMAND FOR FOOD DIVERSITY  3.1 Introduction According to the 2015 Dietary Guidelines for Americans (DGA), food variety is an essential component of a healthy diet (US 2015). The DGA encourages variety across food subgroups, which respectively contain different micronutrients and macronutrients (US 2015). Murphy et al (2006) and Foote et al (2004) find evidence that diversity across food subgroups is positively associated with nutritional adequacy. Further, food diversity is credited with helping maintain a healthy body weight and reducing the risk of diet-related diseases, including heart disease and diabetes (British 2007).  The link between food diversity and health, has served as motivation for several economic studies on the diversity of household food expenditures (Thiele and Weiss, 2003; Lee and Brown 1989; Shonkwiler et al 1987; Lee 1987). In these studies, food diversity is measured using indices of household expenditures across food categories designated based on product codes (i.e. food classification systems created by market intelligence agencies, retailers and government). Under a product code approach, convenience foods (i.e. foods which have processing added by a manufacturer/distributor to provide time-savings to consumers) are classified based on their processed form and subgroup (Lee and Lin 2013). This approach conflicts with the DGAÕs definition of food diversity as variety across food subgroups, irrespective of processed form (US 2015).  Relative to studies conducted in the 1980Õs, classification of convenience foods is of particular importance in contemporary analyses of food diversity demand. Convenience foods have gone from an emerging trend in the 1980Õs, to a staple in !51 household diets, with Okrent and Kumcu (2016) finding that convenience foods comprise nearly 75% of all US food expenditures (Capps et al 1985). Given this changing food landscape, product code-based indices are increasingly capturing diversity in household expenditures across different processed forms as opposed to food subgroups. This raises the question of whether product code-based food diversity indices are still an accurate indicator of nutritional adequacy or whether product-code indices now better represent diversity in product types and form. With this question in mind, it is necessary to reassess household demand for food diversity using an index that classifies foods based on their subgroup composition. Recent studies in the nutrition literature by Vadiveloo et al (2014) and Drescher et al (2007) develop healthy food diversity (HFD) indices based on the DGA and the German Nutrition Society Guidelines. In these indices, food categories are designated based on food subgroups, irrespective of processed form. However, the indices are representative of diversity across the volume of foods individuals consume, not household food expenditures as in economic studies on the demand for food diversity. The purpose of this study is to develop a group-based food diversity index based on the US HFD, which represents diversity in household expenditures across food subgroups. This group-based food diversity index is then applied to examine the relationship between the demand for food diversity, food expenditures and household characteristics. Further, findings using the new group-based food diversity index are compared to those obtained using the product code-based food diversity index.    !52 3.2 Literature Review Household demand for food diversity has been widely examined in the economic literature. Foundational studies by Theil and Finke (1983) and Jackson (1984) examine whether the demand for diversity varies with income. Both studies consider diversity across broad commodity groups, including food. Jackson (1984) develops a hierarchical demand system in which the consumption set changes with income, while Theil and Finke (1983) measure diversity using the Herfindahl-Hirschman and entropy indices. Both studies find evidence that the demand for consumption diversity increases with income.  Focusing specifically on food, Lee (1987) and Shonkwiler et al (1987) examine the relationship between food diversity, expenditures and demographics. Both studies employ count indices, which measure the number of different food categories households purchase. The studies conclude that increased food expenditures are associated with increased demand for food diversity. Lee (1987) further finds a positive relationship between food diversity and household size.  Lee and Brown (1989), Jekanowski and Binkley (2000) and Thiele and Weiss (2003) measure food diversity based on the distribution of expenditure shares across food categories. Types of distributional measures utilized in these studies include the Simpson and entropy7 indices, which are defined as: (1)  !!!!!!!!!!! (2)  !!!!!!"#!!!!!!,                                                                                                   where !! represents the budget share for food category i. The Simpson and entropy !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!7 The entropy index is also commonly referred to as the Shannon Index !53 indices range from !!!!!! and !!!!"#!! respectively. For both indices, zero indicates that households buy from a single food category, while the maximum value refers to households buying equal shares of all n categories. All three studies find that food expenditures or income are positively related to food diversity demand. Lee and Brown (1989) further find that the food diversity demand is positively related to receiving Supplemental Nutrition Assistance Program (SNAP) benefits, as well as household size, age and gender. Lee (1987) finds evidence of an inverse relationship between racial and food diversity, while Thiele and Weiss (2003) find that the food diversity demand is greater among single, male households and households with children.  3. 3 Theoretical Background Following Lee and BrownÕs (1989), household demand for food diversity is derived from the traditional consumer utility maximization problem:  (3) !!"#!!!!!!!                                                                                         !!!!!!!!!!!!                                                                                        !!!,                                                                            0 where q is a vector of household quantity demanded of n commodity categories, p is a vector of prices for the n commodities, m represents household expenditures on all commodities and z is a vector of demographic variables. Solving this problem yields a set of commodity demand equations. Assuming weak separability, the quantity demanded for food can be considered separately from that of other commodity categories and is defined as:  !54 (4)  !!"!!!"!!!!!!!!!                                                                     where !!" denotes household quantity demanded of food category i, !! is a vector of prices and !! represents total household food expenditures. Expenditure shares for each food category, !!", are then defined as: (5)  !!"!!!"!!!"!!!!!"!!!!!!!!!                                                     Thus, a distributional measure of food diversity is given by: (6) !!!!!"!!!!"!!!!!!!!!!!!!!!!,                       where D is a measure for food diversity, typically the entropy or Simpson index.  3.4 An Alternative Group-Based Food Diversity Index In calculating the diversity measure defined in Equation 6, all past economic studies on food diversity demand have used product codes to designate the i food categories. Product codes refer to classification systems created by market intelligence agencies, food retailers and government. Examples include the National Food Consumption Survey 15-digit code system and the Bureau of Labor Statistics 6-digit code system. Classification of convenience foods is central to the product code-based approach. Unlike basic foods, which are raw or minimally processed, convenience foods have been processed or prepared to some extent by a manufacturer or food distributor with the express purpose of creating a time savings or an ease of preparation, or elimination of the need for preparation, by the consumer (Lee and Lin 2013). In the literature, foods are typically categorized into four groups based on their processed form: (1) basic foods, (2) complex ingredients, (3) ready-to-cook and (4) ready-to-eat (Okrent & Kumcu 2016). !55 While basic foods refer to raw or minimally processed foods, the remaining three categories represent different forms of convenience foods. Complex ingredients refer to processed foods used in producing a meal or snack (ex: vegetables, frozen meat), ready-to-cook refers to meals and snacks that require minimal preparation beyond heating or adding hot water (ex: frozen entrees, soup) and ready-to-eat refers to meals and snacks to be consumed as is (ex: refrigerated entrees, food away from home, canned fruit).  Under a product code-based classification scheme, basic and convenience foods are classified into separate categories. For example, instead of a single meat category, raw meat and frozen meat are classified as separate food categories, while frozen dinners are considered a separate category from the meat and vegetables they contain. Because convenience foods are a composition of basic foods in different processed forms, use of a product code approach leads to a diversity measure that not only captures diversity across food subgroups, but also diversity across processed form. This product code-based approach conflicts with the DGAÕs definition of food diversity, in which diversity refers to variety across food subgroups, irrespective of processed form (US 2015). Recent studies in the nutrition literature by Vadiveloo et al (2014) and Drescher et al (2007) develop Healthy Food Diversity (HFD) indices that designate food categories based on subgroups as defined in the DGA and German Nutrition Society Guidelines. These indices are an improvement over existing nutrition measures of food diversity, such as the Healthy Eating Index, which only measure consumption of select subgroups. Mirroring its German predecessor, the US HFD index developed by Vadiveloo et al (2014) is defined as: (7) !"!!"#!!!!!!!!!!!!! !56      s.t. !!!!!!!!!!!!! where !! is the share of each food subgroup i based on the volume of the total diet, hv is the health value of the individualÕs diet and !!! is a health factor based on the DGA daily intake recommendations for each food subgroup. The index is calibrated by dividing by the maximum value of !!!!!!, resulting in a range of !!!!!!!!!.  The US HFD index was validated as a measure of nutritional adequacy through its strong correlation with individual dietary quality indicators (Vadiveloo et al 2014).  While the US HFD index uses a group-based food classification scheme and is a validated measure of nutritional adequacy, it is calculated using volume shares as opposed to expenditure shares which economic studies use to understand the demand for food diversity. The US HFD index reflects diversity across the volume of foods individuals consume and uses actual consumption data from dietary recall surveys. In contrast, economic studies on food diversity demand tend to use food acquisition data from scanner datasets and household surveys to analyze diversity in household food expenditures. Complementing analyses conducted using volume shares, use of expenditure shares allows economists to consider another dimension of food diversity and therefore better inform food assistance and nutrition policy.  Thus, in this study, a group-based food diversity index is developed based on the US HFD, which represents diversity in household expenditures across food subgroups. This group-based index is then applied to reassess household demand for food diversity. Following Vadiveloo et al (2014), we designate the !!!!!!! food categories based on the USDAÕs MyPlate food subgroups (US 2016a). Detailed in Figure 3.1, this group-based approach to classification yields 23 categories. For each basic food item!!! !57  Figure 3.1 Group and Product Code-Based Food Categories CheeseMeatSoyMilkPoultryDark Green Veg.YogurtSeafood (Hi-Fat)TomatoesCitrus, Melon, BerriesSeafood (Low-Fat)Other Red/Orange Veg.Other Whole FruitCured MeatPotatoesFruit JuiceOrgan MeatOther StarchesRefined GrainsEggsOther VegetablesWhole GrainsNuts/SeedsFood Away From HomeFrozen MeatPies/CakesAppetizers - FZaFruit - FZPies - FZAsian FoodGelatin/PuddingPizza -FZBaked BeansHot CerealPizza - RFGBakery Goods - RFGbIce Cream Cones/MixPizza ProductsBakery SnacksIce Cream/SherbetPlain Vegetables - FZBaking MixesInstant PotatoesPopcornBaking NeedsJellies/Jams/HoneyPotatoes/Onions - FZBottled Juices - FZJuices - FZPoultry Substitutes - FZBread/Dough - FZLuncheon MeatsPoultry - FZBreakfast Food - FZLunches - RFGPrepared Deli Food - RFGCake - SSMeat - FZPrepared Vegetables - FZCanned/Bottled FruitMeat - SSProcessed PoultryCheesecakesMexican FoodsSalad ToppingsChocolate CandyMexican SauceSalad/Coleslaw - RFGCold CerealMicrowave DinnersSalty SnacksCookiesNon-Fruit Drinks - SSSeafood - FZCrackersNon-Chocolate CandySeafood - SSDesserts - RFGNovelties - FZSide Dishes - RFGDesserts/Toppings - FZOther Breakfast FoodSnack BarsDinners - SScOther Foods - FZSnack Nuts/SeedsDinners/Entrees - FZOther SnacksSoupDips - SSPancake MixesSoup/Sides - FZDips - RFGPastaSpaghetti/Italian SauceDough/Biscuit - RFGPasta - FZSpecialty Nut ButterDry Dinner - Add MeatPasta - RFGStuffing MixesDry Dinner MixesPastry DoughnutsToaster Pastries/TartsEnglish MuffinsPeanut ButterTomato ProductsEntrees - RFGPickles/Relish - RFGTortilla/Egg Roll WrapsFish/Seafood - FZPickles/Relish/OlivesVegetables - SSFresh Bread/RollsaFrozen bRefrigerated cShelf-stableIRI Basic Food Categories / MyPlate Food Subgroups (N=23)IRI Convenience Food Categories (N=91)!58 !!!!!, we denote the price and quantity demanded as !!" and !!". Similarly, for each convenience food item !!!!!!!, we denote the price and quantity demanded as !!" and !!" Total household food expenditures are denoted by !!. It is important to note that each food item is comprised of at least one food category i. For example, a macaroni and cheese frozen dinner contains refined grains, cheese and milk. We alter Equation 5 to account for this relationship as follows:  (8)  !!"!!!!!"!!!!!!"!!"!!!!!"!!!!!!"!!!"!!                                                          where !!" is a binary variable indicating which food subgroup i  corresponds with basic food item b and !!" represents the share of convenience food item j comprised by food subgroup i. Table 3.1 provides an overview of the share of each food subgroup comprised  by basic and convenience foods; for 13 of the subgroups, convenience foods comprise at least 50% of expenditures and thus will have a large impact on the diversity of household  expenditures across food subgroups. For example, over 65% of household expenditures on the cheese subgroup are attributable to convenience foods. The expenditure shares in Equation 8 are then used to calculate the group-based entropy index of food diversity as follows:  (9)  !!!!!!!"!!!"#!!!!"!!!!!!!"#!!!"!                                                   Deviating from Vadiveloo et alÕs (2014) US HFD index, a health value is not incorporated into Equation 9 given that it is calculated using expenditure shares. In the US HFD index, the health value is calculated based on subgroup volume shares and the DGA subgroup recommended daily values (RDV). Household food expenditures are not an accurate measure of volume, as they represent food acquisitions and not necessarily  !59  individual food consumption. Further, food expenditures depend not only on quantity, but also on price variation due to quality. Thus, weighting Equation 9 by a health value will not result in a food diversity index that also represents adherence to the DGAÕs RDVs, as in Vadiveloo et al (2014). In order to compare the group and product code-based food diversity indices, the entropy index in Equation 9 is standardized by dividing by log(23) (Tuomisto 2012). This results in a group-based food diversity index that ranges from !!!, where zero refers to households buying from a single food category and !"#!refers to buying equal shares of Food SubgroupBasic Food              Categories (N=23)Convenience Food Categories (N=91)Cheese34.9%65.1%Milk61.6%38.4%Yogurt 90.9%9.1%Citrus, Melon, Berries81.8%18.2%Other Whole Fruit84.2%15.8%Fruit Juice67.9%32.1%Refined Grains29.6%70.4%Whole Grains50.0%50.0%Meat58.2%41.8%Poultry34.6%65.4%Seafood (Hi-Fat)50.0%50.0%Seafood (Low-Fat)36.0%64.0%Cured Meat71.7%28.3%Organ Meat20.0%80.0%Eggs50.0%50.0%Nuts/Seeds58.8%41.2%Soy75.0%25.0%Dark Green Vegetables50.0%50.0%Tomatoes34.1%65.9%Other Red/Orange Vegetables4.5%95.5%Potatoes54.1%45.9%Other Starches9.0%91.0%Other Vegetables51.7%48.3%Table 3.1 Share of Food Subgroups Comprised by the Product Code Index's Basic and Convenience Food Categories !60 all 23 food categories.  For comparison, we also calculate a product code-based food diversity index. Under this approach, the i food categories are designated based on the Information Resources Inc.Õs (IRI) food categories. Detailed in Figure 3.1, this results in 114 food categories, of which 23 are basic food and 91 are convenience food categories. Expenditure shares are calculated for each food category following Equation 5. The product code-based entropy food diversity index is then defined as:  (10)  !!"!!!!!"!!"!!!!"#!!!!"!!"!!!!!!!!"#!!!!"!.                                              This product code food diversity index ranges from !!!, where zero refers to households buying from a single food category and one refers to buying equal shares of all 114 food categories.   3.5 Empirical Model As shown in Equation 6, the food diversity indices defined in Equations 9 and 10 can be expressed as a function of expenditure and demographic variables; as in past studies,  prices are excluded due to multicollinearity issues. Following Thiele and Weiss (2003), both food diversity indices are specified as linear in independent variables, with the exception of household size, which is specified in exponential form. This results in the following equations: (11)  !!!!!!!!!"#!!!!!!!!!"#!!"!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!        (12)  !!"!!!!!!!"#!!!!!!!!!"#!!"!!!!!!!                             where !! is the group-based index, !!" is the product code-based index and k=1,É,K  is  !61  the set of all households. Detailed in Table 3.2, explanatory variables are grouped into two categories: (1) expenditures (EXP) and (2) demographics (DEM).  The first category of independent variables, is comprised of expenditures on three types of foods: (1) basic foods, (2) convenience foods and (3) food away from home (FAFH). Past studies find a positive relationship between food diversity and household        expenditures on food (Lee 1987; Lee and Brown 1989; Jekanowski and Binkley 2000).  Table 3.2 Variable Definitions VariableDefinitionUnitBase VariableFood Diversity IndicesGroup IndexGroup based food diversity indexIndex---Product Code Index Product code based food diversity indexIndex---Independent VariablesBasic Food              ExpendituresHousehold weekly expenditures on basic foods$---Convenience Food    ExpendituresHousehold weekly expenditures on convenience foods$---FAFH ExpendituresHousehold weekly expenditures on food away from home $---Household Income Monthly household income $---SNAPHousehold receives SNAPDVNo SNAPFemaleHousehold head is femaleDVMaleHousehold SizeSize of household#---College DegreeHousehold head has college degreeDVNo degreeAgeAge of household head#---HispanicHousehold head is HispanicDVWhiteAfrican-AmericanHousehold head is African- AmericanDVWhiteAsianHousehold head is AsianDVWhiteNortheastHousehold is in NortheastDVWestMidwestHousehold is in MidwestDVWestSouthHousehold is in SouthDVWest!62 Jekanowski and Binkley (2000) explain that as household food expenditures increase,  they purchase a greater variety of higher quality foods.  The second category of variables contains the following demographics: income, SNAP, gender, education, household size, age, race and region. Thiele and Finke (2003) find a positive relationship between household income and food diversity, explaining that households purchase a greater variety of non-essential food products as their income increases. Equations 11 and 12 also include a binary variable indicating whether households receive SNAP food assistance benefits. Lee and Brown (1989) find that increases in SNAP benefits are associated with increases in food diversity.  Past studies further find that food diversity increases at a decreasing rate given an increase in household size, suggesting there are economies of scale in the food diversity (Lee 1987; Lee and Brown 1989; Thiele and Weiss 2003). Thiele and Weiss (2003) further find evidence linking education to food diversity demand. Moon et al (2002) suggest that education increases knowledge of the nutritional benefits of consuming  a varied diet. Similarly, Lee (1987) and Lee and Brown (1989) find that food diversity is greater for female-headed households, and posit that females place greater importance on the nutritional benefits of consuming a varied diet than males. Based on findings by Lee (1987), Lee and Brown (1989), Jekanowski and Binkley (2000) and Thiele and Weiss   (2003), a household headÕs age is expected to have an inverse effect on food diversity.   3.6 Data  Required data on expenditures, food subgroup composition and demographics is obtained from the National Household Food Acquisition and Purchase Survey !63 (FoodAPS). Collected by the USDAÕs Economic Research Service and Food and Nutrition Service between 2012-2013, FoodAPS is a nationally representative survey of 4,826 households. Participating, households recorded all food purchases in a one-week food acquisition diary. Surveys were administered to collect demographic characteristics for each household. The dataset contains a total of 259,124 food purchases and provides descriptions, prices and quantity purchased for each item. Linked with IRI Market  Research data, the dataset is ideal for calculating both group and product code-based diversity indices in that it provides the food subgroup composition for each item, as well as corresponding IRI product code-based categories.  Of the 4,826 households participating in FoodAPS, 165 do not report a food acquisition event and are removed from the sample. Following Murphy et al (2006), we  remove food items whose consumption totals less than one-half of a serving of a food subgroup; this includes goods such as alcoholic and zero-calorie beverages. Free  acquisitions not associated with coupon use are also removed. This results in a final sample of 4,341 households and 200,173 food items. 3.7 Results 3.7.1 Descriptive Statistics Descriptive statistics for food diversity indices, expenditures and demographics are detailed in Table 3.3. On average, FoodAPS households spend the most food dollars,  $49.87 per week, on basic food items, followed by FAFH and convenience foods at  $32.92 and $15.20 per week. Comparatively, households have an average monthly income of $3,950, household size of three people and household head age of forty-six years. Of the households represented, 31% receive SNAP, 34% have a college degree and  !64  38% are from a minority race. A total of 17%, 25% and 36% of households are located in the Midwest, Northeast and Southern regions of the US respectively.  Descriptive statistics strongly suggest that the two indices are representative of  different types of food diversity. Shown in Figure C.1 in Appendix C, the two food diversity indices are only moderately correlated, with a correlation coefficient of 0.54. Using WelchÕs T-Test, the mean group and product code-based food diversity scores  Table 3.3 Descriptive Statistics  VariableMeanSDFood Diversity IndicesGroup Index0.680.12Product Code Index 0.370.17Independent VariablesBasic Food Expenditures49.8749.38Convenience Food Expenditures15.2019.25FAFH Expenditures32.9239.35Household Income 3949.804283.49SNAP0.31---Female0.74---Household Size3.011.73College Degree0.34---Age45.9116.30Hispanic0.20---African-American0.14---Asian0.04---Northeast0.17---Midwest0.25---South0.36---!65  Figure 3.2 Distribution of Group and Product Code Food Diversity Indices differ at the 1% level. With mean values of 0.68 and 0.37, householdsÕ group-based index is nearly double that of their product code-based food diversity score.  Figure 3.2 details the distribution of the two food diversity indices. Both indices are left skewed, with the group and product code indices having a skewness of -1.18 and -0.75. This finding suggests that, for both indices, there are few households with low levels of food diversity. However, the group-based food diversity index is more highly left skewed than the product code-based index. This implies that households distribute their food expenditures more evenly across group-based categories than across product code-based categories. The moderate level of correlation between the two indices, along with differences in means and skewness, supports the notion that the two measures are capturing different forms of food diversity. The group-based index is capturing diversity among food subgroups and thus diversity among the vitamins and minerals needed for a healthy diet. In contrast, the product code-based index is capturing diversity in processed form, as well  !"#!!"$!!"%!!"&!!"'!!!"'#!!"'$!!"'%!!"'&!!"0-.1.11-.2.21-.3.31-.4.41-.5.51-.6.61-.7.71-.8.81-.9.91-1Household FrequencyFood Diversity Index ValueFood Group Food Diversity IndexProduct Code Food Diversity Index!66  as diversity in food subgroups. Given that both indices are calculated using the same set of food items, the lower product code food diversity scores suggest that households have less diversity in the processed form than in the types of food subgroups they purchase.  3.7.2 Regression Results Equations 11 and 12 are estimated using OLS regression techniques, with the resulting  estimates presented in Table 3.4. White-Huber standard errors are calculated to correct for heteroscedasticity. With R2 values of 0.32 and 0.58, the estimated models explain a  significant portion of the variation in the group and product code-based diversity indices.   Coeff.SECoeff.SEBasic Food Expenditures0.001***5.10E-050.002***8.48E-05Convenience Food Expenditures0.001***1.11E-040.003***1.76E-04FAFH Expenditures0.001***3.93E-05-0.001***4.67E-05Household Income 5.26E-083.81E-077.99E-074.66E-07SNAP-0.015***0.004-0.008**0.004Female0.015***0.0040.020***0.004Household Size0.006*0.0040.015***0.001Household Size Squared-0.001***4.11E-04-0.002***0.001College Degree0.013***0.0030.008**0.004Age3.26E-041.01E-050.001***1.17E-04Hispanic0.0040.0040.0050.005African-American-0.017***0.005-0.034***0.006Asian-0.0080.0080.0010.008Northeast-0.0020.005-0.0020.006Midwest0.0030.0050.0050.005South0.0040.0040.0060.005Constant0.558***0.0100.211***0.012R-SquaredNote: *, **, *** denote significance at the 10%, 5% and 1% significance levelTable 3.4 OLS Regressions on Food Diversity Indices (N=4,341)Group Diversity IndexProduct Code   IndexVariable0.330.58!67 3.7.3 Expenditures Expenditures on basic foods, convenience foods and FAFH are found to have a significant, positive effect on householdsÕ group-based food diversity. For all three types  of food expenditures, a $1 increase results in a 0.001 increase in the group-based food diversity index, suggesting that households purchase a greater variety of foods as their  expenditures increase. Putting this into perspective, a $10 increase in any of the three types of food expenditures will increase the householdsÕ group-based food diversity score by 1%. The finding that all three expenditure coefficients are equal implies that, despite different processed forms, similar levels of food subgroup diversity can be obtained from purchasing basic foods, convenience foods and FAFH.  Expenditures on all three types of food also have a significant effect on product code-based food diversity. However, results indicate that expenditures on basic foods,        convenience foods and FAFH have varying effects on product code-based food diversity. Increasing FAFH expenditures by $1 results in a 0.001 decrease in the product code-  based diversity index. This finding is the result of FAFH being comprised of only one  convenience food category under the product code classification scheme. Thus, despite  the fact that FAFH items are composed of a variety of food subgroups, increased expenditures on FAFH more highly concentrates food expenditures in a single food category, thus reducing diversity. Unlike FAFH, a $1 increase in expenditures on basic (convenience) foods results in a 0.002 (0.003) increase in the product code-based diversity index. Convenience food expenditures have a larger effect than basic food expenditures because the product code-based food diversity index is comprised of 91 convenience food categories versus 23 basic food categories; households must spend a !68 greater share of their food dollars on convenience versus basic foods to achieve the maximum food diversity score.  This rationale also explains why coefficients for basic and convenience food expenditures are larger than their group-based food diversity index counterparts. The product code-based food categories are specifically delineated into basic and convenience categories, making it highly responsive to the distribution of food expenditures across basic and convenience foods. In contrast, the group diversity index does not distinguish between basic and convenience foods, defining all food items in terms of their food subgroup composition. As with the descriptive statistics, these findings further suggest that the group-based index is capturing diversity in food subgroups, while the product code index reflects diversity in both processed form and food subgroups.   3.7.4 Demographics Results indicate that household income does not significantly affect either food diversity index. This finding mirrors that of Jekanowski and Binkley (2000) who explain that much of the effect of income on food diversity likely operates through food expenditures. Unlike earned income, receiving SNAP has a significant, inverse effect on group and product code-based food diversity, suggesting that SNAP householdsÕ food purchases are less diverse than those of non-SNAP households. The magnitude of the coefficient for SNAP in the group-based diversity index equation is double that of the corresponding product code-based coefficient. This difference implies that SNAP households have greater diversity across processed forms than across food subgroups. Because food subgroups are representative of nutritional adequacy, use of a product code-based food !69 diversity index in policy analysis may overstate the nutritional adequacy of food insecure householdsÕ food purchases. Supporting prior findings by Lee (1987) and Lee and Brown (1989), female-headed households are found to have greater group and product code-based food diversity scores than male-headed households. While both coefficients are positive, at 0.020 the product-code coefficient for female-headed households is larger than its group-based food diversity counterpart of 0.015. This larger coefficient for the product score suggests that female-headed householdsÕ food purchases are more evenly distributed across processed form than across food subgroups. Household size is found to have significant positive, but decreasing effect on both food diversity indices, suggesting that there are economies of scale in the diversity of food purchases. Specifically, the addition of a household member increases group (product code)-based food diversity at a decreasing rate until a household size of 3 (3.75) is reached, after which food diversity declines with household size. Also of note, the magnitudes of the household size and household size squared coefficients are double that of the product code-based coefficients. This suggests that larger households have lower food diversity across food subgroups than across processed forms. Thus, because food subgroups are representative of nutritional adequacy, use of a product code food diversity index in policy analysis may overstate the nutritional adequacy of larger householdsÕ food purchases. Results further indicate that households with a college-educated head are associated with a 0.013 and 0.008 increase in group and product code-based food diversity. The larger effect of a college education on the group-based index suggests that !70 college-educated households have greater diversity across food subgroups than across processed forms. Given the link between food subgroup variation and nutritional adequacy, this finding supports Moon et alÕs (2002) hypothesis that education increases the likelihood of consuming a varied diet due to having greater knowledge of its nutritional benefits. Thus, use of a product code-based food diversity index to inform nutrition education policy may understate the effect education has on householdsÕ nutritional adequacy.  While a household headÕs age does not affect group-based food diversity, a one-year increase in the age increases a householdÕs product code-based food diversity score by 0.001. Thus, relative to younger households, older households are expected to have greater diversity across processed form, but not across food subgroups. This result conflicts with findings by Lee (1987), Lee and Brown (1989) and Jekanowski and Binkley (2000), who find that food diversity decreases with age. However, these studies use datasets collected from 1977-1990. Thus, it is possible that the current generation of older households have greater variety across product code-based food categories.  While no significant effect is found for Hispanic or Asian households, we find that African-American householdsÕ group and product code-based food diversity indices are 0.017 and 0.034 lower than that of White households. The larger magnitude of the product code coefficient suggests African-American households have less diversity across processed form than across food subgroups. This corresponds with prior studiesÕ findings that African-Americans are less likely to consume convenience foods (Harris and Shiptsova 2007; Lee and Lin 2013; Capps et al 1985).   !71 3.8 Conclusions In past economic studies on food diversity demand, diversity is defined as a measure of food expenditures across categories designated based on product codes. Under a product code classification scheme, foods are classified based on their processed form. This approach conflicts with the DGA, which defines food diversity as variety across food subgroups (US 2015). Given the increasing prevalence of convenience foods in US household diets, this study develops a group-based food diversity index based on the US HFD, which represents diversity in household expenditures across food subgroups, irrespective of processed form. This group-based food diversity index is then applied to reexamine the relationship between food diversity, expenditures and demographics. Further, estimates obtained using the group-based index are compared to those obtained using a traditional product code-based index. The results from this paper confirm that the group and product code-based food diversity indices are capturing different forms of food diversity. Descriptive statistics indicate that the two indices are only moderately correlated and have varying means and skewness. Regression estimates show that food expenditures, SNAP, gender, age, household size, race and education significantly affect the diversity of household food purchases. However, the magnitude and direction of these effects vary for group and product code-based food diversity indices. In particular, receiving SNAP benefits has a larger inverse effect on group-based food diversity, suggesting that food insecure households have greater diversity across processed form than across subgroups. Further, education has a larger positive effect on group-based food diversity, indicating that educated households have greater diversity across subgroups than processed form. Also !72 of note, expenditures on FAFH positively affect group-based food diversity, but have an inverse effect on product code-based food diversity.  Given the differences between the group and product code-based food diversity indices, it is essential that studies select a food diversity index that directly corresponds to their research question. Results suggest that a group-based food diversity index is appropriate for studies seeking to inform food assistance and nutrition policy, while a product code-based index is ideal for guiding food industry managementÕs marketing strategies. Past economic studies overwhelmingly cite food diversityÕs link with nutritional adequacy as their motivation for analysis, explaining that their findings have important implications for policy involving food assistance and nutrition education programs (Thiele and Weiss 2003; Lee and Brown 1989; Shonkwiler et al 1987; Lee 1987). While each of these studies use a product code-based food diversity index, the nutritional motivation of their analysis suggests that use of a group-based food diversity index, which reflects variation across food subgroups and thus essential nutrients, would be more appropriate. The results from this article indicate that the use of product code-based food diversity indices in these past studies likely led to estimates which understate the food diversity issues faced by SNAP households, as well as the potential mediating effects of education. Differing from the majority of economic studies on food diversity, Jekanowski and Binkley (2000) analyze food diversity in order to guide food industry managers in their development of marketing strategies. In particular, understanding how market characteristics affect food diversity can help food industry manufacturers and retailers !73 determine their optimal product mix, in terms of the number, processed form and packaging of the products offered. In this context, use of a product code-based index is ideal in that it captures diversity across both processed form and food subgroups.                                 !74                        APPENDICES                       !75 APPENDIX A Supplementary Material For Chapter 1   !!!!             Table A.1 UK FSA's Nutrient Profiling Model A Points AllocationPointsEnergy (kJ)Sat Fat (g)Total Sugar (g)Sodium (mg)0! 335! 1 ! 4.5! 901> 335> 1> 4.5> 902> 670> 2> 9>1803>1005> 3> 13.5> 2704> 1340> 4> 18> 3605>1675> 5> 22.5> 4506> 2010> 6> 27> 5407> 2345> 7> 31> 6308> 2680> 8> 36> 7209> 3015> 9> 40> 81010> 3350> 10> 45> 900Table A.2 UK FSA's Nutrient Profiling Model C Points AllocationPointsFruit, Veg & Nuts (%)NSP Fiber (g)Or AOAC Fiber (g)Protein (g)0! 40! 0.7 ! 0.9! 1.61> 40> 0.7> 0.9> 1.62> 60> 1.4> 1.9> 3.23.> 2.1> 2.8> 4.84.> 2.8> 3.7> 6.45> 80> 3.5 > 4.7> 8.0!76                 VariableParticipationExpenditure                    Share   Low-Income/SNAP-0.23***-0.06***   Low-Income/Non-SNAP-0.03-0.01   Low-Access-0.03-0.03   FAH Healthy Share-0.050.10***   Share of FAFH Purchases at Supermarkets-0.34**0.14***   Share of FAFH Purchases at Convenience-0.58***-0.08**   Share of FAFH Purchases at FSR0.31***0.25***   Share of FAFH Purchases at Other Retailers-0.33***0.09***   FAFH Visits0.29***1.83E-03Household Characteristics    Age2.15E-032.36E-03***   College-0.080.02   African-American1.57E-030.01   Asian0.180.14***   Hispanic0.16*0.01   Rural -0.08-0.04***   Midwest-0.26***-0.05***   South-0.21**-0.03   West-0.16*-0.02Health Preferences   Nutrition Search0.12*0.04***   Dieting0.080.03***Time Cost   Healthy Time0.030.03***   Inverse Mills Ratio---0.42***Constant0.210.05Significant at the * 0.10 level, ** 0.05 level and *** 0.01 levelTable A.3 Double-Hurdle Model Estimates  (N=3,894)IncomeFood Acquisition Characteristics!77 APPENDIX B Supplementary Material For Chapter 2                         NutrientApple   JuiceCranberry Juice CocktailGrape Juice (Purple)Grapefruit Juice (White)Orange JuicePineapple JuicePrune JuiceEnergy, kcal6%7%8%5%5%7%9%Protein, g0%0%3%2%3%3%3%Total sugars, g76%94%119%49%63%109%109%Dietary fiber, g1%0%0%1%3%2%10%Total fat, g0%0%0%0%1%0%0%Vitamin A, RAE0%0%0%4%1%0%0%Vitamin E, mg0%3%0%1%2%0%2%Vitamin C, mg4%100%0%156%143%42%18%Calcium, mg2%1%0%2%2%3%3%Phosphorous, mg2%0%3%4%4%2%6%Magnesium, mg2%1%6%8%7%8%9%Iron, mg5%1%3%3%6%4%17%Sodium, mg0%0%0%0%0%0%0%Potassium, mg8%1%10%11%12%9%20%Sources: O'Neil & Nicklas (2008); FDA (2013b)Table B.1 % Daily Value of Nutrients in 8oz of Assorted Fruit Juices Based on a 2,000 Calorie Diet!78 APPENDIX C Supplementary Material For Chapter 3    Figure C.1 Correlation of Group and Product Code Food Diversity Indices                            !79                        REFERENCES                       !80 REFERENCES   Abate, G. and C. Peterson. 2005. Rapid Opportunity Assessment: Fruit Sector. The Strategic Marketing Institute. http://productcenter.msu.edu/strategic_support/ research.  Angulo, A.M. and J. M. Gil. 2006. Incorporating Nutrients into Meat Demand Analysis using Household Budget Data. Agricultural Economics 35(2):131Ð144.  Bauer, K.W., M.O. Hearst, A.A. Earnest, S.A. French, J.M. Oakes, and L.J. Harnack. 2012. Energy Content of US Fast-Food Restaurant Offerings: 14-Year Trends. American Journal of Preventive Medicine 43(5): 490-497.  Becker, G.S. 1965. A Theory of the Allocation of Time. The Economic Journal 75(299): 493-517.  Biancuzzo, R. M., A. Young, D. Bibuld, M. H. Cai, M. R. Winter, E. K. Klein, and A. Ameri. 2010. Fortification of Orange Juice with Vitamin D2 or Vitamin D3 is as Effective as an Oral Supplement in Maintaining Vitamin D Status in Adults. The American Journal of Clinical Nutrition 91(6): 1621Ð1626.  Binkley, J.K. 2006. The Effect of Demographic, Economic, and Nutrition Factors on the Frequency of Food Away from Home. Journal of Consumer Affairs 40(2): 372-391.  --- 2008. Calorie and Gram Differences between Meals at Fast Food and Table Service Restaurants. Applied Economic Perspectives and Policy 30(4): 750-763.  Bishai, D. and R. Nalubola. 2002. The History of Food Fortification in the United States: Its Relevance for Current Fortification Efforts in Developing Countries. Economic Development and Cultural Change 51(1): 37Ð53.  Bleich, S.N., J.A. Wolfson, and M.P. Jarlenski. 2015. Calorie Changes in Chain Restaurant Menu Items: Implications for Obesity and Evaluations of Menu Labeling. American Journal of Preventive Medicine 48(1): 70-75.  Box, G.E.P. and D.R. Cox. 1964. An Analysis of Transformations. Journal of the Royal Statistical Society Series B (Methodological): 211Ð252.  British Nutrition Foundation. 2007. Healthy Eating: a Whole Diet Approach. www. nutrition.org.uk (Accessed 3 May 2016).  Byrne, P.J., O. Capps, and A. Saha. 1996. Analysis of Food-Away-from-Home Expenditure Patterns for US Households 1982Ð89. American Journal of Agricultural Economics 78(3): 614-627. !81  Capps, O., J.R. Tedford and J. Havlicek. 1985. Household Demand for Convenience and Nonconvenience Foods. American Journal of Agricultural Economics 67(4): 862-869.  Chau, K. W. and T. L. Chin. 2003. A Critical Review of Literature on the Hedonic Price Model. International Journal for Housing Science and Its Applications 27(2): 145Ð165.  Chau, C.N., J.M. Zoellner, and J.L. Hill. 2013. Availability of Healthy Food: Does Block Group, Race and Income Matter. Journal of Hunger & Environmental Nutrition 8(1): 22-38.  Clemens, R., A. Drewnowski, M. G. Ferruzzi, C. D. Toner and D. Welland. 2015. Squeezing Fact from Fiction about 100% Fruit Juice. Advances in Nutrition: An International Review Journal  6(2): 236Ð243.  Cole, N., J. Jacobson, I. Nichols-Barrer and M. K. Fox. 2011. WIC Food Packages Policy Options Study. Mathematica Policy Research.   Costanigro, M., J. J. McCluskey, J. L. Lusk, J. Roosen and J. F. Shogren. 2011. Hedonic Price Analysis in Food Markets. The Oxford Handbook of the Economics of Food Consumption and Policy. Edited by Jayson L. Lusk, Jutta Roosen, and Jason F. Shogren: 153-161.   Creel, J.S., J.R. Sharkey, A. McIntosh, J. Anding, and J.C. Huber. 2008. Availability of Healthier Options in Traditional and Nontraditional Rural Fast-Food Outlets. BMC Public Health 8(1): 395-403.  Data Points: Supermarkets as a Prepared Food Destination. (2013). Retrieved February 25, 2015 from Supermarket News: http://supermarketnews.com/ datasheet/aug-19-2013-supermarkets-prepared-food-destination  De Lourdes Samaniego-Vaesken, M., E. Alonso-Aperte and G. Varela-Moreiras. 2012. Vitamin Food Fortification Today. Food and Nutrition Research 56 (5459) doi: 10.3402/fnr. v56i0 .5459.  Dennison, B. A., H. L. Rockwell and S. L. Baker. 1997. Excess Fruit Juice Consumption by Preschool-Aged Children is Associated with Short Stature and Obesity.  Pediatrics 99(1): 15-22.  Drescher, L.S., S. Thiele and G.B. Mensink. 2007. A New Index to Measure Healthy Food Diversity Better Reflects a Healthy Diet than Traditional Measures. The Journal of Nutrition 137(3): 647-651.  Drescher, L., S. Thiele and C. R. Weiss. 2008. The Taste for Variety: A Hedonic Analysis. Economics Letters 101(1):66Ð68. !82 Drewnowski, A., and V. Fulgoni. 2008. Nutrient Profiling of Foods: Creating a Nutrient-Rich Food Index. Nutrition Reviews 66(1): 23-39.  Eastwood, D. B., J. R. Brooker and D. E. Terry. 1986. Household Nutrient Demand: Use of Characteristics Theory and a Common Attribute Model. Southern Journal of Agricultural Economics 18(2):235Ð246.  Ervin, R.B. 2011. Healthy Eating IndexÐ2005 Total and Component Scores for Adults Aged 20 and Over: National Health and Nutrition Examination Survey, 2003Ð2004. National Health Stat Report 44: 1-9.  Eshak, E. S., H. Iso, T. Mizoue, M. Inoue, M. Noda and S.Tsugane. 2013. Soft Drink, 100% Fruit Juice, and Vegetable Juice Intakes and Risk of Diabetes Mellitus. Clinical Nutrition 32(2): 300Ð308. doi: 10.1016/j.clnu.2012.08.003.  Euromonitor International. 2015. Juice in the US. Euromonitor Passport Database. 31ÐMar 2015. http://www.euromonitor.com/usa. (Accessed March 31, 2016)  Euromonitor International. 2016. Fortified/Functional Packaged Food in the US. Euromonitor Passport Database. 30ÐMarch 2016. http://www.euromonitor .com/usa. (Accessed March 31, 2016)  Food Standards Agency. 2011. The Nutrient Profiling Model. https://www.gov.uk/ government/publications/the-nutrient-profiling-model. (Accessed August 2016).  Foote, J.A., S.P. Murphy, L.R. Wilkens, P. Basiotis and A. Carlson. 2004. Dietary Variety Increases the Probability of Nutrient Adequacy Among Adults, The Journal of Nutrition 134(7): 1779-1785.  Fox, K. J. and D. Melser. 2014. Non!Linear Pricing and Price Indexes: Evidence and Implications from Scanner Data. Review of Income and Wealth 60(2): 261Ð278.  Frank, L., J. Kerr, B. Saelens, J. Sallis, K. Glanz, and J. Chapman. 2009. Food Outlet Visits, Physical Activity and Body Weight: Variations by Gender and RaceÐEthnicity. British Journal of Sports Medicine 43(2): 124-131.  Gulseven, O. and M. Wohlgenant. 2014. Demand for Functional and Nutritional Enhancements in Specialty Milk Products. Appetite 81: 284-294.  Halvorsen, R. and H.O. Pollakowski. 1981. Choice of Functional Form for Hedonic Price Equations. Journal of Urban Economics 10(1): 37-49.  Harris, J.M. 1997. The Impact of Food Product Characteristics on Consumer Purchasing Behavior: The Case of Frankfurters. Journal of Food Distribution Research 28: 92Ð97.  Harris, J.L., M.B. Schwartz, and K.D. Brownell. 2010. Evaluating Fast Food Nutrition !83 and Marketing to Youth. New Haven, CT: Yale Rudd Center for Food Policy & Obesity.  Harris, J. L., M. B. Schwartz, K. D. Brownell, J. Javadizadeh and M. Weinberg. 2011. Evaluating Sugary Drink Nutrition and Marketing to Youth. New Haven, CT: Yale Rudd Center for Food Policy and Obesity.  Harris, J.M. and R. Shiptsova. 2007. Consumer Demand for Convenience Foods: Demographics and Expenditures. Journal of Food Distribution Research 38(3): 22-36.  Hiemstra, S.J., and W.G. Kim. 1995. Factors Affecting Expenditures for Food Away from Home in Commercial Establishments by Type of Eating Place and Meal Occasion. Journal of Hospitality & Tourism Research 19(3): 15-31.  Hiza, H.A.B, K.O. Casavale, P.M. Guenther, and C.A. Davis. 2013. Diet Quality of Americans Differs by Age, Sex, Race/Ethnicity, Income and Education Level. Journal of the Academy of Nutrition and Dietetics 113(2): 297-306.  Imamura, F., L. OÕConnor, Z. Ye, J. Mursu, Y. Hayashino, S. N. Bhupathiraju and N. G. Forouhi. 2015. Consumption of Sugar Sweetened Beverages, Artificially Sweetened Beverages, and Fruit Juice and Incidence of Type 2 Diabetes: Systematic Review, Meta-Analysis, and Estimation of Population Attributable Fraction: BMJ. 357.   Jackson, L.F. 1984. Hierarchic Demand and the Engel Curve for Variety. The Review of Economics and Statistics 66(1): 8-15.  Jeffery, R.W., J. Baxter, M. McGuire, and J. Linde. 2006. Are Fast Food Restaurants an Environmental Risk Factor for Obesity. International Journal of Behavioral Nutrition and Physical Activity 3(1): 1-6.  Jekanowski, M.D. and J.K. Binkley. 2000. Food Purchase Diversity across US Markets. Agribusiness 16(4): 417-433.  Jensen, H.H. and S.T. Yen. 1996. Food Expenditures Away from Home by Type of Meal. Canadian Journal of Agricultural Economics/Revue Canadienne dÕAgroeconomie 44(1): 67-80.  Kim, T.H., E.K. Lee, and E. Han. 2014. Food Away from Home and Body Mass Outcomes: Taking Heterogeneity into Account Enhances Quality of Results. Nutrition 30(9): 1015-1021.  Kinsey, J. 1983. Working Wives and the Marginal Propensity to Consume Food Away from Home. American Journal of Agricultural Economics 65(1): 10-19.   !84 Kirkpatrick, S.I., J. Reedy, L.L. Kahle, J.L. Harris, P. Ohri-Vachaspati, and S.M. Krebs-Smith. 2014. Fast-Food Menu Offerings Vary in Dietary Quality, but are Consistently Poor. Public Health Nutrition 17(4): 924-931.  Krebs-Smith, S.M. 2001. Choose Beverages and Foods to Moderate your Intake of Sugars: Measurement Requires Quantification. The Journal of Nutrition 131(2):527-535.  Lancaster, K.J. 1966. A New Approach to Consumer Theory. The Journal of Political Economy 132Ð157.  Larson, N.I., M.T. Story, and M.C. Nelson. 2009. Neighborhood Environments: Disparities in Access to Healthy Foods in the US. American Journal of Preventive Medicine 36(1): 74-81.  Lee, J.Y. 1987. The Demand for Varied Diet with Econometric Models for Count Data. American Journal of Agricultural Economics 69(3); 687-692.  Lee, J.Y. and M.G. Brown. 1989. Consumer Demand for Food Diversity. Journal of Agricultural and Applied Economics 21(2): 47-54.  Lee, J.Y. and B.H. Lin. 2013. A Study of the Demand for Convenience Food. Journal of Food Products Marketing 19(1): 1-14.  Lin, B.H., and J.F. Guthrie. 2012. Nutritional Quality of Food Prepared at Home and Away from Home, 1977-2008. United States Department of Agriculture, Economic Research Service: Economic Research Report No. 142361.  Liu, M., P. Kasteridis, and S.T. Yen. 2013a. Breakfast, Lunch, and Dinner Expenditures Away from Home in the United States. Food Policy 38: 156-164.  --- 2013b. Who are Consuming Food Away from Home and Where? Results from the Consumer Expenditure Surveys. European Review of Agricultural Economics 40(1): 191-213.  Martirosyan, D.M., and J. Singh. 2015. A New Definition of Functional Food by FFC: What Makes a New Definition Unique? Functional Foods in Health and Disease 5(6): 209Ð 223.  McCracken, V.A., and J.A. Brandt. 1987. Household Consumption of Food-Away-from-Home: Total Expenditure and by Type of Food Facility. American Journal of Agricultural Economics 69(2): 274-284.  Mintel Report. 2015. Juice, Juice Drinks and Smoothies Ð US Ð November 2015. Mintel Group-Consumer Market Research Reports. http://store.mintel.com/juice-juice-drinks-and-smoothies-us-november-2015. (Accessed March 25, 2016). !85  Mintel Report. 2014. Juice, Juice Drinks and Smoothies Ð US Ð November 2014. Mintel Group-Consumer Market Research Reports. http://store.mintel.com/juice-juice-drinks-and-smoothies-us-november-2014. (Accessed November 30, 2015).  Moon, W., W.J. Florkowski, L.R. Beuchat, A.V. Resurreccion, P. Paraskova, J. Jordanov, and M.S. Chinnan. 2002. Demand for Food Variety in an Emerging Market Economy. Applied Economics 34(5): 573-581.  Morgan, K.J., E.J. Metzen and S. R. Johnson. 1979. An Hedonic Index for Breakfast Cereals. Journal of Consumer Research 6(1): 67Ð75.  Morton, L.W and T.C. Blanchard. 2007. Starved for Access: Life in Rural AmericaÕs Food Deserts. Rural Realities 1(4): 1-10.  Murphy, S.P., J.A. Foote, L.R. Wilkens,  P.P. Basiotis, A. Carlson, K.L. White and K.M. Yonemori. 2006. Simple Measures of Dietary Variety are Associated with Improved Dietary Quality. Journal of the American Dietetic Association 106(3): 425-429.  Nguyen, B.T., K. Shuval, V.Y. Njike, and D.L. Katz. 2014. The Supplemental Nutrition Assistance Program and Dietary Quality among US Adults: findings from a Nationally Representative Survey. Mayo Clinic Proceedings 89(9): 1211-1219.   Okrent, A.M. and A. Kumcu. 2016. US Households' Demand for Convenience Foods. USDA-ERS Economic Research Report 211.  Okrent, A.M. and J.P. MacEwan. 2014. The Effects of Prices, Advertising, Expenditures, and Demographics on Demand for Nonalcoholic Beverages. Agricultural and Resource Economics Review 43(1): 31Ð52.  Oliveira, V., E. Racine, J. Olmsted, and L. M. Ghelfi. 2002. The WIC Program: Background, trends, and issues. U.S. Department of Agriculture, Economic Research Service. No. 33847.  OÕNeil, Carol E., and Theresa A. Nicklas. 2008. A Review of the Relationship between 100% Fruit Juice Consumption and Weight in Children and Adolescents. American Journal of Lifestyle Medicine 2(4): 315Ð354.  OÕNeil, C. E., T. A. Nicklas, M. Zanovec, R. E. Kleinman, and V. L. Fulgoni. 2012. Fruit Juice Consumption is Associated with Improved Nutrient Adequacy in Children and Adolescents: the National Health and Nutrition Examination Survey (NHANES) 2003Ð2006. Public Health Nutrition 15(10): 1871-1878.  Powell, L.M., S. Slater, D. Mirtcheva, Y. Bao, and F.J. Chaloupka. 2007. Food Store Availability and Neighborhood Characteristics in the United States. Preventive Medicine 44(3): 189Ð195. !86 Pudney, S. 1989. Modeling Individual Choice: The Econometrics of Corners, Kinks and Holes. Oxford: Basil Blackwell.  Redman, B.J. 1980. The Impact of Women's Time Allocation on Expenditure for Meals Away from Home and Prepared Foods. American Journal of Agricultural Economics 62(2): 234-237.  Reedy, J. and S.M. Krebs-Smith. 2010. Dietary Sources of Energy, Solid Fats, and Added Sugars Among Children and Adolescents in the United States. Journal of the American Dietetic Association 110(10):1477Ð1484.  Richards, T.J., and L. Mancino. 2013. Demand for Food-Away-from-Home: a Multiple-DiscreteÐContinuous Extreme Value Model. European Review of Agricultural Economics: jbt008.  Sharkey, J.R., C.M. Johnson, W.R. Dean, and S.A. Horel. 2011. Association Between Proximity to and Coverage of Traditional Fast-Food Restaurants and Non-Traditional Fast-Food Outlets and Fast-Food Consumption Among Rural Adults. International Journal of Health Geographics 10(1): 1-11.  Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. The Journal of Political Economy 82(1): 34Ð55.  Sethuraman, R. and C. Cole. 1999. Factors Influencing the Price Premiums that Consumers Pay for National Brands Over Store Brands. Journal of Product & brand Management 8(4): 340Ð351.  Shonkwiler, J. S., J. Lee and T.G. Taylor. 1987. An Empirical Model of the Demand for a Varied Diet. Applied Economics 19(10): 1403-1410.  Singh, G. M., R. Micha, S. Khatibzadeh, P.Shi, S. Lim, K. G. Andrews, R. E. Engell, M. Ezzati, D. Mozaffarian et al. 2015. Global, Regional, and National Consumption of Sugar-Sweetened Beverages, Fruit Juices, and Milk: A Systematic Assessment of Beverage Intake in 187 Countries. PloS ONE 10(8): e0124845.   Siro, I., E. Kapolna, B. Kapolna, and A. Lugasi. 2008. Functional Food, Product Development, Marketing and Consumer AcceptanceÑA Review. Appetite 51(3): 456Ð467.  Stanley, L.R., J.T. Tschirhart and J. Anderson. 1991. A Hedonic Price Analysis of Nutritionally Labeled Breakfast Cereals: Implications for Nutrient Labeling. Journal of Nutrition Education 23(5): 231Ð238.  Stewart, H., N. Blisard, D. Jolliffe, and S. Bhuyan. 2005. The Demand for Food Away from Home: Do Other Preferences Compete with our Desire to Eat Healthfully. Journal of Agricultural and Resource Economics 30(3): 520-536. !87 Szathvary, S. and S. Trestini. 2014. A Hedonic Analysis of Nutrition and Health Claims on Fruit Beverage Products. Journal of Agricultural Economics 65(2): 505Ð517.  Taylor, C. 2014. An Almost Ideal Demand System Analysis of Orange and Grapefruit Beverage Consumption in the United States. Thesis 1913. Clemson University.   Teuber, R. and R. Herrmann. 2012. Towards a Differentiated Modeling of Origin Effects in Hedonic Analysis: An Application to Auction Prices of Specialty Coffee. Food Policy 37(6): 732Ð740.  Theil, H. and R. Finke. 1983. The Consumer's Demand for Diversity. European Economic Review 23(3): 395-400.  Thiele, S. and C. Weiss. 2003. Consumer Demand for Food Diversity: Evidence for Germany. Food Policy 28(2): 99-115.  Todd, J.E., and B. Scharadin. 2016. Where Households Get Food in a Typical Week: Findings From USDA's FoodAPS. USDA-ERS: Economic Report 156.  Tuomisto, H. 2012. An Updated ConsumerÕs Guide to Evenness and Related Indices. Oikos 121(8): 1203-1218.  US Department of Agriculture. 2016a. Choose MyPlate. http://www.choosemyplate. gov/MyPlate. (Accessed June 2016).  US Department of Agriculture. 2016b. FoodAPS National Household Food Acquisition and Purchase Survey. Economic Research Service. http://www.ers.usda.gov/ dataproducts/foodaps-national-household-food-acquisition-and-purchasesurvey/ background.aspx. (accessed April 23, 2016).  US Department of Agriculture. 2016c. WIC Food Packages- Regulatory Requirements for WIC Eligible Foods. Food and Nutrition Service. http://www.fns.usda.gov/wic/wic-food-packages-regulatory-requirements-wic-eligible-foods. (accessed March 31, 2016). US Department of Agriculture. 2013a. Smart Snacks in School: USDAÕs All Foods Sold in Schools Standards. Food and Nutrition Service. http://www.fns.usda.gov/sites/ default/files/allfoods_flyer.pdf  (accessed March 31, 2016).  US Department of Agriculture, 2013b. Food Expenditures: Food Away From Home: Total Expenditures Table: http://www.ers.usda.gov/ data-products/food-expenditures.aspx#26636 (accessed February 24, 2015).  US Department of Health and Human Services. 2015. 2015-2020 Dietary Guidelines for Americans, 8th ed., Washington, DC: US Department of Health and Human Services.  !88  U.S. Department of Health and Human Services. 2010. Dietary guidelines for Americans. https://health.gov/dietaryguidelines/dga2010/dietaryguidelines2010.pdf.  U.S. Department of Health and Human Services. 2005. Dietary Guidelines for Americans. https://health.gov/dietaryguidelines/dga2005/document/.  US Food and Drug Administration (FDA). 2013b. Guidance for Industry: A Food Labeling Guide (14. Appendix F: Calculate the Percent Daily Value for Appropriate Nutrients). http://www.fda.gov/Food/GuidanceRegulation/Guidance DocumentsRegulatoryInformation/LabelingNutrition/ucm064928.htm.  Vadiveloo, M., L.B. Dixon, T. Mijanovich, B. Elbel and N. Parekh. 2014. Development and Evaluation of the US Healthy Food Diversity Index. British Journal of Nutrition 112(9): 1562-1574.  Ver Ploeg, M. et al. 2009. Access to Affordable and Nutritious Food: Measuring and Understanding Food Deserts and their Consequences: Report to Congress. U.S. Department of Agriculture. Economic Research Service. Administrative Publication No. (AP-036) http://www.ers.usda.gov/publications/ap-administrative-publication/ap-036.aspx.  Ver Ploeg, M. 2010. Access to Affordable, Nutritious Food is Limited in "Food Deserts". Amber Waves 8(1).   Wang, Y.C., S.N. Bleich and S.L. Gortmaker. 2008. Increasing Caloric Contribution from Sugar-Sweetened Beverages and 100% Fruit Juices among US Children and Adolescents, 1988Ð2004. Pediatrics 121(6): 1604-1614.  Weemaes, H. and P. Riethmuller. 2001. What Australian Consumers Like About Fruit Juice: Results from a Hedonic Analysis. Paper presented at annual conference of International Food and Agribusiness Management Association. Sydney, Australia. June 27Ð28.  Wojcicki, J.M. and M.B. Heyman. 2012. Reducing Childhood Obesity by Eliminating 100% Fruit Juice. American Journal of Public Health 102(9): 1630Ð1633.  Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.  Yen, S.T. 1993. Working Wives and Food Away from Home: the Box-Cox Double Hurdle Model. American Journal of Agricultural Economics 75(4): 884-895.  Yim, E.S., S. Lee, and W.G. Kim. 2014. Determinants of a Restaurant Average Meal Price: An Application of the Hedonic Pricing Model. International Journal of Hospitality Management 39: 11Ð20.